How the research is built.
The credibility of this programme rests on how the data is gathered, and on the parts we admit it does not yet cover. Both are set out below in full.
The count updates as each assessment completes. Shares are recomputed from the live Integration Index, last on 25 September 2026. See the live index.
What this is
The Big AI Secret is an ongoing research programme that reveals how businesses actually use AI. It is editorial, and it is transparent about how it works: the method is published in full, every figure is sourced, quotes are anonymised before they appear, and nothing is sold during a research conversation. The programme is funded and run by Maybe*. Maybe* is the AI execution layer where real business work gets done, with clear rules, full visibility, and your people in control. That is stated plainly rather than dressed up as independence. It is not a set of dated Guides. It is a continuously updated body of research.
The evidence base
The framework comes from over 1,500 in-depth interviews with business leaders, across sectors including agency, commercial and developers, professional services, retail, financial services, manufacturing, healthcare and technology, published as Edition 1, 2025/26. Every business measured on the Integration Index since it opened is counted separately, because an interview scored against evidence and a two-minute questionnaire are not the same measurement. Findings are cross-referenced against published research from other organisations, including Stanford, McKinsey and BCG, where comparable figures exist.
How we read the data
Two models organise the findings. Six dimensions score how far AI is built into how a business works: Breadth, Autonomy, Flow, Measurement, Governance and Access. Each is scored 0 to 3, by an AI Analyst from the interview or self-reported in the questionnaire. Together they produce the Integration Index, expressed out of 100. Four stages then place a business on the path from experimenting to embedding: Transformer, Integrator, Implementer and Explorer. Most businesses, the research finds, are still in the first two.
How the data is gathered
Every finding originates in a recorded interview or in a structured questionnaire completed by a named leader within a firm. Interviews are booked, recorded and transcribed, and the transcript is matched back to the business automatically before any scoring begins. Interviews are anonymised by role and sector before publication, interviewees see the quotes attributed to them before anything is published, and nothing is printed that has not been checked. Anyone measured keeps a permanent record of their own result, with the scores, the comparisons and the practical next steps written for their business, which they can share with their team.
How scoring is kept honest
Scores from an interview are produced against the transcript, dimension by dimension, with the evidence for each score recorded next to it. A share of interviews is scored blind by two Analysts at all times, and their scores are compared before either is published. That share is not a fixed run at the start of a sector: it is recalculated from how often the two have agreed on each dimension, rising where they disagree and easing back where they do not, and it never falls to nothing. Where they disagree, the research lead adjudicates and the thresholds are rewritten rather than the disagreement being smoothed over. Every published figure carries its source line and its sample size, and the whole scoring model, including the wording of every question and every threshold, is versioned so that any result can be traced to the exact version it was scored under.
The honest limitation
Much of the picture today rests on self-reported estimates. Leaders tell us what they run and roughly what it costs them. We normalise those answers and benchmark them against organisation size and maturity, but it remains estimation, and we treat it as such. Every finding that depends on estimation is labelled that way, next to the figure. Nor can we yet say how far a two-minute self-report differs from an Analyst scoring the same business against evidence. That calibration needs both sides of the measurement, and it grows with every interview added.
Where this goes next
The research is moving from estimated to measured. As businesses choose to share real records of the work their AI actually completes, it shifts from asking what AI did to recording it. The figures on this page move with that: the count rises as each assessment completes, and the shares are recomputed from the live Integration Index rather than being retyped. Taking part is how that picture sharpens.
A working citation, with the figures as they stand today:
Full citation page with Harvard, APA and BibTeX, or watch the figures move on the live dashboard.
Every interview is conducted by an AI Analyst.
AI Analysts are graduates, recruited for curiosity and manner rather than sector experience, and trained under the method set out on this page. They spend their placement doing one thing: sitting down with business leaders and finding out what AI actually does inside their organisation, as opposed to what the press release says.
They also work with AI directly rather than only asking other people about it. Each Analyst spends their first week building and running AI Agents, and then uses them daily to turn interview recordings into structured analysis, correcting and approving every output before it goes anywhere. The accountability for accuracy stays with the Analyst, not the model.
Before an Analyst interviews anyone
- They read the full interview method and sign a confidentiality agreement.
- They complete data protection training.
- They score recorded interviews independently and compare their scoring against the research lead.
- They run practice interviews, recorded and reviewed.
- They run two real interviews with the research lead listening in silence before they run any alone.
A share of interviews is scored blind by two Analysts who compare afterwards, and that share rises for any dimension where they keep disagreeing. When they disagree, the research lead adjudicates, and the scoring thresholds are rewritten rather than smoothed over. We credit our AI Analysts by name in the Guides they contribute to.
Our AI Analysts run the interviews that produce this research.
AI Analysts run the interviews that produce this research as part of our graduate internship. Over a placement that means fifty or more conversations with directors, partners and chief executives about how AI really works inside their organisations, what completes, what quietly does not, and what nobody says publicly.
You will also build and run AI Agents yourself, and use them daily in a working research workflow where being wrong has consequences.
We recruit for curiosity and manner rather than for AI knowledge or sector experience. The single best predictor of a good interview is whether someone can leave a silence alone.
You leave with your interview count confirmed in writing, a named credit in published research, a reference from someone who will take the call, and a first-hand view of AI adoption that very few people at any stage of their career have.
The Big AI Secret is an ongoing research programme, funded and run by Maybe*. Guides publish continuously. Interviews are conducted by AI Analysts, who are credited by name in the Guides they contribute to.