Skip to content
22decisions home

Methodology

Our methodology

Survey data, calibrated twin panels, and real-world signals, validated before anything is published.

The panel, live

Where our twins sit, and who they are

Twins
3,544
Panels
5
Roles
14
Industries
5
Countries
22

Hover over or tap a circle to see the location, its total twins, and the breakdown by panel and industry.

These are modelled twins, not real individuals. Counts are aggregated by country, industry, and role; no company or personal data is shown. Last updated .

  • CIO Twin Panel

    200twins

    CIO twins across four industries, calibrated to more than 7,000 published survey responses and tested blind against independent survey results.

  • CDO Twin Panel

    200twins

    Chief data officer twins across four industries, focused on data, governance, and AI.

  • AI Decision-Makers Panel

    1,373twins

    Executive twins in 13 roles across 120 modelled organisations, mapping who decides, influences, and can veto AI decisions.

  • India AI Decision-Makers Panel

    1,346twins

    Executive twins in 13 roles across 120 modelled Indian organisations in banking and financial services, automotive and manufacturing, retail, healthcare and telecommunications, calibrated to published Indian survey data.

  • CEO Panel

    425twins

    CEO twins in India (125) and the US, UK and Europe (300), calibrated to published figures from the KPMG 2026 India CEO Outlook on AI priorities, payback expectations, barriers and risks.

By role

  • CEO665
  • CIO440
  • CFO240
  • COO240
  • CISO240
  • General Counsel and Privacy240
  • CHRO240
  • Business Unit Head240
  • Chief Procurement Officer240
  • CDO200
  • Chief Risk and Compliance Officer162
  • Chief Data and Analytics Officer155
  • CTO141
  • Chief AI Officer101

By industry

  • Banking and Financial Services936
  • Automotive and Manufacturing857
  • Retail and Consumer Goods814
  • Healthcare and Life Sciences741
  • Telecommunications196

How our research works

From inputs to published research

  1. Stage 1: Inputs

    • Published survey data: 7,000+ responses from leading studies
    • Our pulse surveys
    • Alternative and open data: job postings and open survey datasets
    • Expert input
  2. Stage 2: Model

    • 22decisions twin panels: CIO, CDO, and AI decision-maker twins, calibrated to the inputs
  3. Stage 3: Analysis

    • Sources combined
    • Every figure labelled Measured, Calibrated model, or Cited
  4. Stage 4: Validation

    • Blind tests against independent surveys
    • Prediction checks: frozen forecasts against pulse results
    • Editorial review
  5. Validation back to Model: Results recalibrate the panels
  6. Stage 5: Outputs

    • Research briefs
    • Industry reports
    • Provocations
    • Free tools
Research moves from inputs to the model, analysis, validation, and outputs. Validation results loop back to the model: results recalibrate the panels.

Evidence labels

Every figure carries its source

Each number we publish is labelled with where it comes from.

  • Measured

    Data we collect and analyse ourselves, such as survey responses, benchmark answers and job postings.

  • Modelled

    Output from our modelled decision-maker panels, calibrated against published survey data. Always labelled, never presented as a survey of real people.

  • Cited

    Published third-party research, named alongside the figure so you can check it.

  • Practitioner

    What we hear in interviews and conversations with people making and running AI decisions.

Our panels: their limits

Where we are confident, and where we are not

ConfidenceWhat the panels can answer
High confidenceValidated or measuredCIO priorities, how tech teams use AI, AI hiring patterns.
GoodCalibratedWho decides and who can veto each AI decision; how common chief AI officers and governance committees are.
IndicativeDirectionalHow a given role is likely to react to a proposal; industry differences.
Not yet reliableRanges or new data firstNew behaviours no survey has measured, and spending amounts. We show ranges, or gather new data, before publishing on these.

Validation

How we check our work

  • Blind tests

    We freeze our panel's predictions and compare them with independent survey results, publishing what we find.

  • Prediction checks

    We forecast the results of our own pulse surveys before they close.

  • Editorial review

    Every report is checked before release.

Publishing stages

From scheduled to published

These are the status badges you see on each research brief.

  1. Scheduled

    Scheduled

    The question is set and the brief is in the research calendar.

  2. In modelling

    In modelling

    The twin panels are being calibrated and run on the question.

  3. In analysis

    In analysis

    Sources are being combined and every figure labelled.

  4. In editorial review

    Editorial review

    Findings, labels, and claims are checked before release.

  5. Published

    Published

    The brief is live, with its evidence labels and sources.

Sources

How we choose and label sources

  • Named and dated

    Every cited figure names its source and year, so you can check it yourself.

  • Primary over secondary

    We go to the original study, filing or dataset rather than a summary of it, and prefer large, recent and transparent samples.

  • Vendor-sponsored sources

    Research paid for by a technology vendor is labelled as vendor-sponsored, and we say who paid for it.

  • Lab-sponsored sources

    Figures published by AI labs about their own models are labelled as lab-sponsored and, where we can, checked against independent results.

Independence and disclosure

What we disclose, and what we never do

  • We use AI assistance in our research and writing. We use Claude, made by Anthropic, and we disclose it.
  • Any sponsored work is clearly labelled as sponsored.
  • Tool results, including the AI Decision Benchmark, are never influenced by sponsors.
  • Individual data is never sold or shared without explicit consent.

Transparency

What our twins are, and are not

Our twins are modelled, not real individuals, and contain no personal data. Our research is prepared with the help of AI tools. Clients can request a detailed methodology summary under NDA.

Request the methodology summary

Get our research in your inbox

One considered briefing a month on how organisations decide on, buy and adopt AI. Unsubscribe at any time.