The knowledge that runs your organisation.And the reason it never gets written down.
μῆτις
Met · is · on. The word sits in the name unbroken, which is the point — it carries the argument instead of needing an explanation.
Greek distinguished three ways of knowing. We are interested in the one that does not survive being written down.
- Sophia
- Abstract wisdom.
- Techne
- Formal, writable, transferable craft knowledge. The work instruction. The ERP configuration. The process documentation.
- Mētis
- The know-how people build up by doing the work: how to handle this case, this exception, this awkward customer. Real knowledge, and it survives almost nowhere in writing.
In Seeing Like a State (1998), James C. Scott uses metis for the lived knowledge of practitioners that will not be codified and that the official system cannot see. Large-scale reform fails, he argues, because it replaces metis with techne — perfecting the plan while wrecking the thing it described.
That is this company, already described in a book that boards take seriously.
VisionAn organisation should not have to guess at itself.
Remove the constraint that produces sampling and what appears is not cheaper interviews but a different instrument: one that asks the whole organisation, and in which every conversation is sharper than the last.
MissionFrom how you really work, to processes with AI designed in.
The map is the product. The redesign is what it tells you to do. The build is someone else’s to deliver.
The distinction everything rests onAI in it, not AI beside it.
AI alongside
You put a copilot on step four of a process designed for people with a mouse.
The shape does not change. You win five per cent on one step of a twelve-step process that should not have twelve steps — which is where most AI budget goes, and why it never shows in the numbers.
AI in
You ask what this process would be if you designed it today.
Knowing what a model can do, the answer removes steps more often than it accelerates them. And the redesign has to be about the variants that really exist — the spreadsheet, the group chat, the phone call — not the six steps in the instruction.
You cannot redesign a process you cannot see.
PrinciplesNine rules. Each one rules something out.
They are not decorative. Ruling something out is the only test of whether a principle is real.
01
Ask, don't measure
We do not watch over shoulders, read screens, or install anything on anyone's machine. That is a limitation, and it is the reason this passes a works council.
02
Everyone, not a sample
The sample is precisely what is wrong with the old method. Speak only to the people the client selects and you hear what the client already thought.
03
Role, never person
A statement attaches to a role and a process, never to a name. This is a property of the architecture, not a promise in a contract.
04
The difference is the point
You already have a description of your processes. What you do not have is the distance between it and what actually happens, and that gap is the only part a board cannot produce on its own.
05
AI in, not alongside
If a proposal does not change the shape of the process, it is not a redesign. It is a purchase.
06
Deleting counts
Not every process deserves an AI version. Some deserve to disappear, and that is as valid an outcome as building.
07
Participants get something back
If a conversation vanishes into a board report the participant never sees, participation falls and the method is worthless. That is arithmetic, not goodwill.
08
We say what we could not establish
A number that shows its own margin of error is worth more than a number that merely looks precise.
09
Advice that can be carried out
Two in three process programmes disappoint, rarely because the analysis was wrong. A recommendation that lands nowhere is a cost.
Why nowFour things became true at the same time.
The interview became automatable without becoming worthless
A model can follow up, recognise a contradiction, adapt to who it is talking to, and carry what four hundred earlier conversations established into the next one. Five years ago this was a questionnaire.
Public sources describe organisations better than ever
Job adverts alone name roles, responsibilities, the systems in use, and what is changing. Annual reports, procurement policy and tenders do the rest.
Building bespoke became cheaper than buying off the shelf
Ten years ago custom was the expensive answer. Now it is often the cheap one, precisely because it follows the real process instead of bending the organisation towards a product.
The alternative stayed expensive and slow
The price of a consultant-hour has not fallen. Organisations of two hundred to five thousand people therefore do not buy organisational research — while the need is there, and the AI budget is being spent blind.