Method, apparatus, and system for providing automated solution for local branding and export commerce processes using ai-based country-specific influencer marketing

Abstract

One embodiment of the present invention relates to a method, apparatus, and system for providing a solution for local branding and export commerce process automation utilizing an AI model-based country-specific influencer marketing, which analyzes market data of a target country using an AI model and automates the entire process from establishing an optimized marketing strategy based thereon to creating an export-only sales page, recruiting influencers, and analyzing performance.

Description

Technology Field

[0001] The following embodiments relate to a method, apparatus, and system for providing a solution for local branding and export commerce process automation utilizing AI model-based country-specific influencer marketing, which analyzes market data of a target country using an AI model and automates the entire process from establishing an optimized marketing strategy based thereon to creating an export-only sales page, recruiting influencers, and analyzing performance. Background Technology

[0002] With the recent rapid growth of the cross-border e-commerce market, the demand for companies to expand overseas is increasing explosively. In particular, as the influence of social media expands, marketing utilizing local influencers has become an essential element.

[0003] However, existing export marketing methods had the following limitations. First, brand messages were frequently not effectively conveyed to local consumers due to a lack of understanding of local language, culture, and trends. Second, marketing activities (social media promotion) and sales activities (sales pages) were fragmented, resulting in a weak link to convert marketing traffic into actual purchases. Third, influencer selection relied solely on qualitative judgments based on follower counts, making it difficult to predict Brand Fit or ROI.

[0004] Furthermore, for overseas sales, it is necessary to build sales pages that consider not only language translation but also complex factors such as local currency, customs duties, shipping policies, and legal notices; however, there was a problem in that handling these individually required a massive amount of time and cost. In addition, the inability to clearly filter out fictitious traffic or abusive behavior using bots—a chronic issue in influencer marketing—resulting in inaccurate performance measurement and settlement also remains a challenge that needs to be addressed. Prior art literature

[0005] (Patent Document 0001) KR 10-2917865 B (Patent Document 0002) KR 10-2439666 B (Patent Document 0003) KR 10-2528407 B The problem to be solved

[0006] The problem that an embodiment of the present invention aims to solve is to provide a method, apparatus, and system for providing a solution for local branding and export commerce process automation using an AI model-based country-specific influencer marketing, which utilizes an AI model to analyze the latest trends of a target country and automatically establish a branding strategy, and provides a one-stop solution including the construction of localized sales pages and influencer matching based thereon, as well as performance measurement with guaranteed data integrity, in order to overcome the limitations of conventional export marketing and commerce operations as described above. means of solving the problem

[0007] According to one embodiment, a device includes a processor, memory, a communication module, and a non-transient storage medium, and a method for providing a solution for local branding and export commerce process automation utilizing an artificial intelligence model-based country-specific influencer marketing, which is performed by the processor by executing a program stored in the non-transient storage medium, comprises the steps of: collecting performance data for previously performed marketing activities to generate a monthly report and analyzing performance indicators of the monthly report; inputting the monthly report and local market data of a target country into an artificial intelligence model to generate localized branding strategy data including an optimal budget allocation range for future marketing, a visual concept guide for content based on the preference trends of the target country, and key emphasis keywords for each product; generating an export-only sales page based on the localized branding strategy data, the language of the target country, a local currency payment system, and customs and shipping policies; and generating an influencer campaign recruitment notice based on the localized branding strategy data and the export-only sales page, which includes a posting guide indicating the direction of the content. The present invention provides a method for providing a solution for local branding and export commerce process automation utilizing AI model-based country-specific influencer marketing, comprising the step of collecting result data of influencer activities performed according to the above posting guide and quantitatively calculating the marketing contribution of each influencer.

[0008] In addition, the step of generating the localization branding strategy data comprises: crawling top-ranking posts from social media platforms in the target country to collect a post dataset, and converting post images and text into latent vectors in a shared latent space using a multimodal embedding model; classifying the latent vectors into multiple trend clusters using a clustering algorithm and calculating the centroid vector of each trend cluster; inputting the detailed page text and product images of the product to be sold into the multimodal embedding model to generate product embedding vectors; calculating the cosine similarity between the product embedding vectors and the centroid vectors of each trend cluster, and selecting target clusters where the similarity is above a preset threshold; and extracting the average value of the image filter tone, the distribution probability of object placement, and the most frequently occurring adjective within the text from a set of reference posts belonging to the target clusters, and defining them as localization components. A step of fine-tuning an engagement rate prediction model using the aforementioned set of reference posts as training data, taking visual attributes including image filters, composition, and objects as input, and an engagement rate calculated based on the number of likes and comments on the posts as output; a step of inputting a plurality of virtual design combinations that can be generated based on localization components into the engagement rate prediction model to calculate a predicted engagement rate, and confirming the design combination with the maximum predicted engagement rate as a visual concept guide; a step of generating a keyword pool by extracting nouns and adjectives from the text of the aforementioned set of reference posts, and generating key emphasis keywords for each product by selecting long-tail keywords with low competition intensity relative to search volume among the keywords in the keyword pool.and may include the step of estimating the conversion rate (CVR) curve relative to the budget based on marketing performance data of past similar product families, deriving inflection points for achieving target sales, and determining the optimal budget allocation range.

[0009] And, the step of generating the export-only sales page comprises: converting image filter tone and color contrast ratio data included in the visual concept guide into hexadecimal color codes and Cascading Style Sheets (CSS) style property values, and dynamically injecting the CSS style property values ​​into style sheets defining the background, buttons, and highlighted text of the export-only sales page to render a user interface that aligns with the preference trends of the target country; performing local search engine optimization (SEO) by placing the key highlighted keywords for each product in the meta title and meta description within the head tag of the export-only sales page according to the priority of the key highlighted keywords for each product; translating the detailed description text of the product to be sold into the language of the target country, while restructuring the structure so that the keywords are placed at the beginning of the sentence while maintaining the meaning of the original text for sentences containing key highlighted keywords for each product, and generating localized content by detecting text areas within product images to be included in the export-only sales page, performing background restoration (in-painting), and then overlaying the translated text. A step of calculating an estimated customs duty by linking the HS Code (Harmonized System Code) of the product to be sold with the customs duty rate data of the target country, converting the product price into the local currency by calling a real-time exchange rate API, and calculating the final payment amount in the local currency by reflecting the estimated customs duty and the shipping fee policy based on the shipping zone by country; a step of determining whether the category of the product to be sold falls under a specific regulatory target recorded in the database, and if so, querying the local legal data of the target country to automatically insert mandatory legal notices and disclaimer clauses into the notice section at the bottom of the page;The method may include the steps of: configuring the payment module to bind the final payment amount based on the local currency to be displayed on the payment screen; generating a unique identification token to create a dedicated landing link for each influencer recruited through an influencer campaign recruitment announcement; and setting the unique identification token to be stored in a client storage area when a user accesses the export-only sales page; and inserting a tracking script into the export-only sales page that, upon the occurrence of a payment completion event, calls the unique identification token stored in the client storage area and transmits it to a device along with transaction data.

[0010] In addition, the step of generating the influencer campaign recruitment announcement comprises: loading the target country identifier, target audience, visual concept guide, key emphasis keywords for each product, and optimal budget allocation range from the localization branding strategy data; searching the influencer data pool using the target audience information as a query and calculating the audience fit, which is the ratio of the gender and age distribution of followers for each influencer that matches the target audience; extracting past post images of each influencer from the influencer data pool and inputting the past post images into a multimodal embedding model to convert them into post feature vectors; generating a reference image that visualizes the image filter tone, color contrast ratio, and object placement composition included in the visual concept guide, and inputting the reference image into a multimodal embedding model to generate a target concept vector; and calculating style similarity by calculating the cosine similarity between the target concept vector corresponding to the visual concept guide and the post feature vector. A step of extracting a group of influencer candidates in which the audience suitability and style similarity are above a predetermined standard and the ratio of bots or ghost accounts is below a preset threshold; a step of calculating, for each of the influencer candidates, a reach score based on the number of followers, an engagement score based on the average engagement rate, a similarity score between the category tag and the product category for sale, a performance score based on past sponsorship history, and a content suitability score; a step of calculating an influencer suitability index by weighted summing the reach score based on the number of followers, the engagement score based on the average engagement rate, the similarity score between the category tag and the product category for sale, the performance score based on past sponsorship history, the content suitability score, the audience suitability, and the style similarity.A step of classifying influencers into multiple grade ranges based on the influencer suitability index and automatically setting collaboration conditions for each grade, including payable sponsorship fees, product provision methods, performance-based incentives, and Key Performance Indicators (KPIs); a step of querying the posting registration specifications for each social media platform of the target country to generate posting template parameters including permitted phrase length, number of hashtags, possibility of link insertion, image / video upload specifications, and mandatory advertising disclosure items; a step of generating a list of required and prohibited hashtags combined based on the core emphasis keywords for each product, and generating a content creation guide by automatically inserting the standard image, required hashtags, and prohibited hashtag list into designated fields of the posting guide template; and a step of identifying paid advertising disclosure phrases and locations in the target country's language from the influencer advertising disclosure regulations of the target country stored in the database, and inserting them as mandatory provisions in the legal compliance section of the content creation guide. A step of setting an optimal upload window based on social media active traffic time zone data corresponding to the time zone of the target country and the category of the product to be sold, and specifying this as the recommended upload time that influencers must adhere to when uploading content in the content creation guidelines; a step of generating a basic announcement text including the category, price range, sales conditions, and shipping / return policy of the product to be sold based on the collaboration conditions and announcement template parameters, while restructuring the sentence structure so that key emphasis keywords for each product are placed at the beginning of the sentence and translating it into the target country language to generate a localized announcement text; and a step of generating a dedicated landing link for each influencer by combining a unique identification token with the landing URL included in the content creation guidelines.The method may include the steps of: creating a tracking link allocation slot within the database and mapping a dedicated landing link for each influencer to the tracking link allocation slot on a one-to-one basis to link with the data structure of the recruitment announcement; and creating a recruitment announcement packet containing the dedicated landing link for each influencer and content creation guidelines, and registering it according to the announcement registration API of the influencer recruitment platform.

