The AI pilot go/no-go checklist.
Most AI pilots don't fail on the model — they fail on missing owners, missing metrics, and a missing path to production. Run these 12 checks before you fund the pilot. If a check fails, that's not a reason to cancel — it's the exact thing to fix first.
// Self-apply in ~10 minutes · print it · bring it to the kickoff
The 12 go/no-go checks
-
01 One workflow, not a platform
Go — The pilot targets a single workflow with a visible start and end — one intake queue, one document type, one call flow.
No-go — The charter says "AI transformation," "AI platform," or names more than one department. Scope that wide is how pilots stay pilots.
-
02 A named owner outside the AI team
Go — One operational owner — a person, not a committee — whose team runs the workflow today and whose numbers move if it works.
No-go — The pilot belongs to IT, innovation, or "the AI team." If nobody in operations owns it, nobody adopts it.
-
03 You can state the dollar value out loud
Go — Someone can say what the process costs today — hours × loaded rate, error cost, or revenue leaked — in one sentence, with a number in it.
No-go — The value is "efficiency" or "innovation." A pilot that starts without a dollar figure ends without one.
-
04 One success metric, agreed before the build
Go — A single metric, wired to the P&L, written down before any code exists — and a baseline measurement of it taken now.
No-go — "We'll know it when we see it." No metric before the pilot means no verdict after it.
-
05 Production data exists and you can reach it
Go — Real historical data from the actual workflow is available, accessible, and legally usable — not a curated sample.
No-go — The demo runs on hand-picked or synthetic data. A model that has never seen your messy real inputs has never been tested.
-
06 A path to production, decided up front
Go — Before the build starts you know where it will run, who deploys it, and who answers when it breaks at 2 a.m.
No-go — "We'll figure out deployment after the demo." That sentence is the single most reliable predictor of pilot purgatory.
-
07 The cost of a wrong answer is bounded
Go — Errors are cheap to catch — a human reviews the output, or a wrong answer costs minutes, not clients.
No-go — A wrong answer is expensive (money, compliance, patient safety) and there is no human in the loop. Fix the guardrail first.
-
08 AI is actually the right tool
Go — The problem involves judgment, language, or unstructured data — the things a fixed rule genuinely cannot do.
No-go — A rule, a fixed process, or a spreadsheet would solve it better. Sometimes the honest answer is automation without AI — or no project at all.
-
09 Buy vs. build is already decided
Go — You buy the commodity layers (models, hosting, retrieval) and build only the thin slice that is genuinely your workflow.
No-go — The plan rebuilds something a vendor already sells. Building beats buying far less often than partnering does.
-
10 It lives inside the existing workflow
Go — The output lands in the tools people already use — the CRM, the inbox, the ticket queue. Adoption is the default.
No-go — It is a standalone tool users have to remember to open. Standalone tools get opened for two weeks, then never again.
-
11 Scope is frozen, in writing
Go — The bounded scope is written down, and there is a named place for "can it also do X?" requests to wait until v1 ships.
No-go — Scope is already growing in the kickoff meeting. Every "also" adds weeks and subtracts the ship date.
-
12 A time box with a kill criterion
Go — A 90-day target to production, a first-month checkpoint, and an agreed condition under which you stop and walk away.
No-go — The pilot is open-ended. "Almost done" for months is not a status — it is the failure mode.
How to read your score.
- 10–12 ✓
- Go. Fund it, time-box it, and hold the first-month checkpoint. Teams in this position routinely reach production in about 90 days.
- 7–9 ✓
- Fix first. Don't start the build — spend the next two weeks closing the failed checks (they're usually owner, metric, or data access), then re-run the list.
- ≤ 6 ✓
- No-go. The pilot would join the large majority of AI initiatives that never show a return. Point the budget at the blocking item instead — that money is not wasted; the pilot's would be.
Want a second opinion on your score?
The 2-minute AI-Readiness Scorecard turns this into a 0–100 read on your whole org. Or bring one process you wish were automated to a 30-minute working call — you leave with a hypothesis, a stack pick, and a fixed-scope ballpark.