Why AI Pilots Stall: Data, People, Approval, Sponsor

AI pilots stall on one of four factors: the data wasn't owned, the people doing the work never signed up, nobody knew who approves what evidence, or no one was accountable for the number after launch. The stall is diagnosable in ten minutes — and the weakest factor, not the technology, is what to fix first.

The named framework: the Four Factors

1. Data — AI pays off first where inputs already exist in your systems. If nobody can produce two months of clean numbers for the work, any AI return is an opinion, not a figure (/pillar/ai-pays-off, live 11 Oct 2026).
2. People — the hours AI replaces belong to someone. If no named person's week changes visibly, there is no saving to measure — and no adoption either.
3. Approval — most SME AI pilots die between a promising test and a funded rollout. The gap is who signs, what evidence they require, and what happens to the money if the result disappoints.
4. Sponsor — one person must own the number the initiative is judged on, month after month. Without a named sponsor, the pilot fades the moment something urgent competes with it.

The four stall patterns

Stall pattern	Factor	What it looks like in the wild	Fix first
"The pilot worked but nothing happened after"	Approval	Test impressed; nobody knew what evidence the sign-off needed	Build the CFO-format case before asking
"We bought tools, usage fell off"	People	No named person's week changed; usage was voluntary	Redesign one workflow with its owner
"We can't prove it worked"	Data	No baseline numbers, no two-month CSV	Record the baseline before the next change
"It faded after the reorg"	Sponsor	The sponsor left; nobody inherited the number	Name the sponsor and the monthly check date

Sources and context

• The board-test failure pattern: spend rises, pilots multiply, cost/attrition/revenue/cycle time barely move (skillpilotadvisory.ai/about, live 11 Oct 2026).
• Malaysian enterprises often sequence AI top-down before identifying the problem, turning adoption "into an unnecessary expense" — Manminder Kaur Dhillon, CEO Supernewsroom.ai, World AI Show Malaysia 2026 (The Rakyat Post, 14 Sep 2026).
• Malaysia's National Guidelines on AI Governance and Ethics (MOSTI, Nov 2024) name human oversight, accountability and transparency as core principles — a stalled pilot almost always traces to one of these three being unassigned.
• Malaysia's Cyber Security Act 2024 sets a baseline for critical-infrastructure entities, and NACSA's Azlina Ab Aziz has said publicly the bigger gap is moving from "check-the-box" to continuous resilience (The Rakyat Post, 14 Sep 2026) — the same box-ticking pattern stalls AI pilots.
• The model's own worked example: readiness index 50 ("ready with gaps"), approval the factor to fix first, in a fictional retail case on /scan (fetched 10 Oct 2026).

Diagnose yours in ten minutes

Scan's readiness map scores the four factors into a 0-100 index, names the weakest, and recommends one first move. No calls, no implementation. Then the fix list follows the table above — data first, because no other fix is provable without a baseline.

FAQ

Can a pilot recover after stalling? Yes — name the factor, fix it, and restart with a baseline; the framework exists for restarts, not just first runs.
Which factor is most common? Approval, in owner-run businesses; the case is never in the approver's format.
Is this a change-management framework? It is a readiness-and-proof framework: the four factors decide whether AI pays off provably.
Does the framework apply to non-AI automation? It applies to any measured workflow change; the questions are tool-agnostic.