For heads of AI and AI program leads
Why AI pilots fail to scale: six conditions
AI pilots usually succeed because they are small: a narrow context, curated data, and owners close to the team that built them. The book argues they stall at scale not because the model gets worse but because the enterprise underneath was never clear enough to carry it. It sets out six conditions an initiative should meet before it is scaled.
Kindle edition · 99 pages · a 90-minute readRead a free sample on Amazon
From the book
What the book shows
- 01
The model didn’t get worse
The book describes the pilot that dazzles and never scales: a small team, curated data, an impressive demonstration — then the attempt to extend it across functions, regions, systems and definitions, where it stalls.
“The model did not get worse. The enterprise underneath it was never clear enough to carry it.”
Chapter 6Why Transformations Often Fail Before They Start - 02
Context is what is missing
Without a shared map of where work happens and who owns it, a model cannot tell a legitimate local difference from an error, and nobody can explain or stand behind its recommendations.
“Models do not fail because data is unstructured; they fail because operational context is missing.”
Chapter 8The Operating Tree - 03
Same model, clearer enterprise
In the book’s (anonymized) merger case study, a cross-brand recommendation engine had been paused. Once the two companies agreed shared definitions of customer activity, it was redeployed — and the book reports a 14 percent increase in digital cross-selling revenue within its first 90 days, in that one company.
“The model had not improved. The enterprise feeding it finally had.”
Chapter 20Case Study: When Two Strong Companies Could Not Become One
The six conditions to meet before you scale
Chapter 17 sets out when an AI initiative is ready to leave the pilot. Miss any one, the book says, and what is running in production is still an experiment — with a new kind of organizational risk attached.
“An AI initiative is ready to scale when six things are true.”
- 1The business context is clear.
- 2The data is understood.
- 3The meaning of the output is interpretable.
- 4The decision it supports is known.
- 5The owner of the outcome is accountable.
- 6And the boundaries of acceptable use are defined.
The six conditions as worded in Chapter 17, Scaling AI on a Foundation of Clarity.
Questions to ask
Questions to ask of every initiative before it scales:
- 1Which business decision does this initiative support, and who owns the outcome?
- 2Did the pilot work because its context was narrow — and does that context exist at scale?
- 3Do the terms the model relies on mean the same thing in every market it will reach?
- 4Are the boundaries of acceptable use written down, and owned by someone?
- 5When the model recommends something, can a named person explain it, challenge it and stand behind it?
- 6Which initiatives are still experiments — and are we honest about calling them that?
Where to start in the book
Chapter 17
Scaling AI on a Foundation of Clarity — scaling AI on a foundation of clarity, and the six conditions.
Chapter 6
Why Transformations Often Fail Before They Start — the pilot that dazzles and never scales.
Chapter 8
The Operating Tree — the Operating Tree: the context models are missing.
Kindle edition · 99 pages · a 90-minute readRead a free sample on Amazon
Also see
- Data leadersData readiness for AI isn’t enough
- CIOs & CTOsWhy IT can’t fix what the business hasn’t decided
- Risk & complianceAI governance without bureaucracy
All roles: who the book is for →
About this page: our reading of Your data isn’t broken. Your operating model is. by Adam Ayres and Behrang Karimi, for heads of AI and AI program leads. The book itself is written for the leadership teams of complex enterprises. Quotes are word for word from the book; its case studies describe anonymized companies, and their results are not benchmarks.