Sales opportunity scoring and customer loss early warning method and device, electronic equipment and storage medium

Abstract

The invention provides a sales opportunity scoring and customer loss early warning method and device, electronic equipment and a storage medium, in the method, a unified customer portrait and event sequence are constructed through multi-source data fusion, a high-quality data footstone is laid for accurate analysis, and the problem of analysis one-sidedness caused by information islands is solved from the source. And furthermore, quantitative scoring of sales opportunities and time window prediction of loss risks are realized by utilizing a machine learning model, management decisions are converted from subjective fuzzy judgment depending on personal experience to objective and accurate insight based on data driving, and the accuracy of sales clue screening and the timeliness of risk early warning are directly improved. On the basis, a causal inference model is introduced to generate targeted action suggestions, so that limited sales resources are ensured to be preferentially and effectively put into customer groups which most actively respond to intervention measures, and invalid waste of marketing resources and meaningless disturbance to customers are fundamentally avoided.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent software, and in particular to a method, apparatus, electronic device, and storage medium for sales opportunity scoring and customer churn early warning. Background Technology

[0002] In the modern business environment, customer acquisition costs (CAC) are rising year by year, making lead-to-cash conversion efficiency a core competitive advantage. Sales teams face a massive influx of leads, often relying on gut feeling to select customers for follow-up, resulting in high-intent customers being neglected (missed deals) and low-intent customers wasting significant time (inefficiency). Furthermore, in the subscription-based (SaaS) economy, customer churn is a fatal blow. Traditional CRM systems merely function as "electronic address books" and "ledgers," lacking intelligent analytical capabilities and unable to answer the three fundamental questions: "Who is most likely to buy?", "Who is about to leave?", and "What should we do now?"

[0003] Traditional sales management techniques mainly include: Static rule-based scoring: Set simple scoring rules (e.g., "fill out a form +10 points"). Problem: The rules are rigid, unable to cover complex scenarios, and the scores do not reflect the true conversion rate.

[0004] Traditional machine learning prediction uses logistic regression or random forests for binary classification. Problems: It ignores the time-series nature of behavior, cannot handle unstructured data, and feature engineering is extremely time-consuming.

[0005] Simple churn warning: based on thresholds (e.g., "no login for 30 days"). Problems: highly delayed, customers may have already churned by the time warning is detected, and the false alarm rate is high.

[0006] General recommendation system: Directly apply e-commerce recommendation algorithms. Problem: B2B sales decision chains are long, cycles are lengthy, and data is sparse, making it difficult to directly apply e-commerce algorithms.

[0007] In summary, traditional sales management methods suffer from technical problems such as low efficiency, high customer churn rate, and poor customer experience. Summary of the Invention

[0008] In view of this, the purpose of the present invention is to provide a sales opportunity scoring and customer churn early warning method, device, electronic device and storage medium to alleviate the technical problems of low efficiency, high customer churn rate and poor customer experience of traditional sales management methods.

[0009] In a first aspect, the present invention provides a method for sales opportunity scoring and customer churn early warning, comprising: Collect customer interaction data from multiple channels, and perform fusion and organization processing to generate a unified customer profile and a time-series of events; Based on the event sequence, a first machine learning model is used to predict the conversion probability and value of sales opportunities, and based on the customer profile and the event sequence, a second machine learning model is used to predict the customer churn risk and time. Based on the conversion probability and value of the sales opportunity, and the customer churn risk and time, a causal inference model is used to generate targeted customer follow-up action suggestions. Based on feedback on the effectiveness of the action recommendations, the first machine learning model, the second machine learning model, and the causal inference model are optimized and updated.

[0010] Furthermore, customer interaction data from multiple channels is collected and processed through fusion and organization, including: Obtain customer interaction data from official websites, communication tools, and business systems; Customer mobile phone number, email address and company name are used as key identifiers, and a fuzzy matching algorithm is used to associate the customer interaction data with the identity. For customer interaction data that cannot be directly associated, a graph neural network-based entity linking technology is used to perform cross-channel identity recognition and association on the customer interaction data, thereby constructing a customer profile with the customer ID as the core. The customer interaction data contains records of interactions between customers and sales staff, which are then converted into standardized behavioral events. Each behavioral event is timestamped, and the behavioral events are sorted along a timeline based on the timestamps to obtain a time-series of events; Semantic features are extracted from call recordings and email texts in the customer interaction data to obtain an event sequence with the semantic features, wherein the semantic features include: sentiment tendency, customer intent, and objection points.

[0011] Furthermore, the first machine learning model includes: a behavioral sequence prediction model based on the Transformer architecture, which, based on the event sequence, uses the first machine learning model to predict the conversion probability and value of sales opportunities, including: The behavioral sequence prediction model is used to predict the behavior of the event sequence to obtain the conversion probability and value of the sales opportunity, wherein the value is the expected transaction amount.

[0012] Furthermore, the second machine learning model includes: a deep survival analysis model, which, based on the customer profile and the event sequence, uses the second machine learning model to predict the risk and timing of customer churn, including: The deep survival analysis model is used to predict the churn risk of the customer profile and the event sequence, thereby obtaining the churn risk and time of the customer.

[0013] Furthermore, based on the conversion probability and value of the sales opportunity, and the customer churn risk and time, a causal inference model is used to generate targeted customer follow-up action suggestions, including: Based on historical intervention data, the causal inference model is used to segment customers to identify customer groups that respond positively to intervention measures. For the aforementioned customer groups that respond positively, personalized customer follow-up action suggestions are generated based on the conversion probability and value of their corresponding sales opportunities, as well as the customer churn risk and time.

[0014] Furthermore, based on feedback on the effectiveness of the action recommendations, the first machine learning model, the second machine learning model, and the causal inference model are optimized and updated, including: Collect customer feedback data regarding the action recommendations, and use the customer feedback data to retrain the first machine learning model, the second machine learning model, and the causal inference model.

