Why your AI pilot dies at the handover
Why AI pilots stall before production and use stays shallow: depth of use, stopping criteria, workflow redesign and baseline.
November 4, 2026 · 3 minRead article →
AI adoption · done for you
Adoption isn’t a launch. It’s a routine.
After the assessment, we run everything: we pilot, scale and embed — with measurement and governance from day one.
Scaffold · Signal · Charter
AssessPilotScaleEmbed
4 tracks
Blocking gates
Your team uses it. We operate it.
No system to maintain
Launch · the research
Launching is easy. Becoming routine isn’t.
The closer you get to daily work, the fewer people are left.
- 52%use AI at least a few times a year
- 30%use it a few times a week or more
- 15%use it every day
Routine · the method
Routine gets built one gate at a time.
The next track only opens once the last one holds.
- AssessV-01 · Baseline approved
- PilotV-02 · Pilot measured
- ScaleV-03 · Pattern replicated
- EmbedV-04 · Routine stable
A scheme of the method, not a measured result.
On the books
Nearly nine in ten use AI in some function. 37% attribute any EBIT to it. About 6% reach 5% of EBIT or more, with value they call significant.
On the plan
47% say their company has integrated AI. 25% say it has communicated a clear plan. Where AI is already in place, 14% strongly agree it changed how work gets done.
In Brazil
Half of large companies already use AI. In industry, of those using it in a function, only 9% to 14% use it broadly.
Scaffold · four tracks
Four tracks. Gates that block.
No track opens before the last one passes its gate. That’s what keeps the pilot from dying at the handover.
Association, not cause: none of these studies proves the practice produces the result. That’s why we measure each case against its own week 0.
01 · 2 weeks
Assess
A baseline of each process before anything changes.
V-01 · Baseline approved
Tracking KPIs is the practice with the biggest effect on EBIT, of 12 tested. Fewer than 1 in 5 do it.
McKinsey, 2024. 1,491 respondents, generative AI only.02 · 6 weeks
Pilot
One process, one team, a measured window. We run the pilot.
V-02 · Pilot measured
Only 31% have criteria for a pilot not to reach production. Only 17%, for ending what gives no value.
Gartner, 2026. 2,002 CIOs and technology executives.03 · 1 quarter
Scale
The pattern repeats in other areas, led by us.
V-03 · Pattern replicated
Among those capturing the most value, nearly 3/4 redesigned their workflows. Among the rest, 1/4.
McKinsey, 2026. 1,719 respondents, self-reported.04 · Ongoing
Embed
It becomes routine. We keep operating and reporting the impact.
V-04 · Routine stable
Those who strongly agree their manager supports it: 78% use AI frequently. The rest: 44%.
Gallup, February 2026. Where AI is available.Signal · measure
The number the CFO accepts.
Hours recoveredvs. week 0
Cycle timevs. week 0
Cost per outcomevs. week 0
You see the same dashboard we do, against the same baseline. We do not report activity as outcome.
Charter · govern
Your legal team just approves.
We draft the governance layer. The default answer stops being “wait”.
01Usage policy
02Risk matrix
03Vendor clauses
04Onboarding guide
Nebuloz
We do it.
We run the four tracks
We run the pilots
We measure against week 0
We draft policy and risk
We operate Cosmos
We report every result
Specialists in the design · product in the delivery
Your team
Your team approves and uses.
Approves each gate
Uses the new process
Follows the impact dashboard
No system to maintain
Method · what we leave out
Three numbers we don’t use.
They open a lot of AI presentations. They don’t survive a CFO’s question.
- 95% of pilots fail
It comes from 52 interviews, 153 conference responses and 300 public initiatives. The authors themselves mark it preliminary.
Project NANDA, MIT Media Lab, 2025. - 80% of AI projects fail
RAND cites it as other people’s estimate. Its own study is 65 interviews about causes.
RAND, 2024. - 70% of change programs fail
A review looked for the evidence behind that rate and found no valid empirical basis.
Hughes, in the Journal of Change Management, 2011.