[0011] In addition, the step of quantitatively calculating the marketing contribution for each influencer comprises: a step of collecting raw engagement data including likes, comments, shares, saves, and reach of content posted by influencers by periodically polling the open API of the social media platform of the target country; a step of generating commerce conversion data including the number of incoming clicks, number of items added to cart, number of purchase conversions, and total payment amount for each influencer by parsing the transaction log received from the tracking script and matching access logs containing unique identification tokens with purchase completion logs; and a data validation step of analyzing access IP addresses, user agents, and time elapsed between clicks and purchases for the raw engagement data and commerce conversion data to identify and remove repeated clicks from the same IP range, abnormal traffic suspected of being bots, and exit traffic with a dwell time below a preset valid threshold. Based on the engagement raw data and commerce conversion data for which validation has been completed, the method may include the step of calculating the return on advertising spend (ROAS), cost per click (CPC), and cost per purchase (CPA), which are sales revenue relative to total input costs; calculating the achievement rate by comparing performance with key performance indicators (KPIs); and determining the final settlement amount for each influencer by applying the incentive calculation logic specified in the collaboration conditions.

[0012] A device according to one embodiment may be combined with hardware and controlled by a computer program stored on a medium to execute the method of any one of the methods described above. Effects of the invention

[0013] According to one embodiment, by automating a series of processes that automatically derive budgets, visual concepts, and keyword strategies by analyzing existing marketing performance and market data of target countries using an artificial intelligence model, and then generate sales pages and influencer advertisements based on this and track performance, there is an effect of maximizing the operational efficiency of export companies and supporting sophisticated data-driven decision-making.

[0014] Furthermore, by utilizing a multimodal embedding model to analyze the visual and textual similarity between top-ranking posts and products for sale in target countries, and by deriving optimal design combinations and keywords through an engagement prediction model, it is possible to establish a highly localized branding strategy that aligns with the preference trends of local consumers.

[0015] Furthermore, by converting visual concept guides into CSS styles to render the UI, overlaying text within images with local languages ​​using in-painting technology, and automatically calculating the final price reflecting customs and shipping policies, even users without development knowledge can immediately build high-quality, export-only sales pages optimized for target countries.

[0016] In addition, by analyzing the vector similarity between an influencer's past posts and the target concept to precisely target influencers with a brand fit, and by automatically generating and distributing content guidelines that reflect advertising disclosure regulations and optimal upload times for each country, it is possible to reduce campaign operation resources while complying with marketing compliance.

[0017] In addition, by combining engagement data from social platforms with commerce conversion data through tracking scripts, and by filtering out bots or abnormal traffic through IP address and user agent analysis, it is possible to transparently calculate the actual marketing contribution of influencers and establish a reasonable settlement system. Brief explanation of the drawing

[0018] FIG. 1 is a schematic diagram showing a system for providing a local branding and export commerce process automation solution utilizing artificial intelligence model-based country-specific influencer marketing according to an embodiment of the present invention. FIG. 2 is a flowchart illustrating a method for providing a solution for local branding and export commerce process automation using artificial intelligence model-based country-specific influencer marketing according to an embodiment of the present invention. FIG. 3 is a flowchart illustrating the step of generating localization branding strategy data for a method of providing a solution for local branding and export commerce process automation using an artificial intelligence model-based country-specific influencer marketing according to an embodiment of the present invention. FIG. 4 is a flowchart illustrating the step of creating an export-only sales page in a method for providing a local branding and export commerce process automation solution utilizing an artificial intelligence model-based country-specific influencer marketing according to an embodiment of the present invention. Specific details for implementing the invention

[0019] Hereinafter, embodiments are described in detail with reference to the attached drawings. However, various modifications may be made to the embodiments, and thus the scope of the patent application is not limited or restricted by these embodiments. It should be understood that all modifications, equivalents, and substitutions to the embodiments are included within the scope of the rights.

[0020] Specific structural or functional descriptions of the embodiments are disclosed for illustrative purposes only and may be modified and implemented in various forms. Accordingly, the embodiments are not limited to the specific disclosed forms, and the scope of this specification includes modifications, equivalents, or substitutions that fall within the technical concept.

[0021] Terms such as "first" or "second" may be used to describe various components, but these terms should be interpreted solely for the purpose of distinguishing one component from another. For example, the first component may be named the second component, and similarly, the second component may be named the first component.

[0022] When it is stated that a component is "connected" to another component, it should be understood that it may be directly connected to or joined to that other component, or that there may be other components in between.

[0023] The terms used in the embodiments are for illustrative purposes only and should not be interpreted as intended to be limiting. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this specification, terms such as "comprising" or "having" are intended to indicate the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0024] In particular, where a 'step' in this specification is described as 'comprising' one or more detailed steps or sub-steps, said 'step' may be interpreted as including its own basic processing step while simultaneously performing the described detailed steps as well.

[0025] For example, if it is stated that 'a step of doing B to A' includes 'a step of doing D to C; a step of doing F to E; and a step of doing H to G,' the 'step of doing B to A' may be interpreted not merely as the basic operation of doing B to A, but as a configuration that performs detailed procedures together, such as a step of doing D to C, a step of doing F to E, and a step of doing H to G.

[0026] Accordingly, the above configuration does not exclude various sub-procedures included within the scope of execution of the corresponding step, and may be included within the scope of the present invention even if other procedures or means performing substantially the same or equivalent functions are substituted.

[0027] Expressions such as 'end part', 'both ends', 'one end', 'other end', and 'side end' of a component can be interpreted as referring to at least / any one of the end parts of that component.

[0028] In the description of the present invention, 'a method in which a device comprises a processor, a memory, a communication module, and a non-transient storage medium, and a program stored in the non-transient storage medium is executed by the processor,' the term 'method' may be interpreted as referring to the program stored in the non-transient storage medium itself or a part of the program.

[0029] The term 'Return' as used in the description of the present invention may refer to a result value being output, returned, or returned from a method, procedure, function, etc. used in a given program language / structure.

[0030] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which the embodiments pertain. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in this application. For example, the term 'artificial intelligence model' may be selected from one or more of known general artificial intelligence models.

[0031] In addition, when describing with reference to the attached drawings, identical components are assigned the same reference numeral regardless of drawing symbols, and redundant descriptions thereof are omitted. In describing the embodiments, if it is determined that a detailed description of related prior art could unnecessarily obscure the essence of the embodiments, such detailed description is omitted.

[0032] According to one embodiment, a device includes a processor, memory, a communication module, and a non-transient storage medium, and a method for providing a solution for local branding and export commerce process automation utilizing an artificial intelligence model-based country-specific influencer marketing, which is performed by the processor by executing a program stored in the non-transient storage medium, comprises the steps of: collecting performance data for previously performed marketing activities to generate a monthly report and analyzing performance indicators of the monthly report; inputting the monthly report and local market data of a target country into an artificial intelligence model to generate localized branding strategy data including an optimal budget allocation range for future marketing, a visual concept guide for content based on the preference trends of the target country, and key emphasis keywords for each product; generating an export-only sales page based on the localized branding strategy data, the language of the target country, a local currency payment system, and customs and shipping policies; and generating an influencer campaign recruitment notice based on the localized branding strategy data and the export-only sales page, which includes a posting guide indicating the direction of the content. The present invention provides a method for providing a solution for local branding and export commerce process automation utilizing AI model-based country-specific influencer marketing, comprising the step of collecting result data of influencer activities performed according to the above posting guide and quantitatively calculating the marketing contribution of each influencer.