[0015] Furthermore, the method also includes: Attribution analysis is performed on the final business results, and the first machine learning model, the second machine learning model, and the causal inference model are optimized, updated, and / or optimization suggestions are generated based on the attribution analysis results. The attribution analysis is implemented using SHAP values.

[0016] Secondly, the present invention also provides a sales opportunity scoring and customer churn early warning device, comprising: The data acquisition and processing unit is used to collect customer interaction data from multiple channels, and to perform fusion and organization processing to generate a unified customer profile and a time-series event sequence. The prediction unit is used to predict the conversion probability and value of sales opportunities based on the event sequence using a first machine learning model, and to predict the customer churn risk and time based on the customer profile and the event sequence using a second machine learning model. The generation unit is used to generate targeted customer follow-up action suggestions based on the conversion probability and value of the sales opportunity, the customer churn risk and time, and using a causal inference model. The optimization and update unit is used to optimize and update the first machine learning model, the second machine learning model, and the causal inference model based on feedback on the implementation effect of the action suggestions.

[0017] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the method described in the first aspect.

[0018] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the method described in the first aspect.

[0019] This invention provides a sales opportunity scoring and customer churn early warning method, comprising: collecting customer interaction data from multiple channels, performing fusion and organization processing to generate a unified customer profile and a time-seriesd event sequence; based on the event sequence, using a first machine learning model to predict the conversion probability and value of sales opportunities, and based on the customer profile and event sequence, using a second machine learning model to predict the risk and time of customer churn; based on the conversion probability and value of sales opportunities, and the risk and time of customer churn, using a causal inference model to generate targeted customer follow-up action suggestions; and based on feedback on the effectiveness of the action suggestions, optimizing and updating the first machine learning model, the second machine learning model, and the causal inference model. As described above, this invention's sales opportunity scoring and customer churn early warning method constructs a unified customer profile and event sequence through multi-source data fusion, laying a high-quality data foundation for accurate analysis and fundamentally solving the problem of analytical bias caused by information silos. Furthermore, by using machine learning models to achieve quantitative scoring of sales opportunities and time window prediction of churn risk, management decisions are transformed from "subjective and vague judgments" relying on personal experience to "objective and accurate insights" based on data-driven approaches, directly improving the accuracy of sales lead screening and the timeliness of risk warnings. Building upon this foundation, a causal inference model is introduced to generate targeted action recommendations, ensuring that limited sales resources are prioritized and effectively invested in the customer groups most responsive to intervention measures. This fundamentally avoids the ineffective waste of marketing resources and unnecessary disturbance to customers. Ultimately, by constructing a closed loop of "execution-feedback-optimization," the system acquires the ability to continuously learn and evolve from actual business results. Therefore, this invention significantly improves the work efficiency and output of sales personnel, effectively reduces customer churn rate through early and accurate warnings and targeted interventions, and fundamentally improves customer experience due to relevant recommended content and appropriate timing of outreach. It alleviates the technical problems of low efficiency, high customer churn rate, and poor customer experience associated with traditional sales management methods. Attached Figure Description

[0020] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 A flowchart of a parking path optimization method provided in an embodiment of the present invention; Figure 2 A schematic diagram of a sales opportunity scoring and customer churn early warning device provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0022] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0023] Traditional sales management methods result in low efficiency, high customer churn, and poor customer experience.

[0024] Based on this, the sales opportunity scoring and customer churn early warning method of this invention constructs a unified customer profile and event sequence through multi-source data fusion, laying a high-quality data foundation for accurate analysis and fundamentally solving the problem of analytical bias caused by information silos. Furthermore, it utilizes machine learning models to achieve quantitative scoring of sales opportunities and time window prediction of churn risk, transforming management decisions from "subjective and vague judgments" relying on personal experience to "objective and accurate insights" based on data-driven approaches, directly improving the accuracy of sales lead screening and the timeliness of risk warnings. Building on this, a causal inference model is introduced to generate targeted action suggestions, ensuring that limited sales resources are prioritized and effectively invested in customer groups most responsive to intervention measures, fundamentally avoiding ineffective waste of marketing resources and unnecessary disturbance to customers. Finally, by constructing a closed loop of "execution-feedback-optimization," the system possesses the ability to continuously learn and self-evolve from actual business results. Therefore, this invention significantly improves the work efficiency and output of sales personnel, effectively reduces customer churn rate through early accurate warnings and targeted interventions, and fundamentally improves customer experience due to relevant recommended content and appropriate timing of outreach.

[0025] To facilitate understanding of this embodiment, a sales opportunity scoring and customer churn early warning method disclosed in this embodiment of the invention will first be described in detail.

[0026] Example 1: According to an embodiment of the present invention, an embodiment of a sales opportunity scoring and customer churn early warning method is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0027] This invention proposes an intelligent CRM empowerment system for B2B / B2C hybrid sales scenarios, aiming to solve problems such as the traditional CRM system's "heavy on recording and light on analysis," reliance on subjective experience for sales lead scoring, belated recognition of customer churn, and lack of scientific guidance for sales actions. This system integrates Sequential Recommendation, Natural Language Processing (NLP), Survival Analysis, and causal inference technologies to construct a sales closed loop of "omnichannel behavior collection - temporal feature modeling - dual-track prediction - intelligent decision-making - attribution optimization."

[0028] Figure 1 This is a flowchart of a parking route optimization method according to an embodiment of the present invention, such as... Figure 1 As shown, the method includes the following steps: Step S102: Collect customer interaction data from multiple channels, and perform fusion and organization processing to generate a unified customer profile and a time-series event sequence; This step involves the collection of omnichannel sales touchpoint data and the construction of a comprehensive customer profile, as well as the construction of sales behavior sequences and semantic feature extraction based on event flows. The system collects customer interaction data from multiple data sources both inside and outside the enterprise. Specifically, these multiple channels include: user behavior data from the company's official website, communication data from communication tools such as WeChat or DingTalk, customer follow-up records from the CRM system, order and transaction data from the ERP system, call recording data from the call center, and email correspondence data from the email system. This data is characterized by its multi-source heterogeneity, including structured data (such as order records and login logs), semi-structured data (such as JSON-formatted event tracking data), and unstructured data (such as call recordings and email text).

