Testbusters

The problem

Testbusters is an Italian EdTech company that prepares students for university admission tests and distributes courses, study guides and learning tools through digital channels and bookstores. Its four eCommerce sites ran on a legacy WooCommerce infrastructure. The system worked, but users dropped out at the most critical stages of the funnel and growth costs increased with complexity. Operations also relied heavily on manual work. Post-purchase management, catalog feed synchronization and information flows across teams required constant coordination, making operations harder to scale.

The migration

As Product Manager, I led the progressive, end-to-end headless replatforming of four eCommerce sites from WooCommerce to a composable stack: BigCommerce for transactional logic and payments, Builder.io as the CMS and Prismic for the blog.

The headless approach served three strategic goals: address structural friction in the funnel, particularly around payments, bundles and pricing; improve navigation and performance on mobile; and give marketing and business teams operational autonomy without relying on IT. I completed the infrastructure with technical and semantic work across SEO and GEO, a custom data layer, GTM and GA4 tracking, GSC monitoring, and Looker Studio and Tableau dashboards. Funnel decisions could then be based on shared, verifiable data.

Beyond the storefront

Post-purchase automation

I replaced the sequence of manual tasks after each transaction with a centralized, cross-brand n8n architecture for the four eCommerce sites. The workflows manage catalog feeds, integrations with shipping services and third-party systems, and the delivery of individual services.

To avoid duplication and fragile flows, I designed reusable components, shared helpers, standardized payloads and configuration-driven logic. Idempotency and retries with backoff handle potential failures, while monitoring and alerting detect anomalies before they reach the end user. APIs and webhooks standardize inbound and outbound data exchange. An internal dashboard completes the architecture, providing governance and operational monitoring.

Internal AI assistant

Repetitive internal information requests took time away from the product manager and operations team every day. I designed an n8n-based RAG assistant that could be queried through Slack and connected it to the documented knowledge base in the repository. An ingestion flow collected, indexed and chunked the content into a vector database, while the RAG workflow handled retrieval and response generation.

A feedback loop and iterative evaluation cycles measured relevance and accuracy, allowing the retrieval process to be refined over time. Guardrails restricted answers to validated company documents, reducing the risk of hallucinations. The team could access information in self-service mode without depending on the product manager’s availability.

Multi-brand product feed

Brands and catalogs followed different structures, so every new integration required rebuilding feeds and mappings from scratch. I introduced a normalization layer based on a shared taxonomy, validation systems and brand-specific mappings.

The infrastructure feeds Google Merchant Center and AWIN. Integrating a new catalog now requires configuration rather than new development.

The result

The work on the funnel’s critical friction points increased conversion by 27% year over year, reduced cart abandonment by 34% and raised AOV by 15.1%.

On the operational side, the post-purchase automation went live in two weeks, compared with the month initially estimated to build it as a software layer, and reduced the cost of manual operations by 35%. The RAG assistant freed up roughly three hours a day previously absorbed by repetitive information requests.

These results did not come from a simple storefront redesign. They came from a coordinated intervention across architecture, funnel, data and operational processes.