An ai-based business strategy recommendation method and system

Abstract

The application discloses an AI-based business strategy recommendation method and system, which converts store operation data into multi-dimensional store labels and feature vectors through a data processing and labeling module; a vector database module stores the feature vectors of benchmark stores and their historical operation strategies; a recommendation engine module calculates the similarity between the target store and the benchmark store, filters the candidate benchmark store, calculates the comprehensive score according to the scoring function, and generates a strategy recommendation list; an AI Agent execution system analyzes the selected strategy, converts it into an executable task and automatically executes it; an effect tracking and evaluation module records the execution results to form feedback data for optimizing the recommendation parameters. The application realizes personalized recommendation, high accuracy, strong interpretability, executable landing and continuous optimization capability, reduces the AI usage threshold, and improves the operation decision efficiency.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and big data analysis technology, specifically relating to an AI-based business strategy recommendation method and system. Background Technology

[0002] With the development of big data and artificial intelligence technologies, more and more companies are starting to use AI to support business decision-making. The core logic of existing AI-based business strategy recommendation algorithms is usually to train models based on publicly available data from various industries, and then push general business strategies to users.

[0003] The existing implementation scheme most similar to this invention is as follows: In the data collection stage, macro market data is collected from publicly available industry reports, statistical yearbooks, and other channels; in the model training stage, a general business strategy recommendation model is trained using the collected publicly available data, such as a collaborative filtering or simple regression model; in the strategy generation stage, after the user inputs their basic information, the model generates and pushes business suggestions based on pre-trained rules; in the result display stage, the strategy is pushed to the user in the form of a text list.

[0004] However, through causal reasoning analysis, the existing technology has the following inherent technical defects: First, it lacks personalized causal relationships. Because the data source is publicly available data from a broad industry, there is no direct causal link between it and the specific operating conditions of individual companies. The model cannot understand personalized causal logic such as "because Company A has a low average order value, it needs to offer package deals," and can only provide general suggestions based on statistical correlations, resulting in recommendations lacking specificity.

[0005] Secondly, the causal chains for insufficient recommendation accuracy are missing. Simple recommendation algorithms (such as collaborative filtering) cannot deeply explore the complex causal relationships between a company's historical data and future market trends. For example, the model cannot identify the causal chain of "because competitors recently launched new products, our store's customer traffic has decreased, so we should strengthen marketing to existing customers," resulting in delayed and ineffective recommendation results.

[0006] Third, there's the causal cycle of outdated knowledge bases and lack of feedback. Existing solutions typically lack dynamic data update mechanisms, causing the AI's knowledge base to become disconnected from the rapidly changing market environment. More importantly, because a closed-loop causal relationship of "recommendation-execution-effect" has not been established, the system cannot optimize its model based on causal feedback, resulting in a failure to continuously improve recommendation effectiveness.

[0007] Fourth, the black-box problem leads to a lack of trust. Existing technologies are typically end-to-end black-box models, where enterprise users can only see "what" but cannot understand "why." Due to the lack of causal explanations for the logic behind the recommendation results, enterprises find it difficult to build trust in AI recommendations, resulting in low adoption rates.

[0008] Fifth, the execution gap hinders the effective implementation of strategies. Recommended strategies remain merely textual descriptions, creating an execution gap between the strategies and the company's actual business systems. The transition from "receiving advice" to "implementation" requires significant human intervention, preventing good strategies from being effectively translated into operational improvements.

[0009] Therefore, providing an AI-based business strategy recommendation method and system that can solve the above problems, achieve personalized recommendations, high accuracy, strong interpretability, practical implementation, and continuous optimization capabilities is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0010] To address the aforementioned technical problems, the present invention aims to provide an AI-based business strategy recommendation method and system, overcoming the technical deficiencies of existing technologies, such as lack of personalization in recommendations, insufficient accuracy, outdated knowledge base, lack of interpretability in recommendation results, inability to implement strategies, and lack of feedback optimization mechanisms.

[0011] To address the aforementioned technical problems, this invention constructs a complete causal chain of data, matching, execution, and feedback to achieve personalized recommendations, highly accurate, interpretable, implementable, and continuously optimized business strategy recommendations.

[0012] The system adopts a modular architecture, consisting of a data processing and labeling module, a vector database module, a recommendation engine module, an AI Agent execution system, an effect tracking and evaluation module, and an output module. The modules work together to form a complete closed-loop system.

[0013] Specifically, this invention transforms unstructured enterprise operational data into structured store tags and feature vectors through a data processing and tagging module; stores the feature vectors of exemplary stores and their associated historical operational strategies through a vector database module; filters candidate exemplary stores and generates a strategy recommendation list through a recommendation engine module based on cosine similarity matching and weighted scoring; parses the selected strategies and transforms them into executable tasks through an AI Agent execution system; and records the execution results and generates feedback data through an effect tracking and evaluation module to optimize the recommendation algorithm parameters. Thus, this invention achieves a complete closed loop from data to recommendation, from recommendation to execution, and from execution to feedback, significantly improving the personalization, accuracy, interpretability, and feasibility of recommendations.

[0014] The system also establishes a T+1 periodic update mechanism to ensure the timeliness of the knowledge base; enhances the interpretability of recommendation results by displaying the operating data of similar successful stores; ensures the compliance of strategy execution through store approval processes; and dynamically adjusts the weight parameters of the recommendation algorithm through feedback data to achieve self-optimization and evolution of the model.

[0015] According to one aspect of the present invention, an AI-based business strategy recommendation method is provided, comprising the following steps: Obtain operational data from the target store; Feature extraction is performed on the operational data to generate multi-dimensional store tags for the target store, and the store tags are converted into a first feature vector; Obtain the second feature vector of multiple model stores. The model stores are stores whose operating performance data meets a preset threshold. The second feature vector is generated by converting the store labels of the model stores. Each model store has its historical operating strategies and operating performance data stored in association. Calculate the similarity between the first feature vector and each of the second feature vectors; Based on the similarity, at least one candidate model store with a similarity greater than a first threshold to the target store is selected from a plurality of model stores. Based on the operational performance data of each candidate model store, the historical operational strategies of each candidate model store are sorted, and a strategy recommendation list for the target store is determined based on the sorting results and output to the target store. In response to the target store's confirmation instruction for the selected historical business strategy in the strategy recommendation list, the selected historical business strategy is input into the AI ​​Agent execution system, which then converts the selected historical business strategy into an executable task and executes it automatically.