[0029] Real-time collection of explicit customer behaviors (filling out forms, downloading white papers) and implicit behaviors (duration spent on the website, number of emails opened). Utilizing ID mapping technology, fragmented data is aggregated into a unique customer ID (OneID), constructing a multimodal, comprehensive customer profile that includes static attributes (company size, industry), dynamic behaviors (interaction timeline), and communication content (call recordings, email text).

[0030] Transform customer-sales interaction records into standardized event streams, such as 'T1: Browse the official website → T2: Download materials → T3: Sales call (connected) → T4: Receive quotation'. Utilize NLP and ASR (speech-to-text) technologies to process unstructured communication data and extract key semantic features, such as "customer pain points (price / features)", "emotional tendency (positive / negative)", "objection type", and "decision-maker role".

[0031] Step S104: Based on the event sequence, use the first machine learning model to predict the conversion probability and value of sales opportunities, and based on the customer profile and event sequence, use the second machine learning model to predict the customer churn risk and time. Specifically, the process consists of a lead scoring process based on Transformer and a customer churn warning process based on deep survival analysis.

[0032] A behavioral sequence prediction model based on the Transformer architecture (similar to BERT4Rec, i.e., the first machine learning model) is constructed. The constructed event sequence is used as input, and a self-attention mechanism is used to capture long-term and short-term behavioral dependencies. The model outputs the "Win Rate" (i.e., the conversion probability of the sales opportunity) and "Expected Transaction Amount" (i.e., value) of the current business opportunity in real time, and classifies leads into four levels—PO (High Quality), P1 (Potential), P2 (Nurturing), and P3 (Low Quality)—based on probability to guide the allocation of sales efforts.

[0033] For existing customers, a DeepSurv (i.e., a second machine learning model) is built. Unlike traditional binary classification models that only predict "whether churn will occur," this model predicts the customer's "survival function" and "time-to-event risk." The system can accurately calculate the churn probability of each customer in the next 1 month and 3 months, and identify key nodes in the "churn risk period" (such as 3 months before contract expiration or 3 days after a service ticket complaint).

[0034] Step S106: Based on the conversion probability and value of sales opportunities, and the risk and time of customer churn, generate targeted customer follow-up action suggestions using a causal inference model; Specifically, this process recommends the Next Best Action (NBA). Based on the prediction results and uplift modeling, the system generates personalized action suggestions for sales personnel. For high-scoring opportunities, it recommends "initiating the signing process"; for customers at high risk of churn, it recommends "high-level follow-up" or "offering coupons." The system can also automatically generate standard operating procedures (SOPs), such as automatically sending targeted retention emails or creating CRM follow-up tasks.

[0035] Step S108: Based on feedback on the effectiveness of the action recommendations, optimize and update the first machine learning model, the second machine learning model, and the causal inference model.

[0036] Specifically, this process involves sales funnel attribution analysis and adaptive strategy optimization. The system performs attribution analysis on the results of winning or losing sales opportunities, and customer renewal or churn. It uses SHAP (Shapley Additive Explanations) values ​​to explain the model's decision-making basis (e.g., "The main reason for the score decline is the lack of follow-up for 7 days"). Based on the attribution results, the system automatically adjusts the weight parameters of the models (first machine learning model, second machine learning model, and causal inference model) and optimizes the NBA recommendation strategy, achieving continuous evolution of the sales strategy.

[0037] The above solution, through the design of "time series modeling, dual-track prediction, and causal decision-making", has achieved a transformation from "experience-driven sales" to "data-intelligent sales", significantly improving the win rate and customer retention rate.

[0038] This invention provides a sales opportunity scoring and customer churn early warning method, comprising: collecting customer interaction data from multiple channels, performing fusion and organization processing to generate a unified customer profile and a time-seriesd event sequence; based on the event sequence, using a first machine learning model to predict the conversion probability and value of sales opportunities, and based on the customer profile and event sequence, using a second machine learning model to predict the risk and time of customer churn; based on the conversion probability and value of sales opportunities, and the risk and time of customer churn, using a causal inference model to generate targeted customer follow-up action suggestions; and based on feedback on the effectiveness of the action suggestions, optimizing and updating the first machine learning model, the second machine learning model, and the causal inference model. As described above, this invention's sales opportunity scoring and customer churn early warning method constructs a unified customer profile and event sequence through multi-source data fusion, laying a high-quality data foundation for accurate analysis and fundamentally solving the problem of analytical bias caused by information silos. Furthermore, by using machine learning models to achieve quantitative scoring of sales opportunities and time window prediction of churn risk, management decisions are transformed from "subjective and vague judgments" relying on personal experience to "objective and accurate insights" based on data-driven approaches, directly improving the accuracy of sales lead screening and the timeliness of risk warnings. Building upon this foundation, a causal inference model is introduced to generate targeted action recommendations, ensuring that limited sales resources are prioritized and effectively invested in the customer groups most responsive to intervention measures. This fundamentally avoids the ineffective waste of marketing resources and unnecessary disturbance to customers. Ultimately, by constructing a closed loop of "execution-feedback-optimization," the system acquires the ability to continuously learn and evolve from actual business results. Therefore, this invention significantly improves the work efficiency and output of sales personnel, effectively reduces customer churn rate through early and accurate warnings and targeted interventions, and fundamentally improves customer experience due to relevant recommended content and appropriate timing of outreach. It alleviates the technical problems of low efficiency, high customer churn rate, and poor customer experience associated with traditional sales management methods.

