Case Study · Forward Deployed Engineering
Forward Deployed Engineering: Automating Order Intake for an FMCG Distributor
The distributor asked for a demand forecast. Our embedded engineer found that the real problem was how orders came in. They fixed order intake first, then built replenishment forecasting on the clean data that resulted.
The stated ask vs. the real problem
The distributor asked for a demand forecasting model to reduce stock-outs. Working alongside sales operations and the warehouse, our engineer found that retailers placed orders by WhatsApp message, voice note, email and PDF, and staff re-keyed them into the ERP by hand. Orders arrived late, SKU names were inconsistent and entry errors were common. Any forecast built on that data would have been wrong.
How the engagement ran
Embed
The engineer spent time with order-desk staff, warehouse supervisors and finance to follow an order from message to dispatch.
Identify
Showed that order-entry delay and error rate were the root cause of both stock-outs and poor forecast data.
Design
Proposed an AI order agent that reads incoming messages and documents, matches items to the SKU master using fuzzy and semantic matching, and creates draft orders.
Build
Built and tested on a month of real historical orders, measuring matching accuracy against what staff had actually entered.
Integrate
Draft orders go straight into the ERP. Only low-confidence lines go to a review screen, and customers receive automatic confirmations.
Iterate
Handled real-world shorthand, regional languages and outlet-specific product names, and adjusted confidence thresholds with the order desk every week.
Deploy, then forecast
With clean, same-day order data flowing, delivered SKU-by-outlet demand forecasts that feed replenishment suggestions to the warehouse. Monitoring and hand-over followed.
Why this is Forward Deployed Engineering
- The engineer fixed the operational problem behind the request, not just the model that was asked for.
- Work was sequenced by business impact, so data quality came before forecasting.
- The solution runs inside the client's own ERP and messaging channels, with people reviewing uncertain cases.
- Ownership continued through adoption, monitoring and hand-over.
KPIs we design for
Technology
Have a similar challenge?
Talk to our team about what this could look like for your business.