AI investment decisions are increasingly being made on the basis of vendor demonstrations, industry reports, and executive intuition. None of these answer the question that actually matters: where in our specific organisation will AI generate measurable productivity gains, and in what sequence?

The role-content problem

Most AI readiness assessments analyse the organisation at the function or process level. This is too coarse. The same function contains roles with radically different AI-augmentation potential. A bank’s retail operations function may contain teller roles scoring 70 percent on automation potential alongside branch manager roles scoring 83 percent on AI augmentability. Treating them as one number hides the opportunity.

Genuine AI readiness analysis must start with the work itself. What is this role actually doing, hour by hour? Which of those activities are rules-governed and structured, and which require human judgement in the loop? This is role-content-based analysis, and it produces evidence that is both granular and defensible.

Three lenses, not one

AI augmentability is one of three lenses we apply to every role. The others are digitisation readiness and automation potential. A role can score high on automation potential but low on AI augmentability (routine processing work). Another can score high on AI augmentability but low on automation (advisory work where AI co-pilots add significant leverage but the work itself cannot be automated).

“Digitisation investment unlocks the structured data that automation requires. Automation frees capacity that AI can redirect toward higher-value judgement work.”

The sequence matters. Treating the three as independent programmes, as many organisations do, produces disconnected initiatives that rarely compound. Treating them as a sequence produces transformation programmes that build on each other.

What good evidence looks like

A board-ready AI readiness assessment should meet three tests. First, every score should be traceable to its source data. No black boxes. Second, the methodology should be deterministic. The same input should yield the same classification, not a different answer every time. Third, the analysis should be bounded by what the documented work actually shows, not augmented by assumption.

When those three tests are met, the question of where to invest in AI stops being a belief and starts being a decision.