Resources · 3 min read
AI ROI in retail: how to calculate it without the smoke
“Companies cut inventory 20-30%” is other companies' ROI, on other data, measured by whoever's selling. Yours gets computed from your numbers — or it doesn't get computed.
Brochure ROI isn't your ROI
The proposal comes with a bar chart going up and an industry study: so much less cost, so much more revenue. Those numbers exist — at other companies, with other catalogs, measured by someone with something to sell. They're good for knowing the ground is fertile; they're useless for your board. ROI that survives questions is built from your history, your margin and your payroll.
The three ways AI pays in retail
Revenue that stopped leaking: the stockout avoided, the 10pm question answered. Cost that stops being spent: payroll hours on tasks that now run on their own. Decisions made better: buying what turns instead of what someone's gut says. Each one is measured differently, on its own clock. Adding them into a single “estimated benefit” without showing the parts is the first sign of smoke.
What it's compared against: total cost, not the fee
The cost side carries everything: setup, the monthly, API consumption — and your team's hours on integration, review and approvals, which nobody invoices but somebody pays. Comparing the benefit against the monthly fee alone inflates the result, and it shows. Against total cost, the number comes out smaller and more defensible — which is exactly what you want in a boardroom.
The honest experiment: baseline and control
Before switching anything on, measure a month or two: sales, stockouts, service hours. Afterward, measure the same. And if you can, keep a control — a category or a store where the system doesn't run — to separate the system's effect from the season's. Without a baseline, any December looks like the bot's doing; with a control, even the board's skeptic runs out of arguments.
When the number shows up — and when it doesn't
The honest clocks: after-hours service shows up in weeks; forecasting needs months of retraining to earn its keep. And there are cases where the ROI doesn't close — small catalog, low volume, thin margin — where the right answer is don't buy yet. The vendor who never tells you “this isn't worth it for you” isn't calculating your ROI: they're calculating their own.
Rather go through it on your own case?
A 15-minute audit of your real operation. You leave with an honest read on what to automate first, and what to leave alone.
Book a 15-min audit