[0039] The above provides a brief overview of the sales opportunity scoring and customer churn early warning method of the present invention. The specific details involved are described in detail below.

[0040] In an optional embodiment of the present invention, customer interaction data from multiple channels is collected and then fused and organized, specifically including the following steps: (1) Obtain customer interaction data from official websites, communication tools, and business systems; Specifically, sales don't just happen within CRM; they're distributed across countless touchpoints. This system constructs a comprehensive data collection network (Data Harvester).

[0041] Multi-source data access: Online behavior: Collect customer browsing history across digital channels using a tracking SDK (Web / App / Mini Program). This includes not only recording page views (PV) and unique visitors (UV), but also micro-interactions: scrolling depth (reading interest), mouse hover area (focus), and video playback duration.

[0042] Communication data: Connect to call centers to obtain call recordings; connect to corporate email to obtain incoming and outgoing emails; connect to WeChat / DingTalk to obtain IM chat history; connect to Zoom / Tencent Meeting to obtain meeting minutes.

[0043] Business data: retrieve historical orders and payment records from the ERP system; retrieve work orders and complaint records from the customer service system; and retrieve external bidding information from bidding websites.

[0044] (2) Customer mobile phone number, email address and company name are used as key identifiers, and fuzzy matching algorithm is used to associate customer interaction data with identity; Specifically, fuzzy matching algorithms can employ Levenshtein distance algorithm, Jaro-Winkler similarity algorithm, etc., to handle inconsistent data situations, such as "Zhang San" and "Zhang San", "zhangsan@example.com" and "zhangsan@example.com.cn", etc.

[0045] (3) For customer interaction data that cannot be directly associated, the entity linking technology based on graph neural network is used to perform cross-channel identity recognition and association of customer interaction data, thereby constructing a customer profile with customer ID as the core. Specifically, graph neural networks can learn the representation vectors of entities from different data sources and link entities by calculating vector similarity. Customer profiles include basic customer attributes (such as company size, industry, and region), customer behavioral characteristics (such as activity level, interaction frequency, and product preferences), usage depth features, service characteristics, business characteristics, time-series characteristics, customer value tags (such as historical transaction amount and potential value score), and customer relationship status (such as follow-up stage, intent level, and churn risk level).

[0046] Different channels use different identifiers (cookies, phone numbers, email addresses, WeChat IDs). The system uses graph algorithms (Connected Components) to connect these IDs. For example, if Cookie A has left phone number B on the official website, and phone number B matches customer C in the CRM, then Cookie A's activity will be attributed to customer C.

[0047] Panoramic portrait modeling: Construct a portrait library with a wide table structure.

[0048] Static tags: industry, size, registered capital, technology stack (e.g., the client is using competitors' products), decision-making chain structure (inferred from Qichacha data).

[0049] Dynamic metrics: Activity level over the past 7 days, average response time, cumulative communication duration, and number of days since the last interaction (Recency).

[0050] Intended tags: Product preference (preferring option A or option B), budget range, procurement stage (research / project initiation / price comparison / decision).

[0051] After being cleaned (ETL), this data is stored in HBase or ClickHouse, supporting millisecond-level queries and providing a Feature Store for subsequent models.

[0052] (4) Transform customer-sales interaction records in customer interaction data into standardized behavioral events; Specifically, standardized behavioral event types are defined, including browsing, downloading, inquiring, quoting, and signing contracts. Each event is tagged with a timestamp, event type, event object, and event source. For events with missing timestamps, a time interpolation algorithm is used to complete the timestamp.

[0053] (5) Mark each behavioral event with a timestamp, and sort the behavioral events according to the timeline based on the timestamps to obtain a time-series of events; Specifically, the raw data is fragmented and must be transformed into a "story" with temporal logic.

[0054] Construction of time-series event streams: The system defines a standardized event schema: '{Who, When, What, Channel, Duration, Content, Context}'. All interactions are sorted chronologically to form a customer journey.

[0055] For example: '2023-10-01 10:00' [Official Website] Browse the "Enterprise Edition Pricing Page" (stay for 120 seconds) '2023-10-01 10:05 'Download "White Paper.pdf" from the official website' '2023-10-02 14:00' [Phone] Answered a sales follow-up call (300-second call, positive tone) '2023-10-03 09:00 [Email] Open "Product Quotation" (opened 3 times, forwarded 1 time) (6) Extract semantic features from call recordings and email texts in customer interaction data to obtain event sequences with semantic features, including: sentiment, customer intent and objection points.

[0056] Specifically, for calls and emails, ASR (speech-to-text transcription) and NLP models are used for in-depth analysis.

[0057] Key element extraction (Slot Filling): Extract "Budget Amount", "Estimated Launch Time (BANT Model)" and "Competitor Name".

[0058] Objection detection: Identify customer objections, such as "It's too expensive," "The function is not supported," or "I need to report to my supervisor."

[0059] Sentiment analysis: Calculate the sentiment polarity (Positive / Negative) of each sentence. If the customer frequently interrupts the sales pitch or speaks rapidly during the call, it may indicate impatience.

[0060] Topic modeling: Identifying the distribution of communication topics. Is the focus on "technical details" or "business terms"? The former represents the technical approval stage, while the latter represents the business negotiation stage.

[0061] These semantic features are encoded as high-dimensional vectors and embedded into the event stream, enriching the meaning of the behavioral sequence.

[0062] Through distributed data acquisition and streaming processing technologies, real-time fusion and standardization of multi-source heterogeneous data were achieved, improving the efficiency and timeliness of data processing. Through graph neural network entity linking technology, the problem of cross-channel identity recognition was solved, and a more complete and accurate customer profile was constructed. Through semantic feature extraction technology, deep value information was mined from unstructured data, enriching the semantic dimension of event sequences.

