Case Study: AI-driven legacy online retail platform modernization
An omnichannel retailer of apparel and home goods needed to move away from a restrictive monolith that was slowing campaigns and constraining growth. SumatoSoft modernized the platform and introduced AI-driven recommendations to improve conversion rates.
Project details:
About the Client:
The Client is an omnichannel retailer that sells apparel and home goods both online and offline. The online business had grown faster than the core platform, and the Client needed a flexible architecture that could support AI-driven personalization.
Location: UK
Industry: Retail, eCommerce
Team size: 10 specialists
Project duration: 7 months
Business challenge:
The Client’s eCommerce platform was a monolith, so catalog, pricing, promotions, and checkout logic were hard to change, and integrations with ERP, OMS, and marketing tools relied on batch jobs and ad hoc scripts. All these meant that releases were slow, experiments were expensive, and the team had no reliable event stream for customer behavior.
At the same time, the business wanted to introduce AI recommendations to increase conversion and average order value. That goal was blocked by fragmented data, weak observability, and the risk of breaking critical order and payment flows during modernization. The company needed a safe migration path, better data capture, and an architecture that could support AI in production.
Additional requirements
- Keep checkout, pricing, and order flows stable during migration
- Launch AI recommendations without making them the source of truth for stock or pricing
- Improve campaign time-to-market and make A/B testing a standard process
Our solution
SumatoSoft modernized the Client’s legacy platform with a phased Strangler Fig migration toward headless commerce. We fronted the monolith with a BFF and API facade, then peeled off catalog, pricing, promotions, checkout, and customer profile as independent services. In parallel, we built a unified event pipeline for views, cart actions, and purchases, wired it to the catalog, and shaped the data for AI. On that foundation, an AI recommendation engine serves real-time APIs to the storefront and batch outputs to email and push. Observability, experimentation support, and phased cutover controls keep migration risk low.
Additional info about the case
The new setup treats AI recommendations as an operational layer, bringing behavioral events, catalog data, and campaign context into a single usable flow that makes personalization measurable, testable, and easier to improve over time. The migration was staged to avoid disruption in payments and order processing.
Additional features:
- AI recommendation APIs for home, PDP, and cart
- Event stream for views, cart actions, and purchases
- A/B testing and observability for conversion flows
Business value
Before:
- Monolithic architecture slowed releases and made changes risky
- Recommendations were missing or limited to static rules
- Customer behavior data was fragmented or incomplete
- A/B testing was hard to run and hard to trust
- Low observability made diagnosis and optimization slow
After:
- Headless architecture and domain services improved release speed and reduced change risk.
- AI recommendations improved product discovery across key storefront pages.
- Unified event data created a usable foundation for personalization and analytics.
- A/B testing became part of the normal optimization process
- Better observability helped teams monitor conversion, latency, and errors faster
