The short answer: Before scaling an AI pilot, answer eight questions, one per Friction Default, and score each from 1 (rarely present) to 5 (deeply embedded). Under the Velocity Gap Framework, a total of 29 to 40 means a severe gap: organizational habits need to change before the pilot can succeed at scale.
How to use this checklist
Score each question from 1 (rarely present) to 5 (deeply embedded). Be honest; nobody is watching. Add the eight scores, then read your total in the table below. Better still, have several people score separately and compare.
The eight questions
- Permission Loop. How long does it take to get approval to try something, compared with how long the first rough version would take to build? If approval takes longer than building, score high.
- Polish Paralysis. Can someone outside the team use a rough version this week, or are we polishing before anyone sees it? If nothing is shown until it is polished, score high.
- Meeting Dependency. When we disagree about what to build, do we schedule a meeting, or build a rough demo and show it? If the default is a meeting, score high.
- Structured Waiting. What are we waiting on from other people's calendars that we could move past with a provisional decision? If work stops while we wait for feedback, score high.
- Planning Inversion. Has planning taken longer than a rough build would have? Have we shipped nothing in two weeks? If the plan is longer than the prototype would be, score high.
- Deck Over Demo. Do stakeholders see working software, or slides about it? If decisions are made from decks, score high.
- Consensus Lock. Do we align before acting, or run an experiment first and align on what it showed? If nothing starts until everyone agrees, score high.
- Readiness Hoarding. How long does unfinished work sit before anyone outside the team sees it? If drafts wait for "ready", score high.
Reading your score
| Total | Reading | Do this next |
|---|---|---|
| 8 to 16 | AI-native ready | Focus on the new bottlenecks: clarity, ambition and distribution. |
| 17 to 28 | Moderate friction | Pick the top two or three defaults and intervene on those first. |
| 29 to 40 | Severe Velocity Gap | Change the organization before trying to scale the pilot. |
The scale and bands are from the Velocity Gap Framework. The free diagnostic gives you the same scoring with guidance for each default.
Three more questions from current research
The eight questions look at habits. These three look at the work itself, and come from what the latest research says separates pilots that pay off from those that do not.
- Which workflow changes? McKinsey's 2026 survey found nearly three-quarters of high performers fundamentally redesigning workflows, versus about a quarter of other respondents. If the answer is "none, we are adding a tool", expect limited payoff.
- What is the success metric, and what does one outcome cost? Gartner forecasts that over 40% of agentic AI projects will be scrapped by the end of 2027, citing escalating costs and unclear business value. Write the metric and the cost per outcome down before you scale.
- What is the promote-or-stop date? Choose the date and the evidence that would make you scale or stop. A pilot with no end date drifts into the same cancellation pattern.
For the numbers behind these, see The State of Enterprise AI, Q4 2026. For a full P&L model of an AI agent, including pilot-to-production decision gates, see The AI Agent P&L.
If your score is high
The Velocity Gap Framework guide ($29) is the 90-day plan for removing the defaults you scored highest: a scoring rubric, role-specific action plans for executives, managers and individual contributors, and week-by-week steps. If you are not sure it is the right guide, start here.
Frequently asked questions
How do I know if my AI pilot is ready for production?
Answer the eight checklist questions honestly and add the scores. Totals of 8 to 16 read as ready on the habits that stall pilots; 17 to 28 means target the two or three worst defaults first; 29 to 40 means change the organization before scaling. Then confirm the three research-based questions below: the workflow that changes, the success metric, and the promote-or-stop date.
Who should fill in the checklist?
Ideally several people at different levels, scoring separately and comparing. Gaps between how executives and the people doing the work answer are themselves a signal.