[0063] In an optional embodiment of the present invention, the first machine learning model includes: a behavioral sequence prediction model based on the Transformer architecture, which predicts the conversion probability and value of sales opportunities based on event sequences, specifically including the following steps: A behavioral sequence prediction model is used to predict the behavior of event sequences, thereby obtaining the conversion probability and value of sales opportunities, where the value is the expected transaction amount.

[0064] Specifically, traditional ratings are rule-based (e.g., "see price page +5 points"), which cannot capture complex temporal patterns (e.g., "see price first, then see case studies" indicates a stronger intention than "see case studies first, then see price").

[0065] Model architecture: A sequence prediction model based on Transformer Encoder (similar to BERT) is adopted.

[0066] Input layer: It is obtained by adding three parts of embedding: event embedding (browsing / call / email) + location embedding (sequence order) + time interval embedding (time interval, capturing interaction density).

[0067] Encoding layer: Multi-head self-attention mechanism. The model automatically learns the dependency weights between different events. For example, the model might learn that if "downloading a white paper" is immediately followed by "high-level call," the probability of winning the deal increases dramatically; if it is followed by "long silence," the probability drops sharply.

[0068] Multi-task output layer: Binary Classification: Predicts whether a lead will ultimately result in a win or loss.

[0069] Regression: Predicts the final transaction amount (Deal Size).

[0070] Next Stage Forecast: Forecast the next sales stage (e.g., moving from "Validation" to "Negotiation").

[0071] Dynamic scoring mechanism: The model performs reasoning on all follow-up business opportunities every night (or in real time) and outputs a score from 0 to 100.

[0072] Score Explanation: 90 points = extremely high interest, it is recommended to close the deal immediately; 30 points = lukewarm interest, it is recommended to put the product in the open market or use automated nurturing.

[0073] Trend monitoring: Monitors week-on-week changes in scores. If a major client's score drops sharply from 80 to 50 (possibly due to competitor intervention or key personnel departure), the system immediately triggers a "score drop warning".

[0074] In an optional embodiment of the present invention, the second machine learning model includes: a deep survival analysis model, which, based on customer profiles and event sequences, uses the second machine learning model to predict customer churn risk and timing, specifically including the following steps: A deep survival analysis model is used to predict customer churn risk based on customer profiles and event sequences, thus obtaining the customer churn risk and time.

[0075] Specifically, for SaaS or subscription-based businesses, retaining existing customers is more important than acquiring new ones. Traditional classification models can only predict "whether or not churn," but cannot answer "when churn will occur."

[0076] Deep Survival Analysis: This system uses either the DeepSurv or Cox-Time model.

[0077] Objective function: Maximize the log-likelihood function, considering both lost samples (Events) and non-lost samples (Censored).

[0078] Covariate inputs (the following features are included in the customer profile and event sequence): Usage depth features: login frequency, core function usage rate, API call volume, and storage usage (reflecting stickiness).

[0079] Service characteristics: number of work orders submitted, average resolution time, and NPS score (reflecting satisfaction).

[0080] Business characteristics: remaining days of the contract, historical renewal discount rate.

[0081] Time-series characteristics: Slope of usage decline over the past 3 months.

[0082] Risk forecasting and life cycle value (CLV) calculation: The model outputs the survival probability curve $S(t)$ for each customer.

[0083] Churn warning: Calculate the probability of a customer churning within a future time period T: $P(T)=1-S(t+T) / S(t)$. If $P(30 days)>0.6$, it is marked as high risk.

[0084] Key risk point identification: The model can identify specific risk patterns, such as "the peak of churn occurs 60 days before the contract expires and the work order satisfaction is low".

[0085] CLV prediction: Based on the survival curve integral, it estimates the remaining lifetime value of customers and helps companies determine the upper limit of retention costs (it is not worth spending 10,000 yuan to retain a customer worth 5,000 yuan).

[0086] In an optional embodiment of the present invention, based on the conversion probability and value of sales opportunities, and the risk and time of customer churn, a causal inference model is used to generate targeted customer follow-up action suggestions, specifically including the following steps: (1) Based on historical intervention data, customers are segmented using a causal inference model to identify customer groups that respond positively to intervention measures; (2) For customer groups that respond positively, personalized customer follow-up action suggestions are generated based on the conversion probability and value of their corresponding sales opportunities, the risk of customer churn and time.

[0087] Specifically, the purpose of forecasting is intervention. The system not only tells sales "who is important," but also "how to do it."

[0088] Strategies based on causal inference: Four customer categories were identified using Uplift Modeling: 1. Persuadables: No intervention leads to churn, intervention leads to retention → Focus on intervention targets 2. Sure Things (Solidarity): Retain without interference → Do not disturb.

[0089] 3. Lost Causes: Whether intervention is provided or not, the patient will lose their chance of recovery → give up on treatment.

[0090] 4. Sleeping Dogs (Averse Type): Intervention will only accelerate their loss → Do not touch them.

[0091] The system trains the Uplift model using historical A / B test data and recommends retention actions only for "persuasive" customers.

[0092] SOP automated execution: Combining rule engines with generative AI: Business opportunity advancement: If the score is high but the solution is stuck in the "solution verification" stage, the system recommends "sending a competitor comparison analysis report" and automatically generates an email draft.

[0093] Churn Retention: If churn is predicted to be due to pricing issues, we recommend "applying for an 80% discount renewal coupon"; if it's due to service issues, we recommend "a customer success manager making an in-person visit".

[0094] Automated outreach: For low-scoring opportunities, send a series of Nurture emails via Marketing Automation (MA) tools until the customer takes a new action that triggers a score increase.

[0095] In an optional embodiment of the present invention, the first machine learning model, the second machine learning model, and the causal inference model are optimized and updated based on feedback on the effectiveness of the action suggestions, specifically including the following steps: Collect customer feedback data on action recommendations, and use the customer feedback data to retrain the first machine learning model, the second machine learning model, and the causal inference model.

