A comparative evidence base, not a league table of inevitability

GenAI adoption is fast, broad — and still uneven.

The Atlas separates experimentation from approved production, compares ten sectors, and makes every forecast inspectable. Scores are transparent judgements built from evidence; they are not official statistics.

Forecast disclaimer: probabilities are rounded, reasoned judgements. They are not facts or investment advice and should change as evidence changes.

Current state · July 2026

Adoption index

Seven weighted components; select a sector to inspect the evidence.

Accessible table: sector rankings

Compare

Up to three sectors

Component scores use the weighting described in Methodology.

Observed milestones and forecast markers

Master timeline

Zoom the time window, filter the lanes and open any item for its evidence. Forecast years sit in a shaded region and never use observed-event marks.

Sectors
ObservedForecastForecast regionResistance / reversal
Accessible table: filtered timeline

Ten distinct adoption contexts

Sector detail

Choose a sector. The charity page includes additional organisation-size and safeguarding analysis.

Baseline plus alternatives

Forecasts & scenarios

The baseline is preserved when scenarios change. Scenarios are coherent possibilities, not predictions.

Trajectory comparison

Projected adoption index

Scenario adjustments are bounded to 0–100 and shown as dotted lines.

Accessible table: projected adoption trajectories

Cross-sector questions

Forecast table

Approved production means organisation-sanctioned use in at least one live workflow.

What would change our minds?

A revision ledger, not a silent overwrite

Evidence & forecast updates

New signals are logged with the previous and revised probability. A local proposal form demonstrates the update workflow.

Forecast revision history

Locally proposed signals

Definitions, weights and evidence register

Methodology & sources

This atlas is designed to be challenged and updated. Its strongest evidence measures specific settings; its weakest areas are stated openly.

Scope

GenAI means systems that generate text, images, audio, video, code or structured outputs from learned patterns. The Atlas excludes ordinary analytics unless a cited survey combines AI types; those records are explicitly labelled.

Adoption stages

Experimentation
Individuals or teams testing a tool without an operational commitment.
Piloting
Time-bounded, approved evaluation in a realistic setting.
Workflow integration
Repeated use inside a defined process, still limited in scope.
Production deployment
Approved live use with accountable owners and controls.
Scaled adoption
Production use across multiple teams, sites or workflows.
Institutionalisation
Standards, procurement, training, assurance or labour rules make use durable.
Resistance or reversal
A ban, pause, failure, retrenchment or binding constraint.

Adoption index

The 0–100 index is an editorial synthesis, not an objectively measured statistic. It weights: breadth 20%, workflow depth 20%, investment 15%, workforce capability 10%, governance readiness 10%, sector tools 10%, production evidence 15%.

Scores triangulate surveys, documented deployments, regulation and counter-evidence. A high score can coexist with low trust or high constraint. Sector scores are rounded to avoid false precision.

Forecast process

  1. Define a measurable threshold.
  2. Use comparable enterprise-software diffusion as an outside view.
  3. Decompose drivers and constraints.
  4. Combine sector evidence with the outside view.
  5. Round probabilities and provide ranges.
  6. Record revisions rather than rewriting history.

Evidence strength

High: primary records, regulator data, disclosed-method research or verified operational disclosure. Medium: credible surveys with sampling limits, reputable secondary reporting, or company claims without independent evaluation. Low: commentary, anecdote or ambiguous evidence; rarely used for scoring.

Known gaps

  • Most surveys over-represent large, digitally mature organisations and high-income countries.
  • “Using AI” often conflates employee use, pilots and production.
  • Finance and health sources frequently combine GenAI with older AI.
  • Small charities, local government and lower-income countries are under-measured.
  • Self-reported productivity is not equivalent to causal, quality-adjusted impact.

Update policy

Material signals are added with a source, classification and editor note. Forecast changes record the old value, new value and reason. A proposed signal does not enter the evidence base until reviewed.

Searchable register

Sources