[0033] In the step of collecting performance data on previously performed marketing activities to generate a monthly report and analyzing the performance indicators of the monthly report, result data of past country- or channel-specific marketing activities is collected to generate a report organized on a monthly basis, and the performance indicators included in the report are analyzed and structured into a form usable for subsequent decision-making.

[0034] Here, 'existing marketing activities already performed' refers to a series of execution units for marketing purposes, such as uploading influencer posts, running advertisements, operating landing pages, providing coupons / promotions, and performing retargeting campaigns, and 'performance data' refers to a data set including logs related to impressions, clicks, conversions, sales, and costs generated from the above execution units.

[0035] In the above steps, the device may collect performance data from multiple data sources. For example, the device may collect impressions, reach, likes, comments, shares, saves, profile visits, etc. per post from open APIs or creator / business insight APIs provided by social media platforms, and may collect ad costs, clicks, impressions, click-through rates, conversions, etc. from report APIs of advertising platforms (search ads, social ads, etc.). Additionally, it may collect purchase conversions, total payment amounts, payment amounts per order, and refund counts from logs of sales pages or payment systems. In this case, since different data sources may differ in timestamp standards, aggregation units, and metric definitions, it is desirable for the device to perform normalization and mapping after data collection to integrate them into the same report schema.

[0036] A monthly report refers to document data or structured data suitable for input into subsequent models, which visualizes or summarizes time-series trends, month-over-month growth rates, and efficiency comparisons by channel by aggregating collected performance data on a monthly basis.

[0037] For example, the device may generate a report header containing a target country identifier, campaign identifier, channel identifier, product identifier, and aggregation period (start and end dates of the month) as report metadata, and generate a report table containing aggregated values ​​by metric in the report body.

[0038] The stage of analyzing performance indicators in the monthly report includes a process of processing or deriving indicators beyond simple aggregation so that they can be used as input values ​​for artificial intelligence models in subsequent stages.

[0039] As a specific example of performance indicator analysis, the device can calculate (i) engagement rate (value obtained by dividing the sum of likes, comments, shares, and saves by the reach or impressions), (ii) conversion rate (value obtained by dividing the number of purchase conversions by the number of incoming clicks), (iii) return on advertising (value obtained by dividing the total payment amount by the total advertising cost), and (iv) cost per click (value obtained by dividing the total advertising cost by the number of incoming clicks).

[0040] For example, if the target country is Japan in a specific month, the total payment amount of the campaign is 25,000,000 won, and the total advertising cost is 5,000,000 won, the device can quantify the efficiency of the campaign by calculating the Return on Advertising Spend (ROAS) as 5 (500%). These derived metrics are used as input variables in the next step to determine the optimal budget allocation range or to compare performance by product and channel.

[0041] In the stage of analyzing performance indicators, preprocessing may be performed to ensure the reliability of the indicators. The device ensures data integrity by excluding specific segments or lowering their weights when abnormal surge patterns (such as a sudden surge in reach during a specific time period or repeated clicks based on the same IP range) are detected. Furthermore, to facilitate comparison between different channels, the process may include standardizing or normalizing the indicators to scale the input for subsequent models.

[0042] In the step of inputting the above monthly report and local market data of the target country into an artificial intelligence model to generate localization branding strategy data including the optimal budget allocation range for next marketing, visual concept guides for content based on the preference trends of the target country, and key emphasis keywords for each product, a future marketing strategy is established by combining analyzed past performance data with external data of the currently targeted country.

[0043] Local market data for a target country refers to big data that includes real-time popular search terms on major social media within that country, pricing and promotional information for competitor products, preferred colors or design patterns of local consumers, and seasonal factors (e.g., Black Friday in the U.S., Singles' Day in China, etc.).

[0044] The above artificial intelligence model may be a regression model or a deep learning model trained based on these input values, and calculates the optimal budget allocation range, which is the budget range predicted to be most efficient when deployed for marketing next month.

[0045] The AI ​​model analyzes local trends to derive a 'visual concept guide for content' and 'key emphasis keywords for each product.' The visual concept guide refers to a design guideline that defines image filters (e.g., warm tones, cool tones), composition, font styles, and the placement of key objects preferred by consumers in the target country. Key emphasis keywords for each product are extracted words that local consumers consider important when searching or making purchasing decisions (e.g., 'value for money,' 'eco-friendly,' 'fast delivery,' 'K-beauty,' etc.).

[0046] Localization branding strategy data refers to a comprehensive marketing strategy packet that includes the aforementioned budget, concept guide, keywords, etc.

[0047] In the step of creating an export-only sales page based on the target country's language, local currency payment system, customs, and shipping policies, based on the localization branding strategy data mentioned above, the previously derived strategy data is applied to the webpage where actual sales take place (landing page or shopping mall detail page) to build it.

[0048] An export-only sales page refers to a page that, unlike a domestic sales page, is displayed to overseas visitors and is optimized for the UX / UI environment of local consumers.

[0049] For example, if the target country is 'Vietnam' and 'bright and highly saturated images' and 'emphasis on whitening effects' are derived from the localization branding strategy data, the device automatically adjusts the main image of the sales page to that tone, translates the detailed description text into Vietnamese, and reconstructs the structure to place keywords related to 'whitening' at the top of the page.

[0050] The local currency payment system refers to converting and displaying product prices in Vietnamese Dong (VND) and integrating payment modules commonly used locally, such as e-wallets or credit cards; the customs and shipping policy includes automatically applying logic to either add estimated customs duties and international shipping costs from Korea to Vietnam to the product price or display them separately.

[0051] In the step of creating an influencer campaign recruitment notice including a posting guide that directs the direction of the content based on the above localization branding strategy data and export-only sales page, a job posting and content creation guidelines are created to recruit influencers to perform marketing.

[0052] An influencer campaign recruitment notice is an announcement posted on an influencer matching platform or social media, which specifies the target of recruitment (number of followers, category), benefits provided (product sponsorship, writing fee), activity period, etc.

[0053] The posting guide is a set of guidelines that influencers must strictly adhere to when creating content, and it reflects the previously generated 'visual concept guide' and 'key emphasis keywords for each product.'

[0054] For example, the posting guide includes specific instructions such as "Photograph holding the product under natural light (visual concept)," "Include hashtags '#Koreanskin' and '#Glow' in the post body (key emphasis keywords)," and "Place a link to the sales page at the top of the profile," encouraging influencers to create content that aligns with the brand's strategic intentions.

[0055] In the step of collecting result data of influencer activities performed according to the above posting guide and quantitatively calculating the marketing contribution of each influencer, performance is evaluated by tracking the reactions that occurred after the recruited influencers actually uploaded posts.

[0056] Result data of influencer activities includes engagement metrics such as the number of likes, comments, shares, and saves of posts, as well as the number of clicks that led to sales pages through the posts and the number of conversions that resulted in actual purchases.

[0057] Quantitatively calculating the marketing contribution of each influencer means going beyond a qualitative evaluation such as simply having "many likes" to calculating a numerical score, such as, "Influencer A generated $500 in sales with a cost of $100, so the contribution is 5.0 (ROAS 500%)."

[0058] In this case, weights reflecting the influencer's rank, the quality of the posts (compliance with posting guidelines), and the presence of abuse (fake clicks) identified during the aforementioned data preprocessing are applied to the contribution calculation, thereby deriving a substantial and reliable contribution.

[0059] In addition, the step of generating the localization branding strategy data comprises: crawling top-ranking posts from social media platforms in the target country to collect a post dataset, and converting post images and text into latent vectors in a shared latent space using a multimodal embedding model; classifying the latent vectors into multiple trend clusters using a clustering algorithm and calculating the centroid vector of each trend cluster; inputting the detailed page text and product images of the product to be sold into the multimodal embedding model to generate product embedding vectors; calculating the cosine similarity between the product embedding vectors and the centroid vectors of each trend cluster, and selecting target clusters where the similarity is above a preset threshold; and extracting the average value of the image filter tone, the distribution probability of object placement, and the most frequently occurring adjective within the text from a set of reference posts belonging to the target clusters, and defining them as localization components. A step of fine-tuning an engagement rate prediction model using the aforementioned set of reference posts as training data, taking visual attributes including image filters, composition, and objects as input, and an engagement rate calculated based on the number of likes and comments on the posts as output; a step of inputting a plurality of virtual design combinations that can be generated based on localization components into the engagement rate prediction model to calculate a predicted engagement rate, and confirming the design combination with the maximum predicted engagement rate as a visual concept guide; a step of generating a keyword pool by extracting nouns and adjectives from the text of the aforementioned set of reference posts, and generating key emphasis keywords for each product by selecting long-tail keywords with low competition intensity relative to search volume among the keywords in the keyword pool.and may include the step of estimating the conversion rate (CVR) curve relative to the budget based on marketing performance data of past similar product families, deriving inflection points for achieving target sales, and determining the optimal budget allocation range.