[0096] In an optional embodiment of the present invention, the method further includes the following steps: Attribution analysis is performed on the final business results, and the first machine learning model, the second machine learning model, and the causal inference model are optimized, updated, and / or optimization suggestions are generated based on the attribution analysis results. The attribution analysis is implemented using SHAP values.

[0097] Specifically, SHAP attribution explanation: In order to convince the sales AI of its ratings, the system uses SHAP (Shapley Additive Explanations) values ​​to explain each prediction result.

[0098] For example: 'Client A's score is 92. Positive contributing factors: [Yesterday's high-level call +15 points], [Downloaded white paper +5 points]; Negative contributing factors: [Budget not confirmed -3 points].' This white-box explanation helps sales understand the AI ​​logic and address specific weaknesses (such as confirming the budget).

[0099] Winning / Losing Bet Analysis Report: The system automatically analyzes the reasons for lost orders (Closed-Lost Analysis). Aggregate analysis reveals that "the main sources of lost orders this quarter are 'Competitor X', and the primary reason is 'price'", thus recommending that management adjust pricing strategies or wording.

[0100] Model adaptive evolution: The system establishes a closed loop of "prediction-execution-feedback".

[0101] Whenever a salesperson takes an action based on NBA recommendations (such as sending an email), the system tracks the subsequent effects (Did the customer respond? Did the rating increase?). If a certain type of recommendation remains ineffective for an extended period, the reinforcement learning agent will reduce the recommendation weight of that action and explore new strategies. Simultaneously, as the market environment changes (such as the launch of a new product), the model will automatically fine-tune using new data to maintain the timeliness of the scoring rules.

[0102] This invention aims to address the following core pain points: 1. Challenges in capturing temporal features: Capturing the sequential dependencies of behaviors during long-term sales processes.

[0103] 2. Challenges in utilizing multimodal data: Effectively utilizing massive amounts of communication voice and text data to assist in judgment.

[0104] 3. The challenge of timeliness in churn prediction: Shift from "hindsight" to "early warning" to accurately predict churn time windows.

[0105] 4. The challenge of the effectiveness of strategic intervention: Distinguish between natural conversion and intervention-driven conversion, scientifically guide sales actions, and avoid ineffective harassment.

[0106] The core technological innovations of this invention are: Innovation Point 1: Sales Time-Sequence Behavior Modeling Based on Transformer Breaking away from the limitations of traditional CRM scoring that relies solely on static rules or simple statistics (Logistic Regression), this invention creatively applies the Transformer architecture from the NLP field to sales behavior flow data. Each customer interaction is treated as a "word," and the entire lifecycle as a "sentence." By capturing long-term behavioral dependencies through a self-attention mechanism (such as the impact of "business card from a trade show six months ago" on "signing a contract today"), it achieves accurate modeling of non-linear and complex sales processes, improving scoring AUC (accuracy rate) by over 20%.

[0107] Innovation Point 2: The integrated application of deep survival analysis and uplift modeling In customer churn prediction, the system not only addresses the question of "whether churn will occur" (classification problem) but also the question of "when churn will occur" (time prediction) and further addresses the question of "whether intervention is effective" (causal inference). By integrating DeepSurv and Uplift models, the system can accurately identify high-value customers who are "on the verge of churn and can be recovered through intervention," avoiding wasting resources on "inevitable" or "loyal" customers, and improving marketing ROI by 30%.

[0108] Innovation Point 3: Semantic Fusion of Multimodal Unstructured Interactive Data Traditional CRM systems only record structured fields (amount, stage). This system deeply mines unstructured data such as call recordings, email texts, and meeting minutes. A multimodal fusion network is proposed, which concatenates the emotional features of speech (Audio Embedding), the semantic features of text (Text Embedding), and structured behavioral features (Behavior Embedding). It can identify the true intent of a customer who verbally says they will consider it (text neutral), but whose tone is impatient (negative voice).

[0109] Innovation Point 4: Explainability (xal)-driven Next Best Action Recommendation This solves the problem of sales staff distrust caused by the "black box" nature of AI systems. Based on the SHAP value attribution explanation mechanism, it transforms complex neural network weights into business language that sales staff can understand (e.g., "Your score dropped by 10 points because we haven't contacted you in the last 3 days"). Based on this, it generates specific action instructions (NBA), achieving a leap from "prediction" to "decision-making," truly empowering frontline sales.

[0110] Innovation Point 5: Omnichannel OneID and Behavioral Graph Construction Technology To address the complex "many-to-many" relationships in B2B sales (one company with multiple contacts, one contact with multiple devices), a behavioral graph based on a graph database was constructed. This graph not only links the cross-device behavior of the same person but also the interaction behavior of different decision-makers within the same company (CTO, CEO, procurement). The model can identify complex procurement patterns such as bid-rigging or multi-departmental collaborative decision-making, providing an enterprise-level panoramic opportunity scoring.

[0111] The technical effects are as follows: Sales efficiency is greatly improved: Through precise scoring, sales staff can focus on the top 20% of high-intent customers, increasing the lead conversion rate by 30-50%.

[0112] Significantly reduced customer churn rate: Early identification of churn risks and precise intervention can increase the retention rate of high-value customers by 15-20%, directly increasing enterprise revenue.

[0113] More accurate sales forecasts: Pipeline forecasts based on AI scoring are far more accurate than the "expected win rate" filled in by sales, helping management to formulate inventory and recruitment plans more accurately.

[0114] Newcomers get up to speed faster: SOPs and NBA recommendations act like "AI coaches," guiding new sales staff to follow up with clients according to best practices and shortening the training cycle for newcomers.

[0115] Customer experience optimization: Reduced unwanted phone calls to potential customers, provided appropriate service at the right time, and enhanced brand image.

[0116] The key points are as follows: Omnichannel behavioral event stream standardization technology: Cleans and aligns heterogeneous data such as Web / App event tracking, calls, emails, and ERP transactions into a standardized time-series event sequence.

