AI-POWERED DYNAMIC AND COMPETITIVE TARIFF AND OFFER PERSONALIZATION SYSTEM

Abstract

This invention relates to a system (1) that enables real-time artificial intelligence optimization of customer-specific tariffs and offers and competitive pricing for telecommunications operators.

Description

1 TARIFF AI-POWERED DYNAMIC AND COMPETITIVE PRICING AND OFFER PERSONALIZATION SYSTEM Technical Area 5 This invention enables telecommunications operators to offer customer-specific tariffs and deals. Competitive pricing optimized in real time with artificial intelligence. It is related to a system that enables this. Previous Technique In the current state of the art, existing systems are generally static or segment-based. It is not competitive and does not provide real-time adjustment to competing operator prices. Pricing is mostly based on manual analysis or limited data. Campaign 15 Analytics that correlate performance with actual customer behavior are limited. When implementing dynamic pricing, the price movements of competitors are systematic and It is not monitored in an integrated manner. Competitor data obtained through web scraping. Because their accuracy or freshness has not been checked, the systems are faulty. It can make incorrect pricing decisions by reacting to data. The suggested price or package, 20 Compliance with the client's existing contract or legal regulations is generally required. It is checked later; this leads to processing errors. Standard reinforced. Since learning models focus solely on reward (income), they are subject to aggressive pricing. This can lead to customer churn. The system constantly changes during instantaneous data exchanges. Offering different prices (flickering) undermines customer trust; in current systems 25 The stability (hysteresis) mechanisms are inadequate. Therefore, considering the studies and shortcomings in the current technique... When considered, the customer's usage patterns and the competitors' prices and By analyzing campaign movements, the most suitable, dynamic and competitive pricing and 30 submitting proposals, testing the accuracy of competitor data, and presenting the results. 2 The technical and legal compliance of the proposal is assessed through "Constraint-Based Validation". to guarantee this, to avoid frustrating the user and to maintain price stability Managing bid traffic on digital channels, determining which campaigns that positive feedback was received from the customer, and which campaigns were not approved. Analysis shows that this inadequacy leads to customer loss or operator switching. 5 this, thanks to these analyses, allows for the selection of campaign portfolios and tariff packages. By continuously optimizing, customer satisfaction and loyalty are maximized. the removal of the campaign by integrating it into the digital channels of telecommunications companies automating management, improving customer satisfaction and revenues It appears that a system is needed that will enable its increase. 10 International patent number WO2025079109A1, which falls under the prior art. The document states that customers' Telekom subscription plans can be optimized with the support of artificial intelligence. to manage and provide customers with personalized offers. A system and a method that enables its presentation are being discussed. (Statement 15) The system in question in the invention is a single integrated module from a data integration module. It includes a receiver unit configured to receive data streams. The system is integrated. Organized data using a preprocessor to normalize the data. a normalization that obtains the set and stores the organized data It includes the unit. The system analyzes organized data and provides customers with 20 an analysis unit for conducting trend analysis to separate and categorize It includes the system to estimate customer churn risk and use the model. It includes a prediction unit to detect anomalies. The system allows customers to... categorizes customers based on their mobile subscription plans and estimates the risk of customer churn. It includes a production unit to create a data visualization. 25 Brief Description of the Invention The purpose of this invention is to improve customer tariff management in the telecommunications sector. Competitive price analysis, campaign performance analysis, and personalized offers 30 including AI-based solutions for their work; on digital channels 3 real-time price optimization, offer presentation, and customer feedback. campaign optimization based on reliability of competitor data automatic verification of regulatory and package compliance restrictions. monitoring, customer churn and revenue balance Safe Reinforcement Optimization through learning (Safe RL) and reduction of bid fatigue to 5 The goal is to implement a system developed to ensure prevention. Another aim of this invention is customer-based personalization and competitor price analysis. combined with artificial intelligence and supported by campaign performance analytics. To ensure automated and highly accurate competitive pricing. 10 The goal is to implement a system developed for this purpose. Detailed Description of the Invention The “AI-Powered Dynamic 15” project was carried out to achieve the purpose of this invention. and the "Competitive Tariff and Offer Personalization System" is shown in the attached document. and in this form; Figure 1 shows a schematic view of the system that is the subject of the invention. The parts shown in the figure are individually numbered, and these numbers correspond to... The corresponding answers are given below. 