[0016] Another aspect of the present invention provides an AI-based business strategy recommendation system, comprising: The data processing and tagging module is used to acquire the operating data of the target store, extract features from the operating data, generate multi-dimensional store tags for the target store, and convert the store tags into a first feature vector. The vector database module is used to store the second feature vectors of multiple model stores, as well as the historical operating strategies and operating performance data associated with each model store. The model stores are those whose operating performance data meets a preset threshold. The second feature vectors are generated by converting the store labels of the model stores. The recommendation engine module is used to calculate the similarity between the first feature vector and each of the second feature vectors, select at least one candidate model store from the multiple model stores whose similarity to the target store is greater than a first threshold based on the similarity, sort the historical operating strategies of each candidate model store based on the operating performance data of each candidate model store, and determine the strategy recommendation list of the target store based on the sorting result. The AI ​​Agent execution system is used to parse the historical business strategies selected in the strategy recommendation list, obtain the strategy parsing results using natural language processing technology, transform the strategy parsing results into one or more specific executable tasks according to the preset strategy-task mapping rule base, generate an execution plan, and automatically call the business system to execute the execution plan through the API interface. The effect tracking and evaluation module is used to track and record the execution results of the execution plan, obtain and associate the historical business effect data before and after the execution of the business strategy, form feedback data, and provide the feedback data to the recommendation engine module to optimize the recommendation parameters of the recommendation engine module; The output module is used to output the strategy recommendation list to the target store.

[0017] Compared with the prior art, the present invention has the following beneficial effects: First, it enables personalized recommendations. This invention analyzes a company's historical data, market trends, and industry characteristics to establish multi-dimensional store tags. By establishing causal relationships with similar successful stores, it provides truly targeted business strategy recommendations to meet the specific needs of different companies.

[0018] Second, it improves recommendation accuracy. This invention utilizes vector matching and weighted scoring algorithms to uncover the causal relationship between similar features and excellent business results. By using the strategies of exemplary stores as the basis for recommendations, it significantly improves the accuracy and effectiveness of the recommendation results.

[0019] Third, a data-driven causal feedback loop is established. This invention employs a T+1 data update mechanism to ensure the timeliness of the knowledge base. Simultaneously, by tracking the execution effect of the recommendation strategy, a closed-loop feedback loop between the recommendation strategy and operational results is formed. The recommendation model parameters are dynamically optimized using the effect data, enabling the algorithm to self-evolve and continuously improve recommendation performance.

[0020] Fourth, it enhances the interpretability of recommendation results. By showcasing successful model stores similar to the target store and their operational data, this invention provides clear causal references for the recommendation results, allowing businesses to understand the logic behind the recommendations, thereby enhancing the credibility and acceptance of the recommendations.

[0021] Fifth, it establishes a seamless causal pathway from strategy to execution. This invention introduces AI Agent technology to transform recommended textual strategies into executable operational instructions, which are then automatically executed after an approval process. This bridges the final gap between intelligent decision-making and business implementation, ensuring that good strategies are effectively translated into good results.

[0022] Sixth, it enables traceability and continuous optimization. This invention structurally stores recommendation execution status and performance data into the system, establishes a relationship between recommendation ID, model stores, and strategy type, supports subsequent auditing and algorithm optimization, and provides a data foundation for continuously improving recommendation effectiveness. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of the structure of an AI-based business strategy recommendation system provided in an embodiment of the present invention; Figure 2 This is a flowchart of an AI-based business strategy recommendation method provided in an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] The embodiments of this invention are written in a progressive manner. Example 1: System Architecture

[0027] like Figure 1 As shown, this invention provides an AI-based business strategy recommendation system, the core of which lies in constructing a complete causal chain of data, matching, execution, and feedback through a modular architecture. The system consists of six core modules that work together to achieve a complete closed loop from data input to strategy execution and then to feedback optimization.

[0028] Specifically, the system includes the following functional modules: The data processing and tagging module acquires operational data from target stores, extracts features from this data, generates multi-dimensional store tags, and converts these tags into a first feature vector. The module's inputs include basic store information (brand, type, location, establishment time, area), historical transaction data (sales revenue, gross profit margin, customer traffic, inventory turnover rate, sales growth rate), and market environment data (number of competitors, regional population density, consumption level index). The processing includes data cleaning, standardization, and feature extraction. Based on a pre-defined tagging rule base, features are mapped to corresponding multi-dimensional store tags, including static tags (business format tags, scale tags, regional tags) and dynamic tags (growth tags, efficiency tags, performance tags, problem tags). The final output is structured store tags and a first feature vector. Technically, it employs data preprocessing algorithms and feature engineering techniques, including Z-score standardization and embedding model vectorization.

[0029] Detailed definition of the tag rule base: Static tag rule definition 1. Business Format Tag - Store Type 2. Size Tag - Rent Level 3. Size tag - Employee size 4. Location Tags - City Level 5. Regional Tags - Business District Level 6. Geographic Tags - Location Attributes Is it facing the street? Industry complementarity Dynamic tag rule definition 1. Growth Tag - Revenue Growth 2. Growth Tag - Profit Trend 3. Performance Tags - Performance Rating Assess the overall performance of stores by combining revenue growth and profit trends: 4. Efficiency Tag - Operational Efficiency 5. Problem Tag - Business Issues Identify actual operational problems in stores and support multiple tags: The vector database module stores the second feature vectors of multiple model stores, along with their associated historical operating strategies and performance data. Model stores are those whose performance data meets a preset threshold. The second feature vector is generated from the model stores' store tags. The core of this module is building a vector index, supporting fast similarity queries. Stored content includes store tag vectors, operating data vectors, and strategy vectors. The update mechanism uses a T+1 periodic update to ensure data timeliness. Technically, vector databases such as Milvus or Faiss can be used.