[0117] Transformer-driven opportunity scoring network: It uses a self-attention mechanism to encode long-term and short-term behavioral features, and learns in multiple tasks to simultaneously predict the probability of winning a deal and the transaction amount.

[0118] Multimodal communication content semantic fusion technology: Combining ASR and NLP, it extracts deep semantic features such as emotion, intent, and objection from calls and emails and integrates them into the scoring model.

[0119] Churn time prediction method based on survival analysis: The DeepSurv model is used to predict the customer's survival function and accurately calculate the churn probability and CLV within a specific future time window.

[0120] Causal inference-oriented NBA decision engine: Utilizes Uplift Modeling to differentiate intervention gains, and combines a rules engine to generate specific next best action recommendations and automated SOPs.

[0121] Example 2: This invention also provides a sales opportunity scoring and customer churn early warning device, which is mainly used to execute the sales opportunity scoring and customer churn early warning method provided in Embodiment 1 of this invention. The following is a detailed description of the sales opportunity scoring and customer churn early warning device provided in this invention.

[0122] Figure 2 This is a schematic diagram of a sales opportunity scoring and customer churn early warning device according to an embodiment of the present invention, such as... Figure 2 As shown, the device mainly includes: an acquisition and processing unit 10, a prediction unit 20, a generation unit 30, and an optimization and update unit 40, wherein: The data acquisition and processing unit is used to collect customer interaction data from multiple channels, and to perform fusion and organization processing to generate a unified customer profile and a time-series event sequence. The prediction unit is used to predict the conversion probability and value of sales opportunities based on event sequences using a first machine learning model, and to predict customer churn risk and time based on customer profiles and event sequences using a second machine learning model. The generation unit is used to generate targeted customer follow-up action suggestions based on the conversion probability and value of sales opportunities, and the risk and time of customer churn, using a causal inference model. The optimization and update unit is used to optimize and update the first machine learning model, the second machine learning model, and the causal inference model based on feedback on the effectiveness of the action recommendations.

[0123] This invention provides a sales opportunity scoring and customer churn early warning device, comprising: collecting customer interaction data from multiple channels, performing fusion and organization processing to generate a unified customer profile and a time-seriesd event sequence; based on the event sequence, using a first machine learning model to predict the conversion probability and value of sales opportunities, and based on the customer profile and event sequence, using a second machine learning model to predict the risk and time of customer churn; based on the conversion probability and value of sales opportunities, and the risk and time of customer churn, using a causal inference model to generate targeted customer follow-up action suggestions; and based on feedback on the effectiveness of the action suggestions, optimizing and updating the first machine learning model, the second machine learning model, and the causal inference model. As described above, the sales opportunity scoring and customer churn early warning device of this invention constructs a unified customer profile and event sequence through multi-source data fusion, laying a high-quality data foundation for accurate analysis and fundamentally solving the problem of analytical bias caused by information silos. Furthermore, by using machine learning models to achieve quantitative scoring of sales opportunities and time window prediction of churn risk, management decisions are transformed from "subjective and vague judgments" relying on personal experience to "objective and accurate insights" based on data-driven approaches, directly improving the accuracy of sales lead screening and the timeliness of risk warnings. Building upon this foundation, a causal inference model is introduced to generate targeted action recommendations, ensuring that limited sales resources are prioritized and effectively invested in the customer groups most responsive to intervention measures. This fundamentally avoids the ineffective waste of marketing resources and unnecessary disturbance to customers. Ultimately, by constructing a closed loop of "execution-feedback-optimization," the system acquires the ability to continuously learn and evolve from actual business results. Therefore, this invention significantly improves the work efficiency and output of sales personnel, effectively reduces customer churn rate through early and accurate warnings and targeted interventions, and fundamentally improves customer experience due to relevant recommended content and appropriate timing of outreach. It alleviates the technical problems of low efficiency, high customer churn rate, and poor customer experience associated with traditional sales management methods.

[0124] Optionally, the collection and processing unit is also used to: acquire customer interaction data from official websites, communication tools, and business systems; use customer mobile phone numbers, email addresses, and company names as key identifiers to perform identity association on customer interaction data through fuzzy matching algorithms; for customer interaction data that cannot be directly associated, use entity linking technology based on graph neural networks to perform cross-channel identity recognition and association on customer interaction data, thereby constructing a customer profile with customer ID as the core; convert customer-sales interaction records in customer interaction data into standardized behavioral events; mark each behavioral event with a timestamp and sort the behavioral events according to the timeline based on the timestamp to obtain a time-series event sequence; extract semantic features from call recordings and email text in customer interaction data to obtain an event sequence with semantic features, wherein the semantic features include: sentiment tendency, customer intent, and objection points.

[0125] Optionally, the first machine learning model includes: a behavior sequence prediction model based on the Transformer architecture, and the prediction unit is further used to: use the behavior sequence prediction model to predict the behavior of the event sequence to obtain the conversion probability and value of the sales opportunity, wherein the value is the expected transaction amount.

[0126] Optionally, the second machine learning model includes a deep survival analysis model, and the prediction unit is also used to: use the deep survival analysis model to predict churn risk based on customer profiles and event sequences, thereby obtaining the customer churn risk and time.

[0127] Optionally, the generation unit is also used to: segment customers based on historical intervention data using a causal inference model to identify customer groups that respond positively to intervention measures; and generate personalized customer follow-up action suggestions for customer groups that respond positively based on the conversion probability and value of their corresponding sales opportunities, as well as the risk and time of customer churn.

[0128] Optionally, the optimization and update unit is also used to: collect customer feedback data on action recommendations, and use the customer feedback data to retrain the first machine learning model, the second machine learning model, and the causal inference model.

[0129] Optionally, the device is also used to: perform attribution analysis on the final business results, and optimize and update the first machine learning model, the second machine learning model, and the causal inference model based on the attribution analysis results and / or generate optimization suggestions, wherein the attribution analysis is implemented using SHAP values.

