Most "AI for retail" stops at a dashboard. A chart tells you a product is selling fast; you still do the work. The leverage is in closing the loop — turning the forecast into the next action, with a human gate. That's how we built the AI layer in Develshops, our retail ERP + ecommerce platform.
Forecast on real sales, not vibes
The model projects each product's sales over the next 30 days from its actual velocity, then crosses that against current stock to compute days of coverage — flagged red when stock runs out in under a week. No generative guessing: it's statistics over your own transactions.
The forecast becomes a draft action
Below coverage threshold, the system drafts the suggested purchase order to the usual supplier. You approve, and it creates the PO — one click, not a re-keyed form. Perishables get the same treatment: batches nearing expiry surface a suggested markdown that, on approval, is applied to the price.
A tray, not an autopilot
Everything the AI proposes — reorders, markdowns, prices — lands in a suggestion tray. You review, approve or reject. Approving is what executes the action; nothing is applied silently. The engine is configurable (hosted API, local Claude or Ollama) and gated per store.
Prediction is a feature. Prediction wired to an approved action is an operation.