[0030] The recommendation engine module calculates the similarity between a first feature vector and each second feature vector. Based on this similarity, it selects at least one candidate model store from multiple model stores whose similarity to the target store is greater than a first threshold. It then ranks the historical operating strategies of each candidate model store based on their operational performance data, and determines the strategy recommendation list for the target store based on the ranking results. The core algorithms of this module include cosine similarity calculation, KNN nearest neighbor query, and weighted scoring. The processing involves querying the target store, retrieving similar stores, selecting excellent stores, and generating recommendations. The output is the ranked strategy recommendation list and similarity scores.

[0031] The output module is used to output the strategy recommendation list to the target store.

[0032] The AI ​​Agent execution system parses selected historical business strategies from the strategy recommendation list. It employs natural language processing technology to analyze these strategies, obtaining analysis results. Based on a pre-defined strategy-task mapping rule base, the system transforms these results into one or more specific executable tasks, generating an execution plan. This plan is then automatically executed by calling the business system via an API interface. The core processes of this module include strategy parsing, operationalization, approval processes, and automated execution. Its key features are reduced manual intervention and improved execution efficiency.

[0033] The performance tracking and evaluation module tracks and records the execution results of the implementation plan, acquires and correlates historical operational performance data before and after the implementation of business strategies, generates feedback data, and provides this feedback data to the recommendation engine module to optimize its recommendation parameters. The metrics recorded in this module include strategy execution rate, actual performance improvement, and ROI, using data accumulation to provide feedback for future optimizations.

[0034] The above six functional modules work together to form a complete closed-loop system. Users input operational data from the target store, which is then processed and tagged by the data processing and tagging module to form a first feature vector. The recommendation engine module calculates the similarity between this first feature vector and the second feature vectors of exemplary stores stored in the vector database module, filters out candidate exemplary stores, and sorts their historical operational strategies based on the operational performance data of each candidate exemplary store to determine a strategy recommendation list for the target store. The output module outputs the strategy recommendation list to the target store. In response to the target store's confirmation instruction for the selected historical operational strategy in the strategy recommendation list, the AI ​​Agent execution system converts the selected historical operational strategy into an executable task and executes it automatically. The performance tracking and evaluation module records the execution results and generates feedback data, which is then provided to the recommendation engine module to optimize recommendation parameters.

[0035] All modules communicate with each other through well-defined interfaces, ensuring the scalability and stability of the system.

[0036] Specifically, the main interface definitions between modules are as follows: 1. Data Processing and Tagging Module → Vector Database Module Interface: - Input: Store operating data - Output: Store tag JSON, feature vector (256 dimensions) - Invocation method: Asynchronous write, batch update on T+1 2. Recommendation Engine Module ↔ Vector Database Module Interface: - Query Interface: Input the first feature vector, return the top-K similar stores and their feature vectors. - Invocation method: Synchronous RPC call, response time <200ms 3. Recommended Engine Module → Output Module Interface: - Input: Strategy recommendation list (including strategy ID, strategy content, similarity score, and reason for recommendation) - Output: Formatted front-end display data - Invocation method: Event-driven, real-time push 4. Output module ↔ AI Agent execution system interface: - Input: The selected policy ID and confirmation command - Output: Execution plan approval request - Invocation method: Message queue, asynchronous processing 5. AI Agent Execution System ↔ Business System Interface: - Input: Executable task (including operation type, parameters, and time frame) - Output: Task execution status, execution result - Invocation method: RESTful API, synchronous / asynchronous hybrid 6. Performance tracking and evaluation module → Recommendation engine module interface: - Input: Feedback data (including strategy adoption rate, execution completion rate, ROI, etc.) - Output: Optimized recommended parameters - Invocation method: T+1 batch update, configuration center push In this embodiment, the system is implemented using the following technologies: the front end is implemented using a Web framework, which supports dynamic rendering and real-time updates of table components; the back end is implemented using a server-side framework, which is responsible for the business logic of each module; the vector database uses Milvus or Faiss, which supports second-level queries of millions of vectors; the data storage system uses a relational database, which is used to persistently store store data, strategy data, and feedback data; and the long-connection service uses an independent server to support high-concurrency real-time communication.

[0037] Throughout the system's operation, data flow strictly adheres to temporal and dependency constraints. User input of operational data triggers data processing; the generated feature vectors are used for similarity calculations; the selected candidate model stores and their strategies are used to generate a recommendation list; the selected strategies are transformed into executable tasks by the AI ​​Agent execution system and executed; the execution results are fed back to the performance tracking and evaluation module to form feedback data, which is used to optimize recommendation parameters. All modules communicate through well-defined interfaces, ensuring the system's scalability and stability. Example 2: General Process

[0038] like Figure 2 As shown, this invention provides an AI-based business strategy recommendation method, comprising the following steps: S1: Obtain the operating data of the target store This step is performed by the data processing and tagging module to obtain the target store's operational data. The operational data includes basic store information (brand, type, geographical location, establishment time, area), historical transaction data (sales revenue, gross profit margin, customer traffic, inventory turnover rate, sales growth rate), and market environment data (number of competitors, regional population density, and consumer spending index).

[0039] S2: Generate multi-dimensional store labels and convert them into the first feature vector. This step is performed by the data processing and tagging module, which extracts features from the operational data, generates multi-dimensional store tags for the target stores, and converts the store tags into the first feature vector.

[0040] Specifically, static features are extracted from basic store information and mapped to corresponding static tags according to a pre-defined tag rule base. Static tags include business type tags, scale tags, and region tags. Dynamic features are extracted from historical transaction data and market environment data and mapped to corresponding dynamic tags according to the tag rule base. Dynamic tags include growth tags, efficiency tags, performance tags, and problem tags. The tag rule base stores the correspondence between feature values ​​and tag values.