[0130] The device provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0131] like Figure 3 As shown in the embodiment of this application, an electronic device 600 includes a processor 601, a memory 602, and a bus. The memory 602 stores machine-readable instructions executable by the processor 601. When the electronic device is running, the processor 601 communicates with the memory 602 via the bus. The processor 601 executes the machine-readable instructions to perform the steps of the sales opportunity scoring and customer churn warning method described above.

[0132] Specifically, the memory 602 and processor 601 mentioned above can be general-purpose memory and processor, without any specific limitations. When the processor 601 runs the computer program stored in the memory 602, it can execute the sales opportunity scoring and customer churn warning method mentioned above.

[0133] The processor 601 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 601 or by instructions in software form. The processor 601 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 602, and processor 601 reads the information from memory 602 and, in conjunction with its hardware, completes the steps of the above method.

[0134] Corresponding to the above-mentioned sales opportunity scoring and customer churn warning method, this application embodiment also provides a computer-readable storage medium storing machine-executable instructions. When the machine-executable instructions are called and run by a processor, the machine-executable instructions cause the processor to perform the steps of the above-mentioned sales opportunity scoring and customer churn warning method.

[0135] The sales opportunity scoring and customer churn early warning device provided in this application embodiment can be specific hardware on a device or software or firmware installed on the device. The implementation principle and technical effects of the device provided in this application embodiment are the same as those in the foregoing method embodiments. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the foregoing method embodiments. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can all be referred to the corresponding processes in the above method embodiments, and will not be repeated here.

[0136] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0137] For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0138] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0139] In addition, the functional units in the embodiments provided in this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0140] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the sales opportunity scoring and customer churn warning methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0141] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0142] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application. All should be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A method for sales opportunity scoring and customer churn early warning, characterized in that, include: Collect customer interaction data from multiple channels, and perform fusion and organization processing to generate a unified customer profile and a time-series of events; Based on the event sequence, a first machine learning model is used to predict the conversion probability and value of sales opportunities, and based on the customer profile and the event sequence, a second machine learning model is used to predict the customer churn risk and time. Based on the conversion probability and value of the sales opportunity, and the customer churn risk and time, a causal inference model is used to generate targeted customer follow-up action suggestions. Based on feedback on the effectiveness of the action recommendations, the first machine learning model, the second machine learning model, and the causal inference model are optimized and updated.

2. The method according to claim 1, characterized in that, Collect customer interaction data from multiple channels, and perform integrated and organized processing, including: Obtain customer interaction data from official websites, communication tools, and business systems; Customer mobile phone number, email address, and company name are used as key identifiers, and a fuzzy matching algorithm is used to associate the customer interaction data with the customer identity. For customer interaction data that cannot be directly associated, a graph neural network-based entity linking technology is used to perform cross-channel identity recognition and association on the customer interaction data, thereby constructing a customer profile with the customer ID as the core. The customer interaction data contains records of interactions between customers and sales staff, which are then converted into standardized behavioral events. Each behavioral event is timestamped, and the behavioral events are sorted along a timeline based on the timestamps to obtain a time-series of events; Semantic features are extracted from call recordings and email texts in the customer interaction data to obtain an event sequence with the semantic features, wherein the semantic features include: sentiment tendency, customer intent, and objection points.

3. The method according to claim 1, characterized in that, The first machine learning model includes: a behavioral sequence prediction model based on the Transformer architecture, which predicts the conversion probability and value of sales opportunities based on the event sequence, including: The behavioral sequence prediction model is used to predict the behavior of the event sequence to obtain the conversion probability and value of the sales opportunity, wherein the value is the expected transaction amount.

4. The method according to claim 1, characterized in that, The second machine learning model includes: a deep survival analysis model, which, based on the customer profile and the event sequence, uses the second machine learning model to predict the risk and timing of customer churn, including: The deep survival analysis model is used to predict the churn risk of the customer profile and the event sequence, thereby obtaining the churn risk and time of the customer.

5. The method according to claim 1, characterized in that, Based on the conversion probability and value of the sales opportunity, and the customer churn risk and time, a causal inference model is used to generate targeted customer follow-up action suggestions, including: Based on historical intervention data, the causal inference model is used to segment customers to identify customer groups that respond positively to intervention measures. For the aforementioned customer groups that respond positively, personalized customer follow-up action suggestions are generated based on the conversion probability and value of their corresponding sales opportunities, as well as the customer churn risk and time.

6. The method according to claim 1, characterized in that, Based on feedback on the effectiveness of the action recommendations, the first machine learning model, the second machine learning model, and the causal inference model are optimized and updated, including: Collect customer feedback data regarding the action recommendations, and use the customer feedback data to retrain the first machine learning model, the second machine learning model, and the causal inference model.

7. The method according to claim 1, characterized in that, The method further includes: Attribution analysis is performed on the final business results, and the first machine learning model, the second machine learning model, and the causal inference model are optimized, updated, and / or optimization suggestions are generated based on the attribution analysis results. The attribution analysis is implemented using SHAP values.

8. A sales opportunity scoring and customer churn early warning device, characterized in that, include: The data acquisition and processing unit is used to collect customer interaction data from multiple channels, and to perform fusion and organization processing to generate a unified customer profile and a time-series event sequence. The prediction unit is used to predict the conversion probability and value of sales opportunities based on the event sequence using a first machine learning model, and to predict the customer churn risk and time based on the customer profile and the event sequence using a second machine learning model. The generation unit is used to generate targeted customer follow-up action suggestions based on the conversion probability and value of the sales opportunity, the customer churn risk and time, and using a causal inference model. The optimization and update unit is used to optimize and update the first machine learning model, the second machine learning model, and the causal inference model based on feedback on the implementation effect of the action suggestions.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program is executed by the processor to perform the method of any one of claims 1 to 7.

Images