APPROACH
How we get to the truth
Honesty isn't a personality trait here. It's a process with rules - and the rules are more interesting than the slogans.
01 / THE METHOD
Four steps, each with a rule
The real system, mapped
Every use case, ranked
Sequenced by unlock conditions
Working software, behind gates
01 / AUDIT
Audit the real system
Roadmaps built on interviews alone are fiction. We read the formulas, trace the data lineage, and find where what the system does diverges from what everyone thinks it does. The dark corners are usually where the valuable insights are.
02 / SCORE
Score every use case
Value, complexity, readiness - weighted, compared, ranked. A use case doesn't get built because someone senior likes it. It gets built because the numbers survive scrutiny.
03 / ROADMAP
Roadmap with unlock conditions
Machine learning needs labels, volume and a measurement loop; most organisations don't have them yet, and pretending otherwise is how pilots die. Our roadmaps sequence work behind explicit preconditions - when the condition is met, the lane unlocks.
04 / BUILD
Build behind gates
Deterministic before probabilistic. Measurement before models. Every phase has an exit, and "stop" is always a legitimate verdict - including about our own work.
02 / WHEN WE SAY NO NO-GO
When we tell you not to use AI
These are the tests a use case has to fail for us to write NO-GO. We publish them because you can check our verdicts against them - and because most of the AI disappointment we see was predictable in advance.
- N1
The process isn't stable enough to automate — automating chaos just makes faster chaos.
- N2
There's no measurement baseline, so nobody will ever know if it worked.
- N3
The volume doesn't justify the build — a checklist or a template beats a model.
- N4
The data can't support it yet — and the honest move is fixing the data first.
- N5
The failure cost lands on a customer or a regulator, and the accuracy ceiling isn't high enough.
- N6
A person with better incentives beats the model — the problem is organisational, not technical.
03 / PRINCIPLES
Our three rules that outrank the roadmap
Graduate, don't replace
We improve the tools your team built before proposing their replacement.
People-risk counts
Incentives and change readiness are blocking risks, same as technical ones.
Small data, honest methods
When N is small we say so and pre-commit fallbacks.
04 / IN WRITING
When our approach isn't likely to beat your current process, we put the probability of failure in the report.