[0060] In the step of crawling top-ranking posts on social media platforms of the aforementioned target country to collect a post dataset and converting post images and text into latent vectors in a shared latent space using a multi-modal embedding model, content displayed in the 'Explore tab' or 'Popular Posts' section of the most popular social media (Instagram, TikTok, etc.) in the target country (e.g., the United States, Japan, Vietnam, etc.) is collected.

[0061] Here, top-ranking posts (popular posts; recommended posts) refer to posts recommended by an algorithm due to a surge in views, likes, and shares over a specific period (e.g., the last week), and these serve as the best sample reflecting current trends in the country.

[0062] The device constructs a post dataset by collecting posts from areas where top exposure occurs, such as platform-specific explore tabs, hashtag trends, popular feeds by category, shopping tags, and recommendation feeds for target countries. Top exposure may refer to exposure rankings, recommendation rankings, sorting based on views / reactions, or metrics indicating a surge in a window over a certain period of time provided by the platform, and the device stores top exposure determination rules in a database according to platform-specific policies and filters the collection targets based on those rules.

[0063] Multimodal embedding models (e.g., CLIP, ALIGN, etc.) refer to artificial intelligence models that map text and images to a single common mathematical space (common latent space) rather than to different spaces. Through this process, 'posts as images' and 'hashtags / body text as text' are converted into vector values ​​of the same dimension (latent vectors), making it possible to calculate the semantic similarity between the image and the text.

[0064] Specifically, the device can generate an image latent vector by performing preprocessing (resizing, color normalization, watermark / frame removal, etc.) on a post image and inputting it into an image encoder, and generate a text latent vector by performing language detection, tokenization, stop word removal, emoji / special character normalization, etc. on the post text and inputting it into a text encoder. Additionally, the device can generate a post-unit integrated latent vector by simply concatenating or weighted summing the image latent vector and the text latent vector.

[0065] In the step of classifying the above potential vectors into multiple trend clusters using a clustering algorithm and calculating the center vector of each trend cluster, tens of thousands of collected post vectors are grouped together based on their similar characteristics.

[0066] Clustering algorithms such as K-Means and DBSCAN can be used to classify randomly collected data into meaningful groups (trend clusters), such as 'minimalist sensibility', 'glamorous party atmosphere', and 'eco-friendly / nature mood'.

[0067] A centroid is a vector representing the average position of all vectors included within a cluster, signifying the mathematical definition of a 'typical style' that represents the corresponding trend group. By calculating representative post lists, average reaction indicators, and representative hashtags for each trend cluster, the device can enhance the interpretability of the trends indicated by the centroid.

[0068] In the step of generating product embedding vectors by inputting the detail page text and product images of the product to be sold into the multimodal embedding model, the information of the product to be exported is vectorized using the same model previously used for trend analysis.

[0069] For example, if the product to be sold is 'organic sunscreen', the product's package image and the description (ingredients, efficacy, etc.) written on the detail page are input into a multimodal embedding model to generate a 'product embedding vector', which is a coordinate value indicating where the product is located in the common latent space.

[0070] Specifically, the device can input product text and product images into a text encoder and an image encoder, respectively, and generate a product embedding vector by combining the calculated vectors. If there are multiple product images, the device can construct product image embeddings by calculating embeddings for each image and then averaging or weighting the results.

[0071] In the step of calculating the cosine similarity between the product embedding vector and the center vector of each trend cluster, and selecting target clusters where the similarity is greater than or equal to a preset threshold, it is determined which style among the currently trending trends our product fits best.

[0072] Cosine similarity is a measure that uses the angle between two vectors to represent the degree of similarity as a value between -1 and 1. The device compares the product vector with the center vector of each trend cluster and selects the cluster with the highest similarity or one that is above a preset threshold (e.g., 0.7) as the 'target cluster'. This is not simply to follow trends, but to find 'locally trending styles that do not compromise the identity of our product'.

[0073] If there are multiple candidates, the device can determine the final target cluster by selecting the top N based on similarity scores or by additionally reflecting the cluster's average engagement rate, average purchase conversion rate, brand safety score, etc.

[0074] In the step of defining localization components by extracting the average value of image filter tone, the distribution probability of object placement, and the most frequently occurring adjective in the text from a set of reference posts belonging to the above target cluster, specific success factors are extracted by performing an in-depth analysis of the posts (reference posts) within the selected target cluster.

[0075] The average value of image filter tones refers to the histogram average of color tone (warmth / coolness), saturation, brightness, contrast, etc., applied to the posts. The distribution probability of object placement refers to data analyzed in the form of a heatmap using an object recognition model (e.g., YOLO) to determine where products, people (models), background props, etc., are primarily placed within an image (center, bottom-left, etc.). The most frequently occurring adjectives in text refer to sentiment adjectives (e.g., 'cozy', 'luxury', 'kawaii') that appear most often in the body of a post or in comments, extracted using natural language processing (NLP). The device groups these together and defines them as 'localization components'.

[0076] In the step of fine-tuning an engagement rate prediction model using the above-mentioned set of reference posts as training data, taking visual attributes including image filters, composition, and objects as input, and an engagement rate calculated based on the number of likes and comments on the posts as output, a dedicated AI model capable of predicting "what style of image is responsive?" is trained.

[0077] By additionally training (fine-tuning) an existing vision model by mapping the image data of the target cluster (input) with the actual performance of those images (engagement rate based on likes + comments; output), the accuracy of performance prediction within the corresponding trend is maximized.

[0078] Visual attributes may include numeric parameters extracted from the above localization components (color temperature, contrast, saturation, filter intensity, etc.), compositional features (position of product objects, margin ratio, text overlay area ratio, etc.), and object features (size ratio of product objects, inclusion of people, background type, etc.).

[0079] In the step of calculating the predicted participation rate by inputting multiple virtual design combinations that can be generated based on localization components into the above participation rate prediction model, and confirming the design combination with the maximum predicted participation rate as a visual concept guide, simulations are performed by combining the extracted components in various ways.

[0080] For example, thousands of virtual design combinations, such as [Filter A + Composition B + Text C] and [Filter A + Composition C + Text B], are generated and input into a previously trained engagement prediction model. The best combination predicted by the model to record the highest engagement rate is finalized as the 'Visual Concept Guide'.

[0081] The visual concept guide can be saved in a reusable form for subsequent creation stages, such as color codes, filter intensity ranges, shooting / editing rules, object placement rules, and text length rules.

[0082] In the step of generating a keyword pool by extracting nouns and adjectives from the text of the above set of reference posts, and generating core emphasis keywords for each product by selecting long-tail keywords with low competitive intensity relative to search volume from among the keywords in the keyword pool, the language to which local consumers actually respond is converted into marketing keywords.

[0083] A keyword pool can refer to a set of candidate nouns and adjectives extracted from reference post text, hashtags, top phrases in comments, and the like. A long-tail keyword refers to a keyword that reflects specific needs—such as "long-lasting matte lipstick"—which has lower search volume but high purchase conversion rates and low competition intensity, rather than a common noun (head keyword) like "lipstick" that has high search volume but fierce competition.

[0084] The device checks the search volume and advertising competition of each keyword through search engine APIs, etc., and selects the most efficient long-tail keywords as 'core emphasis keywords for each product.' For example, this can be derived by prioritizing keywords with a high search volume / competition intensity ratio, or by filtering out keywords with competition intensity below a threshold and then selecting the top-ranking keywords by search volume.

[0085] In the step of estimating the conversion rate (CVR) curve relative to the budget based on past marketing performance data of similar product lines, deriving inflection points for achieving target sales, and determining the optimal budget allocation range, an efficient execution plan for the marketing budget is established.

[0086] The device analyzes campaign data from past similar products (same category, similar price range) stored in the database to estimate the CVR curve, which represents the change in conversion rate (Y-axis) according to the amount of marketing expenditure (X-axis). Generally, increasing marketing costs leads to increased sales, but the 'law of diminishing returns' applies, where cost-effectiveness decreases beyond a certain level. The device identifies the point where this efficiency begins to drop sharply (inflection point) to determine the 'optimal budget allocation range' that prevents excessive spending and enables the achievement of the target revenue (ROAS).