[0041] The extracted features are standardized using Z-score, with the following formula: ; in, This is the standard value. The original data values, The mean of all data. denoted as the standard deviation of all data.

[0042] The vectorization method is used to convert the features into the first feature vector: the embedding model is used to convert the text features (brand, type) into vectors, and the numerical features are directly used as vector components. The complete vector dimension of the store is 256 dimensions.

[0043] S3: Obtain the second feature vector of the model store This step is performed by the vector database module, which retrieves the second feature vectors of multiple model stores. Model stores are those whose operational performance data meets preset thresholds, such as stores ranking in the top 10% for indicators like revenue growth rate and profit margin. The second feature vectors are generated by converting the store labels of the model stores, and each model store is associated with and stored with its historical operational strategies and performance data. The vector database uses Milvus or Faiss as its storage and indexing engine, supporting sub-second queries of millions of vectors.

[0044] S4: Calculate the similarity between the first feature vector and each of the second feature vectors. This step is performed by the recommendation engine module, which calculates the similarity between the first feature vector and each of the second feature vectors. The cosine similarity algorithm is used, and the calculation formula is as follows: ; Where A is the first feature vector and B is the second feature vector; The range of similarity values ​​is: A value of 1 indicates that they are exactly the same, a value of 0 indicates that they are completely unrelated, and a value of -1 indicates that they are completely opposite.

[0045] In practical applications, a similarity score > 0.8 indicates strong similarity, > 0.7 indicates moderate similarity, and > 0.6 indicates weak similarity.

[0046] S5: Filter candidate model stores based on similarity This step is performed by the recommendation engine module, which selects at least one candidate model store from multiple model stores whose similarity to the target store is greater than a first threshold, based on similarity.

[0047] S6: Determine and output the recommended strategy list. This step is performed by the recommendation engine module. Based on the operating performance data of each candidate model store, the historical operating strategies of each candidate model store are sorted, and the strategy recommendation list for the target store is determined based on the sorting results and output to the target store.

[0048] Specifically, the first step is to select similar stores using a scoring function, the formula of which is: ; in, For the first The similarity between model stores and target stores. For the first Revenue growth rate of exemplary stores For the first The profit margin of a model store For the first The operational efficiency of exemplary stores , , , The corresponding weighting coefficients are set with initial weights differentiated based on the industry of each store, as follows: Industry Initial Weighting Rules Table The above are the initial weights, which will be dynamically adjusted based on actual feedback data from the stores. Each store will maintain its own set of weights, rather than adjusting them uniformly across the entire store.

[0049] Then, a strategy fit assessment is performed, evaluating the fit of the strategy for each similar store using the following formula: ; in, This refers to the alignment between historical operating strategies and target stores, used to assess whether the strategy is suitable for the characteristics of the target stores. This is used to assess the historical effectiveness of business strategies and evaluate their performance in similar stores. To assess the feasibility of implementing historical business strategies, this is used to evaluate the implementation capabilities and resources of target stores. , , These are the corresponding weighting coefficients, which are 0.4, 0.4, and 0.2, respectively.

[0050] It should be noted that the weighting coefficients , , It is not fixed, but dynamically optimized based on data accumulated during system operation. and Periodic adjustments are made based on similarity matching performance and historical strategy effectiveness data. As feasible data accumulates, the weights are gradually increased, eventually reaching a stable value. The optimized weights take effect in the next recommendation cycle, enabling the recommendation model to continuously evolve.

[0051] Similarity weight The basis for the 0.4 similarity score is as follows: Literature research on similar stores shows that similarity is fundamental to recommendation effectiveness (Koren et al., 2009). Industry practice typically places similarity weights between 30% and 50%. This system's multi-dimensional tags ensure high similarity credibility, thus increasing it to 40% to ensure the recommendation strategy is "appropriate."

[0052] Validity weight The basis for the 0.4 is as follows: T+1 performance data is the most reliable evaluation indicator (actual execution results). Historical performance in industry practice typically carries a weight of 40%-50%. Our system has a sufficiently large sample size (thousands of recommendations per month), ensuring statistical validity and prioritizing strategies that have been proven effective.

[0053] Feasibility weight The basis for a feasibility score of 0.2: Feasibility is difficult to quantify, and initially relies on rule sets and heuristic judgments. In industry practice, the feasibility weight is usually between 10% and 20%. Progressive design: As the system runs, the accumulated feasibility data will gradually increase its weight, preventing the selected solution from being too complex or exceeding the system's capabilities.

[0054] , , Dynamic optimization mechanism of weights The weighting coefficient , , Employ a periodic optimization mechanism based on global performance data: 1. Optimization cycle: Global optimization is performed every T+30 days. 2. Data Statistics: Statistics on recommendation execution data for all stores over the past 30 days, including: adoption rate for each strategy type; execution completion rate for each strategy type; average ROI for each strategy type; and correlation coefficient between similarity matching and final results. 3. Optimize rules: - If the correlation coefficient between the adoption rate and similarity matching degree of a certain strategy is >0.7, then the adoption rate should be appropriately increased. Weight - If the correlation coefficient between the historical effectiveness score and the actual ROI of a certain type of strategy is >0.7, then the score should be appropriately increased. Weight - If the correlation coefficient between the feasibility score and the execution completion rate is >0.7, then the score should be appropriately increased. Weight - Adjust the step size to 0.05. After adjustment, ensure... =1.

[0055] The correlation coefficient is calculated using the Pearson correlation coefficient, and the formula is: ; in, and There are two sets of variable sequences to be evaluated. and These are their means, The sample size. Correlation coefficient. The value range is [-1, 1], and |r|>0.7 indicates a strong correlation.

