Home/How I make claims
How I make claims
A standard I hold myself to, on stage and in writing. It exists because I ask clients to apply the same test to what vendors tell them.
Most of what goes wrong in enterprise AI decisions is not a technical error. It is a claim accepted at a strength it was never entitled to. An observed number becomes an expected saving. One team's result becomes an organisational fact. A pilot's performance becomes a production guarantee.
So every claim I make carries a label saying where it came from and how far it can be pushed. These are the labels.
The labels
| Label | What it means | How far you can push it |
|---|---|---|
| Measured | Counted directly, with a defined method and period. | Reliable for the thing measured, in the conditions it was measured under. Nothing else. |
| Reported outcome | A result reported by the team that produced it, not independently audited. | Evidence that something happened. Not evidence of the size of the effect. |
| Observed activity | Usage: requests, volumes, how often a thing is called. | Shows adoption. Says nothing about value. These are routinely confused. |
| Expected | A projection. The benefit has not been realised yet. | Useful for planning. Never quote it as a result. |
| My own account | First-person recollection of work I did. | Honest, partial, and unverifiable by you. Weigh it accordingly. |
| Illustrative | A worked example with invented figures, used to show a method. | Only ever a demonstration of the shape of a calculation. |
Observed activity and expected benefit are different claims. Most decks present them as the same one.
Why there are no client numbers on this site
You will not find a percentage, a volume, or a saving attributed to a client anywhere here. That is a deliberate choice, and it is worth explaining because the absence looks like a weakness.
Any figure I could publish would be self-reported, not independently audited, and attributed to a client who has not agreed to appear in my marketing. That is precisely the class of claim I would tell a buyer to discount when a vendor puts it in front of them. Publishing it here and asking you to treat it differently would be incoherent.
There is a second reason, less principled and more practical. Numbers from a regulated environment carry context that does not survive being extracted: what was counted, over what period, against what baseline, and who decided the baseline. Stripped of that, a number is decoration.
What I offer instead
- Published method. The frameworks are complete and free, with their failure modes included. You can apply them to your own situation in fifteen minutes and judge whether they hold, without believing anything I say about myself.
- Work you can check. The Living AI Atlas is free to read with no account, maintained continuously rather than published once, and open to contribution: readers can comment to ask for a correction or propose what is missing. Judge the standard of the work without taking anything here on trust.
- Named delivery. Where I have spoken, I say where and when, and you can ask them.
- Where it broke. When I build something hyped, I report where it failed, including when that is inconvenient for me.
What this costs me
A page of statistics would convert better in the short term. Buyers do respond to numbers, including numbers they cannot verify, and I am aware that a competitor's unlabelled chart will look stronger than my labelled sentence.
I would rather be checkable. If you are choosing someone to help you decide what an AI system may do in a regulated process, the way they handle their own evidence is a relevant sample of how they will handle yours.
Holding me to it
If you find a claim on this site that is unlabelled and should not be, or labelled at a strength it does not deserve, tell me and I will correct it in the open with the date on it. That is the same standard the Atlas runs on.
Revisions
- First published. The label vocabulary is the one used in the Amsterdam Business School lecture of 16 September 2026.
This page describes my own practice. It is a standard I set for myself, not a certification, and nobody audits my compliance with it but you.