[0087] For example, if the conversion rate increases from 2 percent to 3.5 percent when the monthly budget is increased from 1 million won to 2 million won, but only increases from 3.5 percent to 3.7 percent when the budget is increased from 2 million won to 3 million won, the device may determine the area around 2 million won as an inflection point and suggest a range such as 1.8 million won to 2.4 million won as the optimal budget allocation range.

[0088] As a result of performing the above steps, the device can generate and store localization branding strategy data including a set of visual concept guide parameters defined by target country, a set of key emphasis keywords by product, target trend cluster identifiers and reference post set metadata, an optimal budget allocation range, and indicators serving as the basis for its calculation; this data is directly input into subsequent steps, such as the creation of export-only sales pages and influencer campaign recruitment announcements, to enable automated localization execution.

[0089] And, the step of generating the export-only sales page comprises: converting image filter tone and color contrast ratio data included in the visual concept guide into hexadecimal color codes and Cascading Style Sheets (CSS) style property values, and dynamically injecting the CSS style property values ​​into style sheets defining the background, buttons, and highlighted text of the export-only sales page to render a user interface that aligns with the preference trends of the target country; performing local search engine optimization (SEO) by placing the key highlighted keywords for each product in the meta title and meta description within the head tag of the export-only sales page according to the priority of the key highlighted keywords for each product; translating the detailed description text of the product to be sold into the language of the target country, while restructuring the structure so that the keywords are placed at the beginning of the sentence while maintaining the meaning of the original text for sentences containing key highlighted keywords for each product, and generating localized content by detecting text areas within product images to be included in the export-only sales page, performing background restoration (in-painting), and then overlaying the translated text. A step of calculating an estimated customs duty by linking the HS Code (Harmonized System Code) of the product to be sold with the customs duty rate data of the target country, converting the product price into the local currency by calling a real-time exchange rate API, and calculating the final payment amount in the local currency by reflecting the estimated customs duty and the shipping fee policy based on the shipping zone by country; a step of determining whether the category of the product to be sold falls under a specific regulatory target recorded in the database, and if so, querying the local legal data of the target country to automatically insert mandatory legal notices and disclaimer clauses into the notice section at the bottom of the page;The method may include the steps of: configuring the payment module to bind the final payment amount based on the local currency to be displayed on the payment screen; generating a unique identification token to create a dedicated landing link for each influencer recruited through an influencer campaign recruitment announcement; and setting the unique identification token to be stored in a client storage area when a user accesses the export-only sales page; and inserting a tracking script into the export-only sales page that, upon the occurrence of a payment completion event, calls the unique identification token stored in the client storage area and transmits it to a device along with transaction data.

[0090] The step of creating the above-mentioned export-only sales page is a process of applying the localization branding strategy data derived in the previous step to the webpage where actual sales take place, thereby automatically building an e-commerce environment optimized for users in the target country and creating a structure capable of influencer-based performance tracking.

[0091] In the step of converting image filter tone and color contrast ratio data included in the above visual concept guide into hexadecimal color codes (Hex Color Code) and CSS (Cascading Style Sheets) style attribute values, and dynamically injecting the CSS style attribute values ​​into style sheets defining backgrounds, buttons, and emphasis text for export-only sales pages to render a user interface that matches the preference trends of the target country, the abstract design strategy derived by AI is converted into actual web page code.

[0092] For example, if 'pastel pink tones' and 'low contrast' are identified as preferred trends for a target country (Japan), the device maps these to hexadecimal color codes (e.g., #FFD1DC) that web browsers can recognize, as well as CSS property values ​​such as opacity and contrast ratio. Subsequently, these values ​​are dynamically injected into CSS variables or style sheets on the sales page to automatically change the colors of background colors, purchase buttons (CTA Buttons), highlighted text, and promotion badges, thereby immediately rendering a user interface (UI) familiar to local consumers without the need for separate publishing work.

[0093] In the step of performing local search engine optimization (SEO) by placing the core emphasis keywords for each product in the meta title and meta description within the head tag of the export-only sales page according to the priority of the core emphasis keywords for each product, technical measures are taken to ensure that the product is displayed at a high level on search engines (Google, Yahoo Japan, Baidu, etc.) in the target country.

[0094] The device selects the top keywords with the highest search volume and conversion rate among the 'product-specific core emphasis keywords' and, within the head tag of the HTML document, ' <title>Place it at the leading position of '(meta title), and the secondary keyword is '< / title> <meta name="description"> Inserts as a natural language sentence within the (meta description) attribute.

[0095] For example, when targeting the US market, if 'Vegan' is the core keyword, instead of simply making the page title "Soothing Cream," structure it as "Vegan Soothing Cream - Eco-friendly & Cruelty-free" to induce search bots to recognize the page as important vegan-related content.

[0096] In the step of generating localized content by translating the detailed description text of the above-mentioned product for sale into the language of the target country, while restructuring sentences containing key emphasis keywords for each product so that the keywords are placed at the beginning of the sentence while maintaining the meaning of the original text, and detecting text areas within product images to be included in the export-only sales page, performing background restoration (in-painting), and then overlaying the translated text, linguistic and visual localization are performed simultaneously.

[0097] Going beyond simple translation, sentence structures are rearranged to consider local consumers' readability and marketing effectiveness. For example, since English-speaking countries prefer the core benefit to appear at the beginning of the sentence, "This product moisturizes the skin" is transformed into the form "Get Hydrated Skin with..." to emphasize the key keyword.

[0098] In addition, it automatically generates content that is perfectly localized, including the text within the image, by detecting text areas within product detail page images using OCR, erasing the corresponding characters and restoring the background using AI in-painting technology, and then overlaying the translated text with a font style similar to the original.

[0099] In the step of calculating an estimated customs duty by linking the HS code (Harmonized System Code) of the product to be sold with the customs duty rate data of the target country, converting the product price into the local currency by calling a real-time exchange rate API, and then calculating the final payment amount in the local currency by reflecting the estimated customs duty and the shipping fee policy based on the shipping zone by country, the price uncertainty experienced by consumers when making direct overseas purchases is eliminated.

[0100] The device identifies the HS code matching the category of the product for sale and queries the estimated customs rate by linking with the customs database of the target country. It then converts the base product price by reflecting real-time exchange rate information and calculates the 'Landed Cost' by adding customs duties and international shipping fees to this. Through this process, consumers can verify and pay a clear price in local currency that includes both taxes and shipping fees.

[0101] In the step of determining whether the category of the product to be sold falls under specific regulations recorded in the database, and if so, querying local legal data of the target country to automatically insert mandatory legal notices and disclaimer clauses into the notice section at the bottom of the page, sales risk is minimized by complying with legal regulations that vary by country.

[0102] By querying a regulatory information database, it automatically inserts, for example, mandatory wording required by U.S. FDA regulations for 'health functional foods' or legal wording regarding functional food labeling in Japan at the bottom of the page or in the product description area. This enables sellers to automatically comply with legal regulations without having to manually check local laws.

[0103] In the step of configuring the payment module to bind the final payment amount based on the local currency to be displayed on the payment screen, generating a unique identification token to create a dedicated landing link for each influencer recruited through an influencer campaign recruitment announcement, and setting the unique identification token to be stored in the client storage area when a user accesses the export-only sales page, the actual payment system is integrated and a technical foundation for tracking marketing performance is established.

[0104] The final amount calculated is linked to the module of the local Payment Gateway (PG) company. At the same time, a unique identification token is defined to be assigned when creating dedicated landing links for influencers to use in the future. When a consumer visits the page via the link, the device extracts the token included in the URL parameters and stores it in a client storage area, such as the consumer's browser cookie or local storage. This is to identify the original contributor to the visit, even if the consumer leaves without making an immediate purchase and returns a few days later to make a purchase.

[0105] In the step of inserting a tracking script into the export-only sales page, which calls a unique identification token stored in the client storage area and transmits it to the device (server) along with transaction data when a payment completion event occurs, a trigger that determines the marketing contribution is installed.

[0106] A script is inserted that executes when the user completes the payment and reaches the 'Order Completion Page (Thank you page)'. This script queries the client storage area to check for a stored influencer identification token, and if the token exists, it sends it to the server along with transaction data such as the order number and payment amount. This data serves as a key basis for future influencer settlements and ROAS analysis, enabling each purchase transaction to be mapped 1:1 to a specific influencer.

