How this writing is made

Everything published here carries my name, and I hold editorial responsibility for all of it. That means every claim, every figure and every argument is one I have examined and would defend in a room. Where a piece is wrong, it is wrong because I got it wrong.

Claims that rest on evidence are traced to the primary source rather than to reporting of it. Where a figure comes from a research paper, the paper is read; where it comes from a survey, the methodology is checked and stated. This matters more than it used to. A number that has passed through two intermediaries acquires confidence it has not earned, and a good deal of published commentary now rests on statistics whose origin nobody has followed.

Four rules apply.

Figures that cannot be verified are not used, however useful they would be to the argument. Several have been dropped from work in progress for that reason.

Working papers are labelled as working papers, because pre-publication research has not been through peer review and a reader is entitled to weigh it accordingly.

Where a survey has a commercial interest in its own findings, that is stated in the text rather than left for the reader to discover. Numbers of that kind are used for the pattern they describe, not for their precision.

Where a claim is contested, the counter-argument is given rather than omitted. Several pieces here concede the strongest objection to their own position, because an argument that cannot survive its best rebuttal is not worth publishing.

Much of what is written here rests on twenty-five years of doing the work rather than on published evidence, and that distinction is marked where it matters. A position drawn from experience is offered as a considered position, not presented as a finding. Where a briefing argues for a practice that is not yet common enough to survey, it says so.

Anonymised examples are drawn from real engagements. They are described at a level that preserves the confidence of the organisations involved, which sometimes means a mechanism is given without the industry or the geography that would identify it.

I use AI tools in this work, as I do in most of my work, and it would be strange to write about enterprise AI while pretending otherwise. They are used for research, for structuring, for drafting and for challenge.

What they do not do is decide anything. The argument, the position taken, the evidence admitted and the words that survive to publication are mine. A tool that proposes a claim I cannot source does not get to keep it, and several of the corrections that shaped these pieces came from checking machine-suggested figures against the underlying papers and finding they did not hold.

It is the standard I would hold anyone working for me to, and the reason is the same in both cases: the tool has no stake in being wrong. The instrument is not the interesting question. Whether the output is right, and whether someone is accountable for it, is.

If something here is wrong, I want to know, and it will be corrected. Substantive corrections to a published document are made in a new version rather than silently; minor corrections to furniture and typography are made in place.

Anything factual you would like to challenge, or any source you think has been read too generously, is welcome at the contact address on this site.