Most retailers run marketing and inventory as separate systems with a meeting between them. So the campaign that works sells out in four days, and the one that fails leaves stock nobody planned for.
Most retail AI starts on the marketing side, because content is the easiest place to start. The bigger number usually sits on the other side of that meeting.
The gap between the campaign and the stockroom
A campaign plan is built weeks ahead, from a brief and a calendar. The buying plan is built from forecasts and supplier lead times. Neither sees what the other is doing this week. The result shows up in two places on the P&L: lost sales when a promoted line runs out, and markdown when a slow line was never pushed.
Dataiku's guide describes an AI agent for regional retail managers that tunes pricing, inventory and promotions across stores with different customers. Neptune's whitepaper gives the same advice we give clients: start with the unglamorous use cases that solve a specific profit-and-loss problem.
What changes
- Before launch, AI checks the campaign plan against stock, open orders and recent demand, and flags promoted lines likely to sell out and lines with too little cover.
- During the campaign, AI watches sell-through and flags lines moving faster or slower than planned, with a suggested response: re-order, shift budget, or hold.
- For slow stock, AI identifies lines that will need markdown and proposes which campaigns could move them first.
Who approves
- The merchandiser approves every price and promotion change and every re-order.
- Marketing approves every piece of content and every budget shift.
- AI proposes, with the reason and the data behind it. It does not change prices or place orders on its own until the proposals have a track record and the limits are written down.
Pick one number
The number we move is picked with you in the diagnostic. In retail it is usually one of three:
- Stockouts on promoted lines
- Markdown on slow stock
- Margin per campaign
Take the baseline from the last few campaigns before anything changes, and report the number after every campaign.
Where content fits
AI-assisted content production still has a place, with an approval step before anything is published. It simply works better once the campaign knows what is in stock.