1. System 2. Data Collection Module 25 3. Competitor Pricing and Campaign Monitoring Module 4. Data Processing Module 5. Prediction Module 6. Offer Validity Check Module 7. Artificial Intelligence Module 30 8. Control Module 4 9. Presentation and Feedback Module A. Customer Interface For telecommunications operators, customer-specific tariffs and offers provide a competitive advantage. pricing is optimized in real time using artificial intelligence. The system in question, developed for the purpose of providing (1); - real-time and historical usage signals of a telecommunications customer to collect and process data using pseudonymization techniques at least one data collection module configured (2), 10 - Implementing pricing and promotions of competing operators through programming. in real time via interface or network scraping methods at least one competitor's price and campaign structured to collect monitoring module (3), - Data completion on collected raw data, outlier 15 to carry out cleaning and normalization processes at least one data processing module configured (4), - the probability of offer acceptance on a customer basis and customer loss after the offer. Estimating the probability, identifying offers with a high probability of acceptance, and the customer 20 proposals that will not increase losses and will provide revenue optimization at least one prediction module configured to select (5), - Regarding the problem of not being able to offer every package to every customer, constraint analyzer before or during proposal generation implementing the technique, package compatibility, switching with the current tariff rules, commitment / contractual restrictions, regulation and fair use 25 rules, campaign budget and stock restrictions, all for the same user Imposing restrictions in the form of transmission frequency limitations is technical. as “invalid” or “operationally impractical” to ensure that bids are eliminated during the production phase at least one configured offer validity check module (6), 30 - using reinforcement learning architecture, making the offer selection decision classical. Instead of rules or static models, with safe reinforcement learning. to optimize, in terms of the situation, the customer profile, the competitor price vector and obtain the network / service terms and conditions, take action by submitting the offer package or At least one 5 configured to select the price / discount parameter artificial intelligence module (7), - not changing offers for small score differences, before the specified time expires mechanisms such as not generating new offers, daily / weekly offer limits By implementing this technical measure, we can both stabilize the user experience. to both keep it stable and prevent the model from behaving erratically 10 at least one control module configured (8), - selected offer, application programming interface with customer at least one presentation configured to transmit to interfaces (A) and return It includes a power supply module (9). The data collection module (2) in the system (1) which is the subject of the invention, Telekom customer data, voice, and text message consumption, roaming behavior, usage details, Internet browsing data, device type, peak usage hours, location information, history bill payment habits, digital channel interactions, previous campaign responses Collecting real-time and historical usage signals in this form, data 20 Processing using pseudonymization techniques; only an offer. retaining the necessary attributes for optimization and data minimization process It is structured to be implemented. The competitor price and campaign monitoring module (3) included in the system (1) which is the subject of the invention, 25 Application programming interface for displaying prices and campaigns of competing operators (Application Programming Interface - API) or web scraping It is structured to collect data in real time using various methods. Competitor Price and campaign monitoring module (3), on competitor data: timeliness score, By calculating the reliability score and latency compensation, 30% of incorrect / outdated competitor information can be corrected. 6 to prevent the weighted competition signal from having a false impact on the price decision. It is structured to be used as a competition vector. The data processing module (4) in the system (1) which is the subject of the invention, the collected raw data Value Assignment with K-Nearest Neighbors (K-Nearest Neighbors Imputation- 5) (KNN Imputation) or Mean / Median Imputation missing data completion, Z-score or Interquartile Range Applying outlier removal and normalization using IQR methods, price sensitivity score, packet overage risk, high peak hour consumption rate, Customer churn early warning signals, campaign fatigue (offer fatigue) 10 to ensure that features with a high impact on the offer are produced in the form of a metric It is structured. The data processing module (4) determines which features are included in the offer selection. To determine if it is critical, Random Forest Attribute Significance Feature Importance) or Shapley Additive Explanations It is configured to use exPlanations (SHAP) methods. 15 The forecast module (5) in the system (1) which is the subject of the invention, offers on a customer basis. Estimating the probability of acceptance and the probability of customer churn after the offer, acceptance High probability bids that won't increase customer churn and revenue optimization. It is configured to select the offers that will provide. The forecast module (5), 20 Gradient Boosting for Forecasting, Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM) in the form of controlled Long Short-Term Memory (Long) for learning models and time series trends Short-Term Memory (LSTM) / Gated Recurrent Unit Using architectures such as GRU (K-Tiered Cross-Route) in the education process Improving generalization capability through validation (K-Fold Cross-Validation), hyper Optimize the parameters using Bayesian optimization, and offer model outputs. generating calibrated probability scores that will serve as input to the decision It is structured to provide this. 7 The offer validity check module (6) in the system (1) which is the subject of the invention, each Regarding the problem of not being able to offer every package to the customer, without generating a proposal... constraint solver technique before or during production implementation, package compatibility, transition rules with the current tariff, commitment / contract restrictions, regulations and fair use rules, campaign budget and stock limitations, 5 Restrictions such as limiting the frequency of sending emails to the same user (anti-spam) are mandatory. to make technically “invalid” or “operational impractical” to ensure that proposals of this type are eliminated during the production phase It is being structured. The artificial intelligence module (7) in the system in question (1) is the Deep Q-Network (Deep Using a reinforcement learning architecture (Q-Network - DQN) or similar, is