Multi-webshop back-office automation and AI shopping assistant

One source of truth instead of fragmented data

The assessment revealed the real bottleneck. The company had no ERP or PIM, and product data lived only in the outdated webshop admin. Overlapping products had to be managed by hand across five separate interfaces, and there was no recorded margin data, even though purchase prices depend on exchange rates. So the solution starts not with another tool but with a shared data layer.

Technical implementation and business impact

Central inventory and data architecture

A lightweight central inventory layer keeps stock in sync across all five webshops via webhooks, updating immediately after every successful purchase. This ends overselling. On top of it sits a margin calculator that links supplier prices to the daily central bank (MNB) exchange rate and gives a real-time view of margins.

Precise financial and logistics automation

A fuzzy matching algorithm pairs daily bank statements with orders by payment reference, amount and name. Refunds are collected into a single bank batch transfer file, and for couriers the system generates API-based waybills or courier-specific export files. An overnight job produces warehouse-optimised pick lists by morning.

Measurable business impact and ROI

The biggest savings potential lies in two people's daily manual bank reconciliation, manual data entry for 4–5 courier companies and spreadsheet-based refund handling. We feed real inventory and margin data back into the AI assistant's product recommendations. So the chatbot does not just drive traffic but builds profitable baskets, and its conversion can be measured end to end.

Hallucination-guarded AI and legacy compatibility

The shopping assistant answers from a vectorised index of the product catalogue, within strict guardrails. It only recommends products that exist and are in stock. The whole system is built around the existing webshop engine rather than replacing it, so the rollout carries no business risk.