[0056] 4. Convergence Criterion: When the adjustment magnitude of three consecutive optimizations is less than 0.01, the weights are considered to have converged, and the cycle transitions to a quarterly optimization period. Dynamic weight adjustment scheme (based on store feedback) In recommendation engines, similarity ( ), income growth ( ), profit margin ), operational efficiency The weights of the four factors have a significant impact on the recommendation results. To enable the recommendation algorithm to self-optimize and adapt to the actual situation of different stores, a dynamic weight adjustment scheme based on store feedback is adopted.

[0057] Weight management system: Global weight: The default weight for newly added stores or stores with no feedback, such as [0.30, 0.30, 0.20, 0.20]. Store-specific weights: Each store maintains a set of personalized weights based on its historical feedback, such as... [0.25, 0.30, 0.20, 0.25] The reasons for not accepting the proposal and the corresponding weighting adjustment rules Weight adjustment process: 1. When generating recommendations, query the store weight configuration table: if a store has a personalized weight, use the personalized weight; otherwise, use the global weight. 2. Calculate the recommendation score ; 3. Push notifications to stores for display; 4. User actions: If the user accepts the application or receives no feedback, the weight remains unchanged; if the user rejects the application, a reason selection form will pop up. 5. After the user selects a reason, the results are obtained according to the rule table. Values, application adjustments: ; 6. Weight normalization: ,make sure =1.0; 7. Ensure the weights are within the valid range (0.10 ≤ ≤0.45); 8. Save to the store weight configuration table, and use the new weight when recommending to this store next time; Example of weight adjustment calculation: Stores Initial feedback indicates that the store belongs to the catering industry. According to the industry's initial weighting rule table, the initial global weight is [0.30, 0.30, 0.20, 0.20]. The user selected reason "1" for rejection (insufficient similarity); Look up the rule table: =-0.05, =0, =0, =0; Weighting adjustment: =0.25, =0.30, =0.20, =0.20, total =0.95; Normalization: Scale factor = 1.0 / 0.95 = 1.053; =0.25×1.053=0.263≈0.26; =0.30×1.053=0.316≈0.32; =0.20×1.053=0.211≈0.21; =0.20×1.053=0.211≈0.21; final Individual weight = [0.26, 0.32, 0.21, 0.21]; Weight adjustment protection mechanism: -Lower limit protection: Any When the value is less than 0.10, it is forcibly set to 0.10, and other weights are rescaled proportionally. -Maximum protection: Any When the value is greater than 0.45, it is forcibly set to 0.45, and other weights are rescaled proportionally. - Imbalance Warning: : A warning is triggered when the ratio is >3:1, and manual review is recommended. - Consecutive Reasons Restriction: If the same reason is reported more than 5 times consecutively, the adjustment of this factor will be suspended, and manual review is recommended.

[0058] S7: AI Agent execution system automatically executes This step is executed by the AI ​​Agent execution system. In response to the target store's confirmation instruction for the selected historical business strategy in the strategy recommendation list, the selected historical business strategy is input into the AI ​​Agent execution system. The AI ​​Agent execution system then transforms the selected historical business strategy into an executable task and executes it automatically.

[0059] Specifically, natural language processing technology is used to analyze the selected historical business strategies, extract key information from the strategies, identify the operation type (sales, marketing, inventory, pricing, etc.), target indicators and time frame, and identify dependencies and preconditions.

[0060] Translate high-level strategies into specific tasks. For example, the original strategy "improve sales conversion rate" can be transformed into: Task 1: Optimize store layout (weight 20%); Task 2: Train sales staff skills (weight 30%); Task 3: Launch promotional activities (weight 50%). Each task includes specific operational steps, responsible persons and timelines, expected performance indicators, risks, and contingency plans.

[0061] Detailed implementation of the approval process Approval notification method: -In-App push notifications- Triggered when new strategies are generated -Mini Program Message- Triggered when a new strategy is generated -System Message Center- Triggered when a new strategy is generated Approval interface functions: The approval interface includes strategy details (strategy name, content description, expected results), execution plan (task list, schedule, responsible person, resource requirements), approval operation area (approve, reject, modify), and feedback reason filling area (preset reason selection + free text input).

[0062] Approval record storage structure: Table: approval_records (Approval Records Table) ├─id: UUID (Primary Key) ├─strategy_id: UUID (Associated Strategy ID) ├─strategy_content: TEXT (Strategy content in JSON format) ├─store_id: UUID (Store ID) ├─reviewer_id: UUID (Approver ID) ├─review_type: ENUM (Approval type: APPROVE / REJECT / MODIFY / PENDING) ├─review_status: ENUM (Current status: PENDING / APPROVED / REJECTED / MODIFIED) ├─feedback_reason: TEXT (Feedback reason, in JSON format) │ ├─selected_reasons: ENUM (Default reason) │ ├─custom_feedback: STRING (Free Text) ├─review_time: TIMESTAMP (Approval Time) ├─created_at: TIMESTAMP (creation time) └─updated_at: TIMESTAMP (update time).

[0063] Automated execution and monitoring include: creating execution tasks and schedules, allocating resources and permissions, monitoring execution progress in real time, and recording execution logs and exceptions. Monitoring metrics include task completion rate, expected progress, cost and resource consumption, and initial performance indicators.

[0064] The detailed process of the recommendation engine includes: querying the first feature vector of the target store; calculating the similarity with the second feature vector of the model stores in the vector database; filtering the preliminary candidate set according to the similarity threshold; calculating the comprehensive score according to the scoring function; selecting the M model stores with the highest comprehensive scores as candidate model stores; obtaining the historical operating strategies of the candidate model stores; calculating the fit score; sorting by fit score; and generating a list of the top N strategy recommendations.

[0065] The AI ​​Agent execution and feedback process includes: receiving the selected historical business strategy; using NLP technology to parse the strategy and obtain the strategy parsing results; matching the task template from the strategy-task mapping rule base; filling in parameters to generate a specific task; generating an execution plan; sending an approval request to the store management terminal; responding to the approval instruction and executing the task by calling the business system via API; monitoring the execution progress in real time; and recording execution logs and results.