proposed. Instead of classical rules or static models, the choice decision is made through safe reinforcement learning. to optimize the customer profile, competitor price vector, and state. Obtain network / service terms, take action by accepting the offer package or price / discount 15 It is configured to select the parameter. The artificial intelligence module (7), income increase in customer satisfaction and acceptance rate, reduction in customer churn risk, together with its objectives of price stability and campaign fatigue control It is structured to optimize. The artificial intelligence module (7), reward The Analytical Hierarchy Process (Analytic 20) is used in weighting the function. Hierarchy Process (AHP) or Ranking Based on Similarity to the Ideal Solution Technique for Order of Preference by Similarity to Ideal Solution- Using multi-criteria decision-making methods (TOPSIS) for specific risks Guardrail policy to prevent actions that exceed certain thresholds implementing; automatically rejecting offers where the risk of losing a customer exceeds a certain threshold. 25 to eliminate or move back to a safer offer category It is being structured. The control module (8) in the system (1) of the invention, in small score differences No changes to offers, no new offers to be generated before the specified period expires, daily / weekly offers 30 By implementing mechanisms in the form of limits, this technical measure benefits both the user. 8 to keep the experience stable and to prevent the model from behaving erratically It is being structured. The presentation and feedback module (9) included in the system (1) which is the subject of the invention, selected transmitting the proposal to the customer interfaces (A) via the application programming interface, 5 For low latency, use "Apache Kafka" or "Spark Streaming". using frames, accept / reject, view, click, purchase, cancel Collecting customer responses in this form through the event flow and artificial intelligence providing feedback to the model, the concept if a decrease in model performance is detected. Triggering the update with a drift detection approach and providing a secure 10 when needed. "It is being structured to revert to the default policy." Industrial application of the invention In the system in question (1) data collection module (2), customer's Telekom network 15 usage details, internet browsing data, device type, location information and collects past bill payment habits. Data is protected under the Turkish Personal Data Protection Law (or GDPR). Within the framework of compatibility, it is anonymized and processed. Competitive price and campaign monitoring module (3), this module works simultaneously, competitor API integration or Web 2.0 from operators' websites and digital channels It extracts current tariff information using scratching methods. The collected data is marked as "up-to-date". By labeling the data with "score," the system is prevented from making erroneous decisions based on outdated data. The processing module (4) combines the incoming raw data. Missing data is processed with KNN Imputation. The process is completed, and outliers are cleaned using ZS Score analysis. Then the machine 25 derived variables for learning models such as "price sensitivity" and "package completion speed". Features are calculated and the Online Product Store is designed for low-latency access. It is kept on. The forecast module (5) uses the processed data to predict the customer's possible It calculates the likelihood of accepting an offer and the risk of changing operators. This In the calculation, gradient boosting algorithms such as XGBoost / LightGBM are used, and LSTM networks are used for time series analysis. Proposal validity check 30 In module (6), the technical and legal suitability of the possible proposals to be produced is checked. 9 This module functions as a "constraint solver"; regulatory rules, customer Checks current commitment status and campaign budget limits. Applicable. Non-existent options are eliminated (by the "masking" method). Artificial intelligence module (7), Deep Q is the decision-making center that selects the most optimal option from among the filtered candidate proposals. This engine, which uses a Network (DQN) architecture, maximizes revenue while serving customer 5. It chooses the action that minimizes the risk of loss. At critical risk levels, it selects "Safe". This is where the barrier mechanisms that prevent people from leaving the "Zone" come into play. Decision criteria are weighted using AHP / TOPSIS methods. Control module (8), The firmness of the decision made is tested here. Frequent changes of offer to the customer. To avoid reflection, resistance to change in small score differences and a certain period of 10 Waiting filters are applied. This prevents the user from experiencing offer fatigue. obstacles. In the presentation and feedback module (9), the final decision is RESTful or gRPC through APIs based on the customer's front-end interface (Mobile Application, Web) The information is transmitted to the customer via the Online Transaction Center or Call Center screen. Customer response to the offer (View, Accept, Reject, Ignore) Apache Kafka 15 It is fed back in real time via similar streaming platforms. This data is both a It is used to train the next decision, and also results in a decrease in model performance. It feeds the module that monitors whether (Concept Drift) is occurring. If performance drops, the system... It automatically triggers the retraining process. The data collection module (2) in the system (1) which is the subject of the invention, competitor price and campaign monitoring module (3), data processing module (4), forecasting module (5), offer Validation check module (6), artificial intelligence module (7), control module (8), Presentation and feedback module (9) and customer interface (A) Personal Data It operates within the scope of the Data Protection Law (KVKK). 25 Around these fundamental concepts, the invention's subject is "AI-Powered Dynamic and..." There are many different things related to the "Competitive Tariff and Offer Personalization System (1)". It is possible to develop applications, and the invention is illustrated with the examples described here. It cannot be restricted, it is essentially as stated in the claims. 30

