Guide
Retail AI
What AI changes in retail, what it does not, and how to start without a science project.
I have spent 15 years in retail pricing, rebates, point of sale and the ERP platforms underneath them, and I now build agents and tooling with AI. This is the short version of what I tell retailers who ask where to begin.
What retail AI is, plainly
Retail AI is software that reads, predicts or decides using retail data, then advises a person or acts for them. There are three kinds, and it helps to know which one you are buying.
- PredictDemand, price response, which customers will leave. Old idea, better tools.
- GenerateProduct search, descriptions, service replies. Fast to demo, easy to get subtly wrong.
- ActAgents that create an order, match a delivery or route an approval. The newest kind, and the one that needs the most care.
Where it earns its keep first
- Search and discoveryCustomers find the right product when search understands intent, and it only works with clean product data.
- Pricing and rebatesHeavy on rules and data, so the gain is large and measurable. I have seen a monthly rebate run drop from days to hours.
- PlanningDemand and replenishment, where a better forecast pays for itself in stock held.
- Store operationsRostering and task allocation, where small gains repeat across many stores.
- Back officeFinance and procurement, where documents arrive in every format.
What usually goes wrong
Four things, in the order I meet them. The data sits in silos, so the model sees half the picture. Nobody owns the result, so the pilot ends when the sponsor moves on. The pilot never gets a path to production, so it stays a demo. And trust is never designed: nobody decided who checks what, so the first visible mistake ends the project.
How I would start
- 1. Pick one processWith a number you can measure today.
- 2. Draw itTwo or three boxes and an arrow. If you cannot draw it, you cannot automate it.
- 3. Name the ownerOne person who is accountable for the result, not a committee.
- 4. Decide who checksWhich decisions stay with a person, and why.
- 5. Measure before and afterThen widen, and only then.
Agentic commerce
The next shift is agents that shop and transact for a customer or a business. That depends on open standards such as the Model Context Protocol, and on retailers having product, price and stock data an agent can read and act on safely. Retailers who tidy that data now will be ready, and the rest will find out the hard way.
Questions people ask
What is retail AI?
Retail AI is software that reads, predicts or decides using retail data, such as prices, stock, customers, orders and invoices, and then advises a person or acts on their behalf. It comes in three kinds: models that predict (demand, price response), models that generate (search, product content, service replies), and agents that act (create an order, chase a supplier, route an approval).
Where should a retailer start with AI?
Start with one process that has a clear number attached, such as the days a monthly rebate run takes, the time spent answering the same service question, or the share of searches that return nothing. Draw the process first, decide who approves what, measure before and after, and only then widen it.
Is retail AI only for large retailers?
No. The cost of the models has fallen far enough that the limit is usually data and ownership, not budget. A mid-sized retailer with clean product and price data and one accountable owner can get further than a large one with five pilots and no owner.
How long until retail AI shows a result?
For a narrow, well-defined process, a first measurable result is realistic within weeks. Anything that touches several systems, such as pricing, ERP and the website together, takes longer, and the time is mostly spent on the data and the approvals, not the model.
What is agentic commerce?
Agentic commerce is AI agents finding, comparing and buying on behalf of a customer or a business. It depends on standards such as the Model Context Protocol (MCP), an open standard for connecting agents to business systems, and on retailers having product, price and stock data that an agent can read and act on safely.
More from me: the work, how I use AI to test strategic thinking, about me.