[0066] Example of task templates for different strategy types The data update and optimization feedback loop process includes: the effect tracking and evaluation module obtains execution results and operational performance data; the feedback data is updated using a T+1 mechanism; the recommendation engine module reads the feedback data; the effectiveness of different strategy types is statistically analyzed; the weight coefficients in the scoring function are dynamically adjusted; and the optimized parameters take effect in the next recommendation.

[0067] Among them, the formula for calculating the effect tracking and evaluation indicators. Direct effect indicators 1. Strategy Adoption Rate Definition: The percentage of system-generated recommendations that are actually adopted and implemented by stores. Calculation formula: Adoption rate = (Number of adopted recommendations / Total number of recommendations issued) × 100% Example: Total recommendations: 150, 117 accepted, acceptance rate = 117 / 150 × 100% = 78% 2. Execution Completion Rate Definition: The proportion of adopted recommendations that are implemented on time and with high quality. Calculation formula: Completion rate = (Number of successfully completed recommendations / Number of adopted recommendations) × 100% Example: 117 recommendations were adopted, 94 were completed, and the completion rate = 94 / 117 × 100% = 80.34% 3. Target Achievement Rate Definition: The ratio of actual results to expected results Calculation formula: Expected achievement rate = (Actual effect value / Expected effect value) × 100% Achievement rate calculation for each dimension: - Sales growth achievement rate = Actual sales growth % / Expected sales growth % × 100% - Profit improvement achievement rate = Actual profit improvement amount / Expected profit improvement amount × 100% - Overall achievement rate (weighted) = (Single item achievement rate × weight) Business performance indicators 4. Revenue Growth Definition: Change in sales before and after implementing the recommendation Calculation formula: - Sales growth = Sales revenue during the later stage of implementation - Sales revenue during the earlier stage of implementation - Sales growth rate = (Later stage of implementation - Earlier stage of implementation) / Earlier stage of implementation × 100% Example: Sales of ¥50,000 in the first 7 days and ¥54,000 in the last 7 days, an increase of ¥4,000, representing a growth rate of 8%. 5. Profit Improvement Definition: Change in net profit before and after implementing the recommendation Calculation formula: -Profit increase = Post-implementation profit - Pre-implementation profit -Profit Improvement Rate= % Three typical scenarios: -Price increase boosts sales: Sales increased from ¥100k to ¥98k, gross profit margin increased from 25% to 28%, profit increased by ¥2.44k, a 9.76% increase. - Cost reduction and profit improvement: Sales remain unchanged at ¥100k, expenses increase from ¥15k to ¥12k, resulting in a profit increase of ¥3k, a 30% improvement rate. -Comprehensive Plan: Sales increased from ¥100k to ¥110k, expenses increased from ¥15k to ¥13.5k, profit increased by ¥6.2k, a 62% increase. 6. ROI (Return On Investment) Definition: The ratio of the benefits generated by the recommended implementation to the input costs. Calculation formula: ROI = (Revenue - Input) / Input × 100% Example 3: Recommended Operation Strategies for Chain Stores In this embodiment, a chain retail enterprise uses the system of the present invention to provide business strategy recommendations for its stores distributed throughout the country.

[0068] Taking a community store in a second-tier city as an example: Basic store information: Brand "XX Convenience Store", Type "Community Store", Location: Residential area of ​​a second-tier city, Established for 2 years, Area: 80㎡ Historical transaction data: Sales of RMB 8 million in the past 12 months, up 8% year-on-year; gross profit margin of 25%; average daily customer traffic of 300 people; and inventory turnover rate of 6 times per year. Market environment data: 3 competitors in the surrounding area, population density of 5000 people / square kilometer, and consumption level index of 85. The data processing and tagging module extracts features from operational data and generates multi-dimensional store tags for the target store: Size label: Medium-sized store (area 80㎡) Business type tag: community store Regional tag: Second-tier city Growth Tag: Stable Growth (8% growth rate) Efficiency label: Normal Problem tag: Low gross profit (gross profit margin of 25%, lower than the industry average) The execution process is as follows: Step 1: Feature Vector Generation The data processing and tagging module converts store tags into 256-dimensional first feature vectors. Textual features (brand, type) are converted into vectors using an embedding model, while numerical features (area, sales revenue, growth rate, etc.) are standardized using Z-scores and used as vector components.

[0069] Step 2: Similarity Calculation The recommendation engine module calculates the cosine similarity between the first feature vector and the second feature vectors of 100,000 exemplary stores stored in the vector database. The vector database uses Milvus, which supports queries of millions of vectors in seconds.

[0070] The similarity formula is: ; Step 3: Selection of candidate model stores The recommendation engine module employs an efficient retrieval strategy: The first step is to return the top 1000 similar stores based on vector retrieval, with a time complexity of [missing information]. .

[0071] The second step is to filter by a similarity threshold of 0.6 to obtain 500 stores as a preliminary candidate set.

[0072] The third step is to calculate the comprehensive score of each exemplary store in the initial candidate set based on the scoring function. This store belongs to the convenience store industry (fast-moving consumer goods category), and according to the industry's initial weighting rule table, the initial weight is... =0.25, =0.35, =0.2, =0.2: ; in, To ensure similarity to the target store, For stores % of revenue growth rate For stores profit margin For stores Operational efficiency.

[0073] The top 10 stores with the highest overall scores were selected from the initial pool of candidates as exemplary stores.

[0074] Step 4: Generating the Strategy Recommendation List The recommendation engine module retrieved the historical operating strategies of 10 candidate model stores. Analysis revealed that these model stores implemented the following historical operating strategies: "Optimize store layout" strategy "Training Salesperson Skills" Strategy "Launching promotional activities" strategy Membership Marketing Strategy "Inventory optimization strategy" The recommendation engine module calculates the suitability score between each strategy and the target store: ; in, The matching degree between historical operating strategies and target stores is calculated by the cosine similarity between the feature vector of historical operating strategies and the first feature vector. The historical effectiveness of historical business strategies (based on statistics of the implementation effect of historical business strategies in exemplary stores). Assess the feasibility of implementing historical business strategies (based on the target store's size, performance, and problem labels). , , These are the corresponding weighting coefficients, which are 0.4, 0.4, and 0.2, respectively.

