Merchant-centric personalisation and offer generation, built on AWS

A FinTech business working with Cloud Combinator on AWS. The client is anonymised at their request.

The challenge had three focus areas.

Member, demographic and merchant offer data is collected, prepared and stored in Amazon S3. From there, two segmentation pipelines run in parallel. Amazon Personalize builds a model that ranks members and applies threshold-based segmentation into high, medium and low propensity groups. In parallel, a large language model on Amazon Bedrock performs the same segmentation task, so the two approaches can be compared on the same data. Each pipeline returns member IDs grouped by propensity, ready for merchants to target.

Cloud Combinator delivered an end-to-end system, deployed and validated in the client's AWS environment, that segments members by propensity to shop through both Amazon Personalize and Amazon Bedrock, and generates personalised offers via Bedrock. The client came away with a working service and a like-for-like comparison of the two segmentation approaches, plus training and documentation for its team.

• Propensity segmentation, two ways — Members grouped into high, medium and low propensity to shop using both Amazon Personalize and a Bedrock large language model, each returning member IDs by group.
• Automated offer generation — Amazon Bedrock dynamically creates personalised offers per member, combining behaviour and propensity with each merchant's rules.
• A clear approach comparison — Building both pipelines on the same data gave the client an evidence-based view of which method fitted its needs.
• Cost-efficient at scale — The system was modelled at roughly 0.0031 US dollars per offer, sized for 1,000 merchants reaching 1,000 members each per month.
• Handover — The end-to-end system deployed and validated, with training sessions and documentation for the client's team to use and maintain it.

With a working personalisation system and a clear view of the best segmentation approach, the client has a foundation it planned to take into live testing and towards a production-ready service, giving its merchants a sharper way to reach the members most likely to shop.