[0107] In addition, the step of generating the influencer campaign recruitment announcement comprises: loading the target country identifier, target audience, visual concept guide, key emphasis keywords for each product, and optimal budget allocation range from the localization branding strategy data; searching the influencer data pool using the target audience information as a query and calculating the audience fit, which is the ratio of the gender and age distribution of followers for each influencer that matches the target audience; extracting past post images of each influencer from the influencer data pool and inputting the past post images into a multimodal embedding model to convert them into post feature vectors; generating a reference image that visualizes the image filter tone, color contrast ratio, and object placement composition included in the visual concept guide, and inputting the reference image into a multimodal embedding model to generate a target concept vector; and calculating style similarity by calculating the cosine similarity between the target concept vector corresponding to the visual concept guide and the post feature vector. A step of extracting a group of influencer candidates in which the audience suitability and style similarity are above a predetermined standard and the ratio of bots or ghost accounts is below a preset threshold; a step of calculating, for each of the influencer candidates, a reach score based on the number of followers, an engagement score based on the average engagement rate, a similarity score between the category tag and the product category for sale, a performance score based on past sponsorship history, and a content suitability score; a step of calculating an influencer suitability index by weighted summing the reach score based on the number of followers, the engagement score based on the average engagement rate, the similarity score between the category tag and the product category for sale, the performance score based on past sponsorship history, the content suitability score, the audience suitability, and the style similarity.A step of classifying influencers into multiple grade ranges based on the influencer suitability index and automatically setting collaboration conditions for each grade, including payable sponsorship fees, product provision methods, performance-based incentives, and Key Performance Indicators (KPIs); a step of querying the posting registration specifications for each social media platform of the target country to generate posting template parameters including permitted phrase length, number of hashtags, possibility of link insertion, image / video upload specifications, and mandatory advertising disclosure items; a step of generating a list of required and prohibited hashtags combined based on the core emphasis keywords for each product, and generating a content creation guide by automatically inserting the standard image, required hashtags, and prohibited hashtag list into designated fields of the posting guide template; and a step of identifying paid advertising disclosure phrases and locations in the target country's language from the influencer advertising disclosure regulations of the target country stored in the database, and inserting them as mandatory provisions in the legal compliance section of the content creation guide. A step of setting an optimal upload window based on social media active traffic time zone data corresponding to the time zone of the target country and the category of the product to be sold, and specifying this as the recommended upload time that influencers must adhere to when uploading content in the content creation guidelines; a step of generating a basic announcement text including the category, price range, sales conditions, and shipping / return policy of the product to be sold based on the collaboration conditions and announcement template parameters, while restructuring the sentence structure so that key emphasis keywords for each product are placed at the beginning of the sentence and translating it into the target country language to generate a localized announcement text; and a step of generating a dedicated landing link for each influencer by combining a unique identification token with the landing URL included in the content creation guidelines.The method may include the steps of: creating a tracking link allocation slot within the database and mapping a dedicated landing link for each influencer to the tracking link allocation slot on a one-to-one basis to link with the data structure of the recruitment announcement; and creating a recruitment announcement packet containing the dedicated landing link for each influencer and content creation guidelines, and registering it according to the announcement registration API of the influencer recruitment platform.

[0108] The step of creating the aforementioned influencer campaign recruitment notice is a process of packaging and distributing a notice, content guidelines, collaboration conditions, and a performance tracking system to recruit influencers who will carry out the actual marketing, based on the previously established localization branding strategy.

[0109] In the step of loading target country identifiers, target audiences, visual concept guides, key emphasis keywords for each product, and optimal budget allocation ranges from the above localization branding strategy data, strategic assets derived by the artificial intelligence model are imported into the recruitment announcement generation module.

[0110] Here, the target audience includes demographic characteristics such as age, gender, and interests (e.g., K-beauty, vegan, parenting, etc.), while the visual concept guide includes previously derived design parameters (filter tone, composition, etc.). Based on this data, the device automatically selects influencer candidates optimized for country, product, and audience characteristics and utilizes them as foundational data for generating customized advertisements.

[0111] In the step of searching the influencer data pool (or influencer database) using the above-mentioned target audience information as a query and calculating the audience fit, which is the ratio of the gender and age distribution of each influencer's followers matching the target audience, the characteristics of the 'followers' rather than the influencer's own characteristics are analyzed to predict actual marketing effects.

[0112] The device searches an influencer database to load the gender and age distribution data of each influencer's followers, and then calculates the audience fit by calculating the intersection ratio between the distribution of the target audience and the distribution of the influencer's followers.

[0113] For example, if the target audience is 'American women in their 20s,' the device calculates the proportion of 'American women in their 20s' among each influencer's followers and assigns a score to this ratio. This is because, no matter how famous an influencer is, there is no marketing effect if potential customers who will purchase my product do not follow them.

[0114] In the step of extracting past post images of each influencer from the above influencer data pool and inputting the past post images into a multimodal embedding model to convert them into post feature vectors, the visual style of the influencer is mathematically quantified.

[0115] By collecting images previously uploaded by an influencer and inputting them into a multimodal embedding model (e.g., CLIP), information regarding the mood, color scheme, and subjects the influencer primarily handles is converted into high-dimensional vector values ​​called 'post feature vectors'.

[0116] In the step of generating a reference image that visualizes the image filter tone, color contrast ratio, and object placement configuration included in the above visual concept guide, and generating a target concept vector by inputting the reference image into a multimodal embedding model, a reference point for comparison is created.

[0117] Based on the 'Visual Concept Guide' derived earlier in the form of text or parameters, a virtual 'Reference Image' is synthesized using generative AI (e.g., Stable Diffusion). Subsequently, this reference image is passed through the same multimodal embedding model as the influencer's post to generate a 'Target Concept Vector,' thereby making it directly comparable to the influencer's post vector.

[0118] In the step of calculating style similarity by calculating the cosine similarity between the target concept vector and the post feature vector corresponding to the visual concept guide above, the degree of agreement (Brand Fit) between the brand image and the influencer style is measured.

[0119] The device calculates the cosine similarity between the target concept vector and the post feature vector to produce 'style similarity,' which quantifies how similar the visual concept pursued by our brand is to the influencer's usual style.

[0120] In the step of extracting a group of influencer candidates in which the audience suitability and style similarity are above a predetermined standard and the ratio of bots or ghost accounts is below a preset threshold, unsuitable targets are primarily filtered out.

[0121] Even if the audience fit and style fit are excellent, if the proportion of bot accounts or ghost accounts (inactive accounts) calculated by comprehensively analyzing posting frequency patterns, repeated hashtag usage, abnormal click-through rates, and the proportion of followers in the same IP range is above a certain level (e.g., 30%), they are excluded from the candidate pool due to the risk of abuse.

[0122] For each of the above-mentioned influencer candidate groups, the steps of calculating a reach score based on the number of followers, an engagement score based on the average engagement rate, a similarity score between the category tag and the product category to be sold, a performance score based on past sponsorship history, and a content suitability score; and calculating an influencer suitability index by weighting and summing these, thereby precisely evaluating the candidate groups.

[0123] The value of each influencer is quantified into a single score called the 'Influencer Suitability Index' by comprehensively considering not only the simple number of followers but also actual reach, engagement rate, sales or click performance during past promotions of similar products, and content image quality.

[0124] In the step of classifying influencers into multiple grade ranges based on the above influencer suitability index and automatically setting collaboration conditions including sponsorship fees payable for each grade, product provision methods, performance-based incentives, and target performance indicators (KPIs), contract conditions based on objective data are established.

[0125] For example, the group with a suitability index in the top 1% is classified as 'Tier 1' to set high fixed manuscript fees and free product provision, while the group in the top 10–30% is classified as 'Tier 3' to automatically match conditions based on product provision and sales commissions (affiliate). This reduces the time required for negotiation and establishes a reasonable compensation system.

[0126] In the step of generating announcement template parameters by querying the announcement registration specifications for each social media platform of the above-mentioned target country, announcement template parameters are generated by querying the allowed phrase length, maximum number of hashtags, possibility of link insertion, image / video upload specifications, and mandatory advertising disclosure items for each social media platform of the target country.

[0127] In the step of generating a content creation guide by creating a list of required and prohibited hashtags combined based on the core emphasis keywords for each of the above products, and automatically inserting the standard image, the list of required and prohibited hashtags, and the post guide template into designated fields, specific work instructions to be delivered to the influencer are completed.

[0128] The device automatically generates essential hashtags (e.g., #MustHave) to maximize marketing effectiveness and prohibited hashtags (e.g., competitor names) for brand safety. Additionally, by attaching previously generated 'reference images,' it helps influencers intuitively understand the visual concept and produce content while adhering to shooting compositions and tone and manner.

[0129] In the step of identifying paid advertisement disclosure phrases and locations in the language of the target country from the influencer advertisement disclosure regulations of the target country stored in the above database, and inserting them as mandatory provisions in the legal compliance section of the content creation guidelines; global regulatory compliance is automated.

[0130] For example, when targeting the U.S., provisions mandating the setting of "Paid Partnership" tags or the "#Ad" label at the top of the content are forcibly inserted into the guidelines in accordance with FTC regulations to preemptively block legal issues.

