Customer Segmentation and Product Recommendations Using GenAI

Quick Summary

Challenge

A financial institution needed to segment customers across a wide range of traits — from demographics to behavior and preferences — and generate actionable insights to guide marketing and product strategy.

Solution

Tatras Data developed a pipeline that uses GenAI to select key features and describe customer subgroups, paired with unsupervised ML clustering to identify segments within a larger cohort.

Result

Customer segmentation with tailored product recommendations, deployed into the bank’s marketing systems.

Tech Stack

AI: OpenAI LLMs |

ML: Feature selection Unsupervised clustering (K-means, DBSCAN) |

Data & Retrieval: Customer profile databases Descriptive stats |

Dev: Scikit-learn LangChain Transformers |

Viz: Integrated dashboards for subgroup insights |

Security: Role-based access Internal data sandbox deployment

The Challenge

A major financial institution wanted to understand its customers on a deeper level. Not just who they were, but, what they wanted. Their goal: identify meaningful customer segments and customize offerings based on traits like behavior, lifestyle, and product preferences.

Traditional segmentation was static and coarse. The bank needed a more dynamic solution that could:

Automatically discover subgroups within larger segments
Describe each group with clear, explainable characteristics
Personalize product recommendations accordingly

A Day in the Life: Before Our Solution

A product team wants to target urban millennial customers with a new investment product. They ask the analytics team:

Which micro-segments within that demographic are most likely to respond to a risk-focused savings product?

What follows is the usual scramble. The team runs SQL queries, exports data, and manually builds clusters in Excel. Then they try to interpret them using charts and pivot tables which are often based on outdated assumptions.
Campaigns go live with broad targeting, slow iterations, and no clear sense of which segments truly moved the needle.

Pain Points:

Manual segmentation processes were slow and prone to bias
Marketing efforts lacked personalization and often targeted too broad an audience
Product teams had no clear, up-to-date insights into micro-segments
Lack of explainability in cluster outputs made it hard to act on results

Solution

1. Core Innovation

Tatras built a modular pipeline combining unsupervised ML and generative AI. The system dynamically identifies relevant features, segments users, and creates natural language summaries of each subgroup’s characteristics.

Step 1: Feature Selection via GenAI

The system identifies which features (e.g., credit behavior, digital engagement, lifestyle signals) matter most within a selected demographic or cohort.

Step 2: ML Clustering Pipeline

Using those features, a clustering engine segments customers into distinct subgroups using unsupervised learning techniques.

Step 3: GenAI-Powered Subgroup Descriptions

Each cluster is described using a language model that outputs a human-readable summary — including habits, risk profiles, product affinities, and behavior markers

2. Key Features:

GenAI-Driven Feature Selector: Dynamically chooses relevant features per target group
Unsupervised Clustering Pipeline: Segments users using models like K-means or DBSCAN
Cluster Descriptor Engine: Converts statistical summaries into human-readable insights
Real-Time Segmentation: Enables responsive campaigns and personalized experiences
Modular Deployment: Easily integrates into CRM, marketing, and product stacks.

3: Workflow Integration

The system is deployed as an internal API and visualization tool for marketing teams. Product and strategy teams can now query any customer group and instantly view subgroups with live descriptors, enabling highly targeted engagement.

Outcomes

✅ Customer segmentation across demographic and behavioral groups

🎯 Personalized product recommendations per subgroup

🕒 Reduced time-to-campaign with automated subgroup discovery

📈 Enhanced marketing precision and product uptake