Claims

REQUESTS 1. Customized tariffs and offers for customers by telecommunications operators. Competitive pricing optimized in real time with artificial intelligence. developed to enable this; 5 - real-time and historical usage signals of a telecommunications customer to collect and process data using pseudonymization techniques including at least one configured data collection module (2); - Implementing pricing and promotions of competing operators through programming. 10 in real time via interface or web scraping methods at least one competitor's price and campaign structured to collect monitoring module (3), - data completion and outlier analysis on collected raw data. to carry out cleaning and normalization processes at least one data processing module configured (4), 15 - the probability of offer acceptance on a customer basis and customer loss after the offer. Estimating the probability, identifying offers with a high probability of acceptance, and the customer proposals that will not increase losses and will provide revenue optimization at least one prediction module configured to select (5), - Regarding the problem of not being able to offer every package to every customer, 20 constraint analyzer before or during proposal generation implementing the technique, package compatibility, switching with the current tariff rules, commitment / contractual restrictions, regulation, and fair use rules, campaign budget and stock restrictions, all for the same user Imposing restrictions in the form of transmission frequency limitations, technical 25 as “invalid” or “operationally impractical” to ensure that bids are eliminated during the production phase At least one proposed validity check module is configured (6), - using reinforcement learning architecture, making the offer selection decision classical. Instead of rules or static models, with safe reinforcement learning 30 to optimize, in terms of the situation, the customer profile, the competitor price vector 11 and obtain the network / service terms and conditions, take action by submitting the offer package or at least one configured to select the price / discount parameter artificial intelligence module (7), - not changing offers for small score differences, before the specified time expires mechanisms such as not generating new offers, daily / weekly offer limits 5 By implementing this technical measure, we can both stabilize the user experience. to both maintain and prevent the model from behaving erratically at least one control module configured (8), - selected offer, application programming interface with customer At least one presentation configured to transmit to interfaces (A) and back 10 a system characterized by a supply module (9) (1).

2. The telecommunications customer's data, voice, and text message consumption, and roaming behavior, Usage details, internet browsing data, device type, peak hours usage, location information, past billing habits, digital channel 15 Interactions, in real-time and in the form of previous campaign responses. Gathering historical usage signals, data alias identification technique to process with; retaining only the attributes necessary for bid optimization. and data collection structured to implement data minimization processes. A system like the one in Claim 1 characterized by module (2) (1). 20 3. Application programming tools to analyze the prices and campaigns of competing operators. to collect in real time using facial or web scraping methods characterized by the configured competitor price and campaign monitoring module (3) a system like any of the above requests (1). 25 4. On competitor data: timeliness score, reliability score, and latency. By calculating the compensation, incorrect / outdated competitor information can erroneously influence pricing decisions. to prevent it from having an effect, weighted competition signal Competitive price 30 structured to enable its use in vector form. 12 and characterized by the above campaign monitoring module (3) a system like any of the requests (1).