[0131] In the step of setting an optimal upload window based on social media active traffic time zone data corresponding to the time zone of the target country and the category of the product to be sold, and specifying it in the content creation guidelines as the recommended upload time that the influencer must adhere to when uploading content; the reach efficiency of the content is increased.

[0132] By analyzing local time zones and user activity patterns by category, the time zone with the best response (e.g., Friday 8 PM EST) is designated as the 'upload window', and guidance is provided to adhere to it.

[0133] In the step of generating a localized announcement text by creating a basic announcement text including the category, price range, sales conditions, and shipping / return policy of the products to be sold based on the above collaboration conditions and announcement template parameters, while restructuring the sentence structure so that key emphasis keywords for each product are placed at the beginning of the sentence and translating it into the target country language, the announcement text itself for influencer recruitment is localized.

[0134] Clarify collaboration conditions (compensation), structure sentences to effectively highlight the product's selling points (key keywords), and translate into the local language to increase the application rate of local influencers.

[0135] In the step of creating a dedicated landing link for each influencer by combining a unique identification token with the landing URL included in the above content creation guidelines; and the step of creating a tracking link allocation slot within the above database (specifically, a campaign management database) and mapping the dedicated landing link for each influencer to the tracking link allocation slot on a one-to-one basis to link with the data structure of the recruitment announcement; the system preparation for performance tracking is completed.

[0136] A unique tracking link is generated for each influencer in advance during the recruitment announcement phase and mapped to an 'assignment slot' within the system. This is to ensure that as soon as an influencer applies for and is selected for a campaign, the unique link is issued to them without any delay, and performance tracking begins immediately.

[0137] In the step of generating a recruitment announcement packet containing a dedicated landing link and content creation guidelines for each influencer and registering it according to the announcement registration API of an influencer recruitment platform, all finally generated data (announcement, guidelines, tracking link, collaboration conditions, etc.) are bundled into a packet form and automatically posted through an influencer recruitment platform or social media API in the target country to start campaign recruitment.

[0138] In addition, the step of quantitatively calculating the marketing contribution for each influencer comprises: a step of collecting raw engagement data including likes, comments, shares, saves, and reach of content posted by influencers by periodically polling the open API of the social media platform of the target country; a step of generating commerce conversion data including the number of incoming clicks, number of items added to cart, number of purchase conversions, and total payment amount for each influencer by parsing the transaction log received from the tracking script and matching access logs containing unique identification tokens with purchase completion logs; and a data validation step of analyzing access IP addresses, user agents, and time elapsed between clicks and purchases for the raw engagement data and commerce conversion data to identify and remove repeated clicks from the same IP range, abnormal traffic suspected of being bots, and exit traffic with a dwell time below a preset valid threshold. Based on the engagement raw data and commerce conversion data for which validation has been completed, the method may include the step of calculating the return on advertising spend (ROAS), cost per click (CPC), and cost per purchase (CPA), which are sales revenue relative to total input costs; calculating the achievement rate by comparing performance with key performance indicators (KPIs); and determining the final settlement amount for each influencer by applying the incentive calculation logic specified in the collaboration conditions.

[0139] In the step of periodically polling the open API of the social media platform of the target country to collect raw engagement data including likes, comments, shares, saves, and reach of content posted by an influencer, the diffusion power of the content created by the influencer and the public's reaction are tracked in real time.

[0140] The device performs a polling method to retrieve the latest data by periodically (e.g., every hour) sending requests to major social media platform servers in target countries. The Engagement Raw Data collected at this time is an unprocessed, raw response indicator that is stored along with post identifiers, influencer identifiers, and timestamps, and is accumulated in a time-series format that allows for the analysis of response trends over time.

[0141] In the step of generating commerce conversion data including the number of clicks, cart additions, purchase conversions, and total payment amount per influencer by parsing the transaction log received from the above tracking script and matching the access log containing a unique identification token with the purchase completion log; the extent to which the reaction on social media is connected to actual economic value (sales) is determined.

[0142] Transaction log parsing refers to the process of extracting only meaningful data from the vast amount of access records stored on a server and converting it into an analyzable format. The device uses a 'unique identification token' as an identification key from logs transmitted by tracking scripts embedded in sales pages to track which influencer's link a consumer clicked on (incoming click), whether they subsequently added items to their cart, and whether they ultimately made a payment (purchase conversion). For example, if cart events and payment completion events occur sequentially within a session initiated by a specific unique identification token, the corresponding revenue is attributed to the conversion performance of the corresponding influencer.

[0143] In the data validation step, which analyzes the access IP address, User-Agent, and time-to-purchase pattern for the above engagement raw data and commerce conversion data to identify and remove repeated clicks from the same IP range, abnormal traffic suspected of being bots, and exit traffic with a dwell time below a preset valid threshold, technical filtering is performed to prevent performance inflation or fraudulent receipt.

[0144] The device analyzes access IP addresses to invalidate instances where an abnormally high number of clicks occur within a specific range (suspected click farms), and analyzes User Agents (browser and device identification information) to identify and remove access by crawlers or script bots rather than standard web browsers. Additionally, it ensures data integrity by excluding instances from performance aggregation where the Time-to-Purchase is physically impossible to achieve (suspected macros) or the page dwell time is less than 3 seconds (simple accidental clicks or exits), deeming these to be below the valid threshold.

[0145] In the step of calculating the return on advertising spend (ROAS), cost per click (CPC), and cost per purchase (CPA) based on the above-mentioned engagement raw data and commerce conversion data for which validation has been completed, comparing the performance with the target performance indicators (KPIs) to calculate the achievement rate, and applying the incentive calculation logic specified in the collaboration conditions to determine the final settlement amount for each influencer, the final performance is evaluated and the reward is determined based on the validated data.

[0146] Return on Advertising Spend (ROAS) is calculated as 'Total Payment Amount / Sponsorship Fees and Commissions Paid', Cost Per Click (CPC) as 'Total Cost / Number of Valid Clicks', and Cost Per Purchase (CPA) as 'Total Cost / Number of Valid Purchases'. The device compares the Key Performance Indicators (KPIs) set at the start of the campaign with the actual achievement rate and automatically applies pre-agreed incentive logic (e.g., an additional 10% of sales paid upon exceeding KPIs, a reduction in the performance coefficient for non-achievement, etc.) to calculate the final settlement amount to be paid to the influencer. This automates the complex settlement process and realizes a transparent, data-driven compensation system.

[0147] An apparatus according to one embodiment includes a processor and memory. The processor may include at least one apparatus described above through the drawings or perform at least one method described above through the drawings. The memory may store information related to the method described above or store a program in which the method described above is implemented. The memory may be volatile memory or non-volatile memory.

[0148] The processor can execute a program and control the device. The code of the program executed by the processor can be stored in memory. The device can be connected to an external device (e.g., a personal computer or a network) through an input / output device (not shown in the drawing) and exchange data.

[0149] The embodiments described above may be implemented as hardware components, software components, and / or combinations of hardware and software components. For example, the devices, methods, and components described in the embodiments may be implemented using one or more general-purpose or special-purpose computers, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing unit may execute an operating system (OS) and one or more software applications executed on said operating system. Additionally, the processing unit may access, store, manipulate, process, and generate data in response to the execution of the software. For ease of understanding, the processing unit may be described as being used as a single unit, but those skilled in the art will understand that the processing unit may include multiple processing elements and / or multiple types of processing elements. For example, the processing unit may include multiple processors or one processor and one controller. Additionally, other processing configurations, such as parallel processors, are also possible.

[0150] The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., either alone or in combination. The program instructions recorded on the medium may be those specifically designed and configured for the embodiment, or they may be those known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The hardware devices described above may be configured to operate as one or more software modules to perform the operation of the embodiment, and vice versa.

[0151] Software may include computer programs, code, instructions, or a combination of one or more of these, and may configure a processing unit to operate as desired or instruct the processing unit independently or collectively. Software and / or data may be permanently or temporarily embodied in any type of machine, component, physical device, virtual equipment, computer storage medium, or device so as to be interpreted by the processing unit or to provide instructions or data to the processing unit. Software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

[0152] Although the embodiments have been described above with reference to the limited drawings, those skilled in the art can apply various technical modifications and variations based on the above. For example, suitable results may be achieved even if the described techniques are performed in a different order than described, and / or if the components of the described system, structure, device, circuit, etc. are combined or assembled in a form different from described, or replaced or substituted by other components or equivalents.

