# First AI workflow checklist Fill in one card per candidate AI workflow, starting with the busiest internal task. Fill in what you know; every blank line gets an owner and a date. Any blank line means: this workflow is not ready to be the first lever. | Question | Answer | Owner / by when | | --- | --- | --- | | The candidate workflow (one task, done at volume) | | | | Margin per decision: what money rides on one decision | | | | Decision volume: how many of these decisions a week | | | | Does the output change a price, a promotion, an order or a shelf? | ☐ Yes ☐ No | | | If yes: guardrails in place before anything goes live | | | | If yes: the named owner to call when it goes wrong | | | | Forecasting, optimization or RPA already running on this task | | | | What the new buy adds over what already runs | | | | Vendor's specific problem already solved for a retailer (not the platform demo) | | | | When the vendor's claim runs on our data | | | | The one KPI we will measure | | | | The one business process the pilot runs in | | | | Baseline snapshot: value, date, taken before day one | | | | Where the pay sits on the margin map (price and markdown, stock and availability, assortment, the product page, store labor, media, shrink) | | | | Decision and date | ☐ FIRST LEVER ☐ WAIT FOR GUARDRAILS ☐ NOT READY, OWNER BY ______ Date ______ Signed ______ | | ## Example (fictional): product-page content rewrite at a housewares chain Fictional example. | Question | Answer | Owner / by when | | --- | --- | --- | | The candidate workflow (one task, done at volume) | Rewrites product descriptions and bullets for the web shop, about 2,000 items a month | n/a | | Margin per decision: what money rides on one decision | Small: no direct price change, each rewrite nudges conversion on one page | n/a | | Decision volume: how many of these decisions a week | About 500 rewrites a week across 60,000 listed items | n/a | | Does the output change a price, a promotion, an order or a shelf? | No: pages change, no shopper-visible price, promotion, order or shelf does | n/a | | If yes: guardrails in place before anything goes live | n/a (the boundary test answered No) | n/a | | If yes: the named owner to call when it goes wrong | n/a (the boundary test answered No) | n/a | | Forecasting, optimization or RPA already running on this task | A 6-year-old rules template writes item titles; nothing AI-named runs on content | n/a | | What the new buy adds over what already runs | Rewrites bullets and attributes, not just titles; the old template stays | n/a | | Vendor's specific problem already solved for a retailer (not the platform demo) | Platform demo only; no named solved problem **Fails the bar** | E-commerce lead, answer by 4 Dec | | When the vendor's claim runs on our data | A 200-item trial on our own catalog, week of 8 Dec | E-commerce lead | | The one KPI we will measure | Conversion rate on rewritten pages | n/a | | The one business process the pilot runs in | Weekly product-content refresh | n/a | | Baseline snapshot: value, date, taken before day one | **Blank: not ready** | Data analyst, by 27 Nov | | Where the pay sits on the margin map (price and markdown, stock and availability, assortment, the product page, store labor, media, shrink) | The product page | n/a | | Decision and date | ☐ FIRST LEVER ☐ WAIT FOR GUARDRAILS ☒ NOT READY, OWNER BY 11 Dec Date 20 Nov "No baseline snapshot and the vendor claim is untested on our data." | E-commerce director | From the course AI in retail and consumer brands, Dr. Tony Hoang · v1 · Oct 2026