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.