Methodology
How the number is produced, and what keeps it honest.
Four models do the work: a digital maturity language everyone can hold in their head, a capability model with a defined readiness threshold, a wave roadmap prioritised by dependency, and a value model with rules it cannot break.
Three vocabularies, three jobs, never mixed. Maturity (Aware→Transform) answers where you are. Waves (W1–W4) answer in what order. Autonomy levels (L1–L3) answer how much an agent may do in a given workflow.
Maturity model
Five stages. One sentence to place you.
Executives do not need a scoring rubric; they need to know where they stand and what the next stage requires.
1. Aware
AI is discussed, not governed. Isolated experiments, no shared value case.
2. Investigate
Structured pilots begin. Data, skills and governance gaps become visible.
3. Activate
First workflows run in production, with human oversight and measured value.
4. Accelerate
Capability spreads across functions. Value compounds and the operating model becomes explicit.
5. Transform
An AI-native operating model. Agents work alongside teams under governed, reversible autonomy.
One language
The same five stages across the report, the dashboard and the conversation. No parallel vocabulary to reconcile.
Placed by evidence
Your maturity level comes out of an analysis of the whole evidence base, its sources and the assessment as a whole, not out of a conversation. The report states it in a single sentence on page one.
The engine stays behind
There is a scoring engine underneath. It is deliberately not shown to you, because the score you get is not a decision. What counts is taking the first step towards the next stage.
Capability model
Eight capabilities, read against one threshold.
Five business capabilities and three technology capabilities. What matters is not the exact position but which side of the readiness threshold it falls on.
Business capabilities
Whether the organisation can absorb and govern AI.
- Executive LeadershipAt readiness
- AI GovernanceBelow threshold
- People & CultureBelow threshold
- Business IntegrationBelow threshold
- AI Operating ModelBelow threshold
Technology capabilities
Whether the technical foundation can carry it.
- Data FoundationAt readiness
- Technology PlatformAt readiness
- AI FactoryBelow threshold
The enterprise reference threshold. Activate is the standard enterprise reference point: owned, funded and in production in places. It is not a universal permission to deploy every use case. Each workflow also has a risk- and dependency-specific threshold. A low-risk assistant may proceed with bounded controls while a regulated decision requires stronger data, governance, evaluation and oversight. A high average never cancels a failed mandatory gate.
Roadmap
Four waves, sequenced by dependency.
Waves are not ambition levels. The next wave starts once the capability it rests on clears the readiness threshold, which is why roadmaps that skip ahead stall in the next wave.
- W1
Productivity
Individual and team productivity. Assistants inside existing work, measured against a real before/after rather than enthusiasm.
Depends on: Access, basic guardrails and a way to measure.
Proof it worked: Habit formation: the same people still using it in month three.
- W2
Process Automation
Process automation. End-to-end workflows redesigned around human–agent collaboration, with autonomy calibrated per step.
Depends on: Data foundation and governance clearing the threshold.
Proof it worked: Cycle time and exception rate move, and stay moved.
- W3
Decision Intelligence
Decision intelligence. AI in the decisions themselves, such as forecasting, pricing and allocation, under explicit model risk management.
Depends on: Trustworthy data, evaluations and model risk management in place.
Proof it worked: Decision quality improves against a pre-agreed baseline.
- W4
AI-Native
AI-native operating model. New services and cadences that were not staffable before, with agents operating inside governed bounds.
Depends on: An operating model, a measurement pipeline and agent-scale security.
Proof it worked: Capability that competitors cannot copy by buying the same tools.
Value model
Added value is calculated bottom-up.
The value answer is built from function-level ranges and nothing else. There is exactly one value model in the report, and every figure in every section resolves back to it.
- 1
Value per function
Each function gets a low and a high, derived from its own workload and team size, not from a market average.
- 2
Annual run-rate
The sum of the function ranges once capability is in place. The headline is the sum; it is never chosen first.
- 3
Three-year cumulative value
Run-rate multiplied by a ramp that assumes value arrives gradually, because it does.
- 4
Payback
Cumulative value plotted against the investment envelope, with the break-even point stated rather than implied.
Five rules the model cannot break
One source of truth
If the sum of the functions does not equal the headline, the report fails its own test and is not delivered. This is enforced in code, not by review.
Visible assumptions
Cost ranges, team sizes and their sources sit next to the number, not in an appendix nobody opens.
Cost side only in the core
Revenue upside is deliberately excluded from the core case and reported separately, because it is the easiest number to inflate.
A qualitative account
Capability gains that cannot be counted honestly are described in words. They never get a € or a %, however tempting.
Evidence before precision
Every financial input is labelled Hypothesis, Supported estimate or Validated baseline. A precise payback date is not shown when the inputs do not support that precision.
First, learn the value you can create. Then build the operating model that delivers it.
Insight gives your executive team a reviewed answer within two business days once the required answers are complete.