5. Value Assignment using K-Nearest Neighbor on the collected raw data or Mean / Median Assignment for Missing Data, Z-score or 5 Outlier removal using Interquartile Range methods and implementing normalization, price sensitivity score, package overflow risk, High peak hour consumption rate, early warning signals of customer loss, Highly influential features on the offer, such as the campaign fatigue metric. data processing module (4) configured to enable its production and 10 as in any of the above characterized claims system (1).

6. To determine which features are critical in the selection of proposals. Random Forest Feature Significance or Shapley Cumulative Explanations 15 data processing module (4) configured to use its methods as in any of the above characterized claims system (1).

7. Probability of offer acceptance and post-offer customer loss on a customer basis. 20 Estimating the probability of acceptance, identifying offers with a high probability of acceptance, and customer loss. to select offers that will not increase costs and will optimize revenue The above is characterized by the structured prediction module (5) a system like any of the requests (1).

8. Gradient Boosting, Extreme Gradient Boosting, and Slight Gradient Boosting for Estimation Supervised learning models in the form of Gradient Boosting Machines and Long Short Term Memory / Gated Recurrent Transactions for Time Series Trends Using unit-shaped architectures, K-Tier Crossover in the educational process Improving generalization capability through validation, hyperparameters, Bayes 30 13 Optimizing through optimization, using model outputs, and incorporating them into the proposal decision. to enable the generation of calibrated probability scores The above is characterized by the structured prediction module (5) a system like any of the requests (1).

9. Regarding the problem of not being able to offer every package to every customer, the offer... constraint analysis technique before or during production implementing, package compatibility, transition rules with the current tariff, commitment / contract restrictions, regulations and fair use rules, campaign budget and stock constraints, as well as restrictions on the frequency of sending to the same user, are among the 10 factors. imposing restrictions that are technically “invalid” or “operational” proposals deemed "not feasible" should be eliminated during the production phase. with the offer validity check module (6) configured to provide as in any of the above characterized claims system (1). 15 10. Using a deep Q-Network or similar reinforcement learning architecture, The proposal selection decision is made using safe reinforcement instead of a classical rule or static model. Optimizing through learning, considering the customer profile and competitor pricing. Obtain the vector and network / service conditions, and submit the offer package as action on 20 or artificial intelligence configured to select the price / discount parameter any of the above requests characterized by module (7) a system like one of them (1).

11. Increased revenue, customer satisfaction and acceptance rate, and a 25% reduction in customer churn risk. reduction, price stability and campaign fatigue control an AI module structured to optimize goals together (7) like any of the above-mentioned claims characterized by system (1). 14 12. The Analytical Hierarchy Process or Weighting of the Reward Function This is a technique for ranking preferences based on their similarity to the ideal solution. using criteria-based decision-making methods, exceeding certain risk thresholds Implementing a policy of putting up barriers to prevent actions; customer loss. Automatically eliminate bids with a risk above a certain threshold or less than 5 AI structured to pull back into a safe offer category any of the above requests characterized by module (7) a system like one of them (1).

13. Not changing offers in case of small score differences, setting a new 10 before the specified time expires. mechanisms such as not generating bids, daily / weekly bid limits By implementing this technical measure, we can both keep the user experience stable. and structured to prevent the model from behaving erratically any of the above requirements characterized by the control module (8) a system like one of them (1). 15 14. Submit the selected proposal to the application programming interface and customer interfaces. (A) transmit using “Apache Kafka” or “Spark Streaming” for low latency Using frames like these: accept / reject, view, click, buy Collecting customer responses such as purchase and cancellation through the event flow and 20 Providing feedback to the AI ​​model, detecting a drop in model performance. If this is done, the approach to detecting conceptual shifts can trigger the update, and to return to a safe "default policy" when necessary characterized by the structured presentation and feedback module (9) a system like any of the above requests (1). 25

Images