[0075] Sort the strategies by suitability score from highest to lowest, and select the top 5 strategies as the strategy recommendation list: "Launching promotional activities" strategy Membership Marketing Strategy "Optimize store layout" strategy "Training Salesperson Skills" Strategy "Inventory optimization strategy" The system outputs the recommendation list to the target store management terminal and displays the recommendation reason: "XX community store, which is 92% similar to your store, saw a 15% increase in sales after adopting the 'launch promotional activities' strategy."

[0076] Step 5: AI Agent Execution Store managers can view the strategy recommendation list in the management terminal, select the "Launch Promotional Activity" strategy, and click confirm.

[0077] The AI ​​Agent execution system receives the selected strategy, uses NLP technology to parse it, and extracts key information from the strategy: Operation type: promotion Target metric: sales Timeframe: 7 days Based on the strategy parsing results, the corresponding task template is matched from the preset strategy-task mapping rule base. The rule base stores three task templates corresponding to the promotional activity strategy: coupon creation task, user filtering task, and message push task.

[0078] The AI ​​Agent execution system populates the target metrics and timeframe into the task template, generating a specific, executable task: Task 1: Create a "¥20 off for every ¥100 spent" coupon, valid for 7 days. Task 2: Filter users within a 3km radius who have not made any purchases in the past 30 days. Task 3: Send promotional information to the selected 2000 users. The AI ​​Agent execution system generates an execution plan, which is then sent to store managers via in-app push notifications for secondary confirmation. Store managers review the strategy details and execution plan on the approval interface, select "Approve," and confirm. The system records the approval information in the `approval_records` table, including approver ID, approval time, and approval status.

[0079] Step 6: Results Tracking and Feedback Optimization The performance tracking and evaluation module monitors the task execution status in real time and records the execution log. On T+1 day, the system obtains the operational performance data before and after execution. Sales revenue in the first 7 days: 100,000 yuan Sales revenue in 7 days after implementation: 120,000 yuan Growth rate: 20% Investment cost: 5000 yuan Gross profit growth: 8,000 yuan ROI: (120000-100000-5000) / 5000 = 3.0 The effect tracking and evaluation module links and stores strategies, execution results, and effect data to form feedback data: Recommended ID: REC_20240114_STORE_001 Execution start date: 2024-01-14 Execution completion date: 2024-01-20 Actual investment cost: ¥5,000 Sales during the execution period (7 days prior to execution): ¥100,000 Sales during the execution period (7 days after execution): ¥120,000 Growth rate: 20% Gross profit growth: ¥8,000 Related exemplary store: STORE_005 Strategy ID: STRATEGY_PROMOTION Similarity score: 0.85 Feedback data is provided to the recommendation engine module. The performance tracking and evaluation module calculates various metrics for this recommendation: adoption rate, execution completion rate, expected achievement rate, ROI, etc. Based on the feedback data, the recommendation engine module calculates that the average effectiveness of "promotional activities" strategies is 80%. Furthermore, if a store manager rejects a recommendation and selects a reason (such as "significant differences in industry / business format among similar stores"), the system will dynamically adjust the weighting scheme to update the store's personalized weights, making future recommendations more aligned with the store's preferences.

[0080] Through continuous improvement cycles, the system achieves more accurate similarity matching, more optimized strategy selection, and continuously improved recommendation accuracy.

[0081] One or more embodiments in this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments in this application should be included within the protection scope of this application.

[0082] If a flowchart is used in this application, it is used to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the steps can be processed in reverse order or simultaneously. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0083] The foregoing has provided a detailed description of an AI-based business strategy recommendation method and system provided in this application. The above description of the disclosed embodiments enables those skilled in the art to implement or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An AI-based business strategy recommendation method, characterized in that, Includes the following steps: Obtain operational data from the target store; Feature extraction is performed on the operational data to generate multi-dimensional store tags for the target store, and the store tags are converted into a first feature vector; Obtain the second feature vector of multiple model stores. The model stores are stores whose operating performance data meets a preset threshold. The second feature vector is generated by converting the store labels of the model stores. Each model store has its historical operating strategies and operating performance data stored in association. Calculate the similarity between the first feature vector and each of the second feature vectors; Based on the similarity, at least one candidate model store with a similarity greater than a first threshold to the target store is selected from a plurality of model stores. Based on the operational performance data of each candidate model store, the historical operational strategies of each candidate model store are sorted, and a strategy recommendation list for the target store is determined based on the sorting results and output to the target store. In response to the target store's confirmation instruction for the selected historical business strategy in the strategy recommendation list, the selected historical business strategy is input into the AI ​​Agent execution system, which then converts the selected historical business strategy into an executable task and executes it automatically. Track and record the execution results of the executable tasks, obtain and associate the operational performance data before and after the execution of the historical business strategies, and form feedback data, which includes strategy adoption rate, execution completion rate, expected achievement rate and return on investment. The recommendation parameters are optimized based on the feedback data, and the recommendation parameters include the weight coefficients of the scoring function used to screen candidate model stores.

2. The AI-based business strategy recommendation method according to claim 1, characterized in that, The operational data includes basic store information, historical transaction data, and market environment data; The basic information about the stores includes brand, type, geographical location, establishment time, and area; The historical transaction data includes sales revenue, gross profit margin, customer traffic, inventory turnover rate, and sales revenue growth rate. The market environment data includes the number of competitors, regional population density, and consumption level index.