[0153] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims set forth below. Explanation of the symbols

[0154] 1 : Device 11: Data Collection Unit 12 : AI Processing Unit 13: Strategy Formulation Department 14 : Page creation section 15 : Campaign Management Department 16 : Performance Analysis Department 17 : Database 2 : Social Media Server 3 : Korea Customs Service Server 4 : Payment Server S100: The step of analyzing performance indicators for the monthly report S200: Step to generate localization branding strategy data S210: Step to convert post images and text into latent vectors in a common latent space S220: Step to calculate the center vector S230: Step to generate product embedding vectors S240: Step to select the target cluster S250: A step of extracting the average value of image filter tones, the distribution probability of object placement, and the most frequently occurring adjective in the text, and defining them as localization components. S260: Step to fine-tune the participation rate prediction model S270: The stage of finalizing the design combination with the highest predicted participation rate as a visual concept guide. S280: Step to generate key emphasis keywords for each product S290: Step to determine the optimal budget allocation range S300: Step to create an export-only sales page S310: The step of rendering the user interface S320: Steps to perform local search engine optimization S330: Step to generate localized content S340: Step to calculate the final payment amount in local currency S350: Step to automatically insert mandatory legal notices and limitation of liability clauses into the notice section at the bottom of the page S360: A step of configuring the unique identification token to be stored in the client storage area when the user accesses the export-only sales page. S370: Step to insert the tracking script into the export-only sales page S400: Step to create an influencer campaign recruitment notice S500: The step of quantitatively calculating marketing contribution

Claims

Claim 1 A method for providing a solution for local branding and export commerce process automation utilizing an artificial intelligence model-based country-specific influencer marketing, wherein the device includes a processor, memory, a communication module, and a non-transient storage medium, and the processor executes a program stored in the non-transient storage medium, comprising: a step of collecting performance data for previously performed marketing activities to generate a monthly report and analyzing performance indicators of the monthly report; a step of inputting the monthly report and local market data of the target country into an artificial intelligence model to generate localized branding strategy data including an optimal budget allocation range for future marketing, a visual concept guide for content based on the preference trends of the target country, and key emphasis keywords for each product; a step of generating an export-only sales page based on the localized branding strategy data, the language of the target country, a local currency payment system, and customs and shipping policies; and a step of generating an influencer campaign recruitment notice based on the localized branding strategy data and the export-only sales page, including a posting guide indicating the direction of the content. The method includes the step of collecting result data of influencer activities performed according to the above posting guide and quantitatively calculating the marketing contribution of each influencer; and the step of generating the localization branding strategy data comprises: crawling top-ranking posts on social media platforms of the target country to collect a post data set, and converting post images and text into latent vectors in a shared latent space using a multi-modal embedding model; classifying the latent vectors into multiple trend clusters through a clustering algorithm and calculating the center vector of each trend cluster; and inputting the detailed page text and product images of the product to be sold into the multi-modal embedding model to generate product embedding vectors.A step of calculating the cosine similarity between the product embedding vector and the center vector of each trend cluster, and selecting target clusters where the similarity is above a preset threshold; a step of defining localization components by extracting the average value of image filter tone, the distribution probability of object placement, and the most frequently occurring adjectives within the text from a set of reference posts belonging to the target clusters; a step of fine-tuning an engagement rate prediction model using the set of reference posts as training data, wherein the model takes visual attributes including image filters, composition, and objects as input and outputs an engagement rate calculated based on the number of likes and comments on the posts; a step of calculating a predicted engagement rate by inputting multiple virtual design combinations that can be generated based on the localization components into the engagement rate prediction model, and confirming the design combination with the maximum predicted engagement rate as a visual concept guide; a step of generating a keyword pool by extracting nouns and adjectives from the text of the set of reference posts, and generating product-specific core emphasis keywords by selecting long-tail keywords with low competition intensity relative to search volume from among the keywords in the keyword pool. and includes the step of estimating the Conversion Rate (CVR) curve relative to the budget based on marketing performance data of past similar product lines, deriving inflection points for achieving target sales, and determining the optimal budget allocation range; and the step of creating the export-only sales page comprises: converting image filter tone and color contrast ratio data included in the visual concept guide into hexadecimal color codes and Cascading Style Sheets (CSS) style property values, and dynamically injecting the CSS style property values ​​into style sheets defining the background, buttons, and highlighted text of the export-only sales page to render a user interface that matches the preference trends of the target country;A step of performing local search engine optimization (SEO) by placing the key emphasis keywords for each product in the meta title and meta description within the head tag of the export-only sales page according to the priority of the key emphasis keywords for each product; a step of generating localized content by translating the detailed description text of the product to be sold into the language of the target country, while restructuring the sentence containing the key emphasis keywords for each product so that the corresponding keywords are placed at the beginning of the sentence while maintaining the meaning of the original text, detecting the text area within the product image to be included in the export-only sales page, performing background restoration (in-painting), and then overlaying the translated text; a step of calculating the estimated customs duty by linking the HS Code (Harmonized System Code) of the product to be sold with the customs rate data of the target country, converting the product price into the local currency by calling a real-time exchange rate API, and calculating the final payment amount in the local currency by reflecting the estimated customs duty and the shipping fee policy based on the shipping zone by country. A step of determining whether the category of the product to be sold falls under a specific regulatory target recorded in the database, and if so, querying local legal data of the target country to automatically insert mandatory legal notices and disclaimer clauses into the notice section at the bottom of the page; a step of configuring the payment module to bind the final payment amount in the local currency to be displayed on the payment screen, generating a unique identification token to create a dedicated landing link for each influencer recruited through an influencer campaign recruitment announcement, and setting the unique identification token to be stored in the client storage area when a user accesses the export-only sales page;The method includes the step of inserting a tracking script into the export-only sales page, which retrieves a unique identification token stored in the client storage area and transmits it to the device along with transaction data when a payment completion event occurs; and the step of generating the influencer campaign recruitment announcement comprises: loading a target country identifier, target audience, visual concept guide, key emphasis keywords for each product, and optimal budget allocation range from the localization branding strategy data; searching an influencer data pool using the target audience information as a query and calculating an audience fit rate, which is the ratio of the gender and age distribution of followers for each influencer that matches the target audience; extracting past post images of each influencer from the influencer data pool and inputting the past post images into a multimodal embedding model to convert them into post feature vectors; generating a reference image that visualizes the image filter tone, color contrast ratio, and object placement configuration included in the visual concept guide, and inputting the reference image into a multimodal embedding model to generate a target concept vector. A step of calculating style similarity by calculating the cosine similarity between a target concept vector and a post feature vector corresponding to the visual concept guide; a step of extracting a group of influencer candidates in which the audience suitability and style similarity are above a predetermined standard and the ratio of bots or ghost accounts is below a preset threshold; and a step of calculating, for each of the influencer candidates, a reach score based on the number of followers, an engagement score based on the average engagement rate, a similarity score between the category tag and the product category for sale, a performance score based on past sponsorship history, and a content suitability score.A step of calculating an influencer suitability index by weighted summing the reach score based on the number of followers, the engagement score based on the average engagement rate, the similarity score between the category tag and the product category for sale, the performance score based on past sponsorship history and the content suitability score, and the audience suitability and style similarity; a step of classifying influencers into multiple grade ranges based on the influencer suitability index, and automatically setting collaboration conditions including payable sponsorship fees, product provision methods, performance-based incentives, and Key Performance Indicators (KPIs) for each grade; a step of generating announcement template parameters including allowed phrase length, number of hashtags, whether links can be inserted, image / video upload specifications, and mandatory advertising disclosure items by querying the announcement registration specifications for each social media platform of the target country; and a step of generating a content creation guide by generating a list of required and prohibited hashtags combined based on the core emphasis keywords for each product, and automatically inserting the reference image, the list of required and prohibited hashtags, into designated fields of the posting guide template. A step of identifying paid advertisement disclosure phrases and locations in the language of the target country from the influencer advertisement disclosure regulations of the target country stored in the database, and inserting them as mandatory provisions in the legal compliance section of the content creation guidelines; a step of setting an optimal upload window based on social media active traffic time zone data corresponding to the time zone of the target country and the category of the product to be sold, and specifying in the content creation guidelines the recommended upload time that the influencer must comply with when uploading content;A method for providing a solution for automating local branding and export commerce processes utilizing AI model-based country-specific influencer marketing, comprising: generating a basic announcement text including the category, price range, sales conditions, and shipping / return policy of the product to be sold based on the collaboration conditions and announcement template parameters, reconstructing the sentence structure so that key emphasis keywords for each product are placed at the beginning of the sentence, and translating it into the target country language to generate a localized announcement text; generating an influencer-specific landing link by combining a unique identification token with the landing URL included in the content creation guidelines; creating a tracking link allocation slot within the database and mapping the influencer-specific landing link to the tracking link allocation slot on a one-to-one basis to link with the data structure of the recruitment announcement; and generating a recruitment announcement packet including the influencer-specific landing link and content creation guidelines and registering it according to the announcement registration API of the influencer recruitment platform. Claim 2 delete Claim 3 delete

Images