Consumer
The store already has software. The exception still needs a person.
Inventory mismatches, vendor claims, returns, and store questions still move through email and spreadsheets. A recommendation engine on the site does not reduce that load.
How I see it
Retail AI fails when it stays in merchandising slides
Retail operators feel AI as a merchandising conversation. The expensive repeatable work is often elsewhere: a claim packet, an allocation exception, a wholesale order that does not match, a store that needs a decision.
AI is useful when it interprets the exception, assembles the evidence, and writes the next action into the system. It is not useful as another personalization project that never reaches the people who close the books on a week.
The first workflow should have volume and a completed result: a closed claim, a resolved mismatch, a completed store request. Measure labor and cycle time. Do not start with a brand-new customer experience program.
Keep merchants on taste and buy decisions. Move the packet work that currently sits around those decisions.
Common mistakes
What teams usually get wrong.
Starting with personalization because that is the industry story
Personalization can be a later revenue bet. The first operational win is usually an exception queue.
Leaving stores on email
If the new path is a portal the store will not open on Saturday, you designed for headquarters.
Automating a report nobody uses to act
A prettier dashboard is not a workflow change.
A useful diagnostic
Five questions before you fund the work.
Which exception queue grows when sales grow?
That queue is the leverage point.What is a completed case: a paid claim, a corrected allocation, a closed store ticket?
If the unit is unclear, stop.How much of the time is gathering evidence versus deciding?
Gathering is usually the AI-addressable slice.Will stores or vendors change how they submit?
If intake cannot change, the redesign is smaller than it looks.Who owns the queue after launch?
Merchandising and IT together is not an owner.
Economic model
Exception loadexceptions per year × hours to evidence and close × loaded cost, plus margin lost while the case sits
If margin loss is the larger term, cycle time is the case, not headcount.
Three credible paths
How far should you go?
Do not force one solution. Choose the path the economics, the risk, and the organization can support.
Improve the work you already have
Keep the process mostly intact and use AI on the bottlenecks that create delay, rework, or follow-up.
The workflow is already sound and a few steps create most of the friction.
Gains are usually incremental. The operating economics do not change much.
Redesign the workflow around AI
Question every handoff, queue, and duplicate step, then rebuild the process around what AI can now do.
The process grew over years and coordination now costs more than the work itself.
Requires process change, clearer ownership, and a willingness to retire old steps.
Put an agent on a high-value outcome
Give an agent responsibility for one valuable result across systems, with humans at the control points that require judgment, authority, or risk acceptance.
The workflow is high-value, variable, multi-step, and worth engineering for production.
Needs stronger architecture, evaluation, controls, and monitoring. This is not a prompt project.
When this is the wrong next step
Do not start with a sector program.
- The only acceptable project is a customer-facing recommendation demo.
- Store teams will not be part of discovery.
- There is no system of record for the exception, only inboxes.
- Volume is a handful of wholesale accounts with relationship-only service.
A useful next step
Bring one expensive workflow from this industry.
We guide the implementation, go deep on the technical path, and stay hands-on through operations — or tell you when a simpler answer is better.
Discuss an AI opportunity