3. The AI-based business strategy recommendation method according to claim 2, characterized in that, The step of extracting features from the operational data to generate multi-dimensional store tags for the target store includes the following steps: Static features are extracted from the basic store information, and the static features are mapped to corresponding static tags according to a preset tag rule library; Dynamic features are extracted from the historical transaction data and the market environment data, and the dynamic features are mapped to corresponding dynamic tags according to the tag rule base; The static tags include business type tags, scale tags, and regional tags, while the dynamic tags include growth tags, efficiency tags, performance tags, and problem tags; the tag rule base stores the correspondence between feature values ​​and tag values.

4. The AI-based business strategy recommendation method according to claim 1, characterized in that, The similarity between the first feature vector and each of the second feature vectors is calculated using the cosine similarity algorithm, and the formula is as follows: ； Where A is the first feature vector and B is the second feature vector; The range of values ​​for the similarity is: .

5. The AI-based business strategy recommendation method according to claim 4, characterized in that, The step of selecting at least one candidate model store from the plurality of model stores whose similarity to the target store is greater than a first threshold based on the similarity includes the following steps: From the multiple exemplary stores, select exemplary stores whose similarity is greater than the first threshold as a preliminary candidate set; The comprehensive score of each model store in the preliminary candidate set is calculated according to the scoring function. The input of the scoring function includes the similarity score and the operating performance score. The operating performance score is generated based on the operating performance data of the model stores. The M exemplary stores with the highest comprehensive scores are selected from the initial candidate set and used as the candidate exemplary stores. Wherein, the first threshold and M are both preset values.

6. The AI-based business strategy recommendation method according to claim 5, characterized in that, The operational performance data includes revenue growth rate, profit margin, and operational efficiency; The formula for the scoring function is: ； in, For the first The similarity between the model store and the target store. For the first Revenue growth rate of exemplary stores For the first The profit margin of a model store For the first The operational efficiency of exemplary stores , , , These are the corresponding weighting coefficients; The weighting coefficient , , , The initial settings are differentiated based on the industry of each store, and a set of personalized weights based on historical feedback is maintained for each store.

7. The AI-based business strategy recommendation method according to claim 1, characterized in that, The step of sorting the historical operating strategies of each candidate model store based on their operating performance data, and determining the strategy recommendation list for the target store based on the sorting results, includes the following steps: Obtain the historical operating strategies and corresponding strategy feature vectors of each of the candidate model stores; Calculate the cosine similarity between each of the strategy feature vectors and the first feature vector of the target store, and use it as the matching degree between the historical operating strategy and the target store; Based on the historical performance data of each of the historical business strategies in the candidate model stores and other stores, the average success rate or average return on investment of each of the historical business strategies is calculated as the historical effectiveness of the historical business strategies. Based on the multi-dimensional store tags of the target store, assess the degree of matching between the resources required for the target store to implement each of the historical operating strategies and the available resources of the target store, as a measure of the feasibility of implementing the historical operating strategies; Based on the matching degree, the historical validity, and the implementation feasibility, the suitability score of each of the historical business strategies is calculated according to a preset weighting coefficient. The historical business strategies are sorted in descending order of their suitability scores. The top N ranked historical business strategies are determined as the strategy recommendation list for the target store; Where N is a preset positive integer.

8. The AI-based business strategy recommendation method according to claim 7, characterized in that, The formula for calculating the fit score is: ； in, The degree of matching between the historical operating strategy and the target store. To determine the historical effectiveness of the aforementioned historical business strategies, To assess the feasibility of implementing the aforementioned historical business strategy, , , These are the corresponding weighting coefficients; The weighting coefficient , , Periodic optimization is performed based on global performance data. Optimization rules include adjusting the correlation coefficient between similarity matching and acceptance rate. Adjusted based on the correlation coefficient between historical effectiveness scores and actual return on investment. Adjustments based on the correlation coefficient between feasibility score and execution completion rate .

9. The AI-based business strategy recommendation method according to claim 1, characterized in that, The step of converting the selected historical business strategy into an executable task and automatically executing it through the AIAgent execution system includes the following steps: Natural language processing technology is used to analyze the selected historical business strategies to obtain strategy analysis results, which include operation type, target indicators and time frame; Based on the strategy parsing results, the corresponding task template is matched from the preset strategy-task mapping rule base; Fill the task template with the target metrics and the timeframe to generate a specific executable task; An execution plan is generated based on the specific executable task, and after passing the store approval process, the execution plan is executed by calling the business system through the API interface. The store approval process includes: sending an approval request to the store management terminal, receiving approval operation instructions from the store manager, and recording the approval status and feedback reasons in the approval record table.

10. An AI-based business strategy recommendation system, characterized in that, include: The data processing and tagging module is used to acquire the operating data of the target store, extract features from the operating data, generate multi-dimensional store tags for the target store, and convert the store tags into a first feature vector. The vector database module is used to store the second feature vectors of multiple model stores, as well as the historical operating strategies and operating performance data associated with each model store. The model stores are those whose operating performance data meets a preset threshold. The second feature vectors are generated by converting the store labels of the model stores. The recommendation engine module is used to calculate the similarity between the first feature vector and each of the second feature vectors, select at least one candidate model store from the multiple model stores whose similarity to the target store is greater than a first threshold based on the similarity, and sort the historical operating strategies of each candidate model store based on the operating performance data of each candidate model store, and determine the strategy recommendation list of the target store based on the sorting result; the recommendation engine module is also used to optimize the recommendation parameters based on feedback data, the recommendation parameters including the scoring function weight coefficient; The AI ​​Agent execution system is used to parse the selected historical business strategies in the strategy recommendation list. It uses natural language processing technology to parse the selected historical business strategies to obtain strategy parsing results. According to the preset strategy-task mapping rule base, the strategy parsing results are transformed into one or more specific executable tasks, generating an execution plan. After passing the store approval process, the execution plan is automatically called by the business system through the API interface. The effect tracking and evaluation module is used to track and record the execution results of the execution plan, obtain and associate the historical business effect data before and after the execution of the business strategy, form feedback data, which includes strategy adoption rate, execution completion rate, expected achievement rate and return on investment, and provide the feedback data to the recommendation engine module. The output module is used to output the strategy recommendation list to the target store.

Images