Methodology

Published in full, so the reasoning can be checked

A framework that produces investment recommendations should be open to inspection. The dimensions, the gate model, the comparison set and the confidence calculation are all documented here.

Eleven scoring dimensions

Each dimension is scored from 0 to 100 with a written rationale and its own confidence value. There is no single overall score, because collapsing eleven different judgements into one number hides exactly the disagreement a reader needs to see.

The eleven dimensions. Governance exposure is the one dimension where a higher score means greater concern.
Dimension	What it measures	A low score means
Problem clarity	Whether the problem, its scope and its success measure are defined.	The work has not yet been specified well enough to invest against.
AI necessity	Whether the outcome genuinely requires AI rather than simpler means.	A conventional or existing capability would likely do this.
Data readiness	Availability, quality, coverage, lineage and lawful basis for use.	The data foundation would have to be addressed before any build.
Technical feasibility	Whether the capability can be delivered in this environment.	Integration, latency or infrastructure constraints are not yet solved.
Organisational readiness	Skills, ownership, change capacity and operating model fit.	There is no sustainable home for the capability once it is built.
Regulatory exposure	How far the use case sits inside regulated territory.	Obligations are unclear or unmet for the intended deployment.
Governance exposure	Oversight, accountability and control burden. Higher means more concern.	Read inversely: a low value here indicates lighter governance exposure.
Value credibility	Whether the claimed benefit is evidenced and attributable.	The benefit rests on assumptions nobody has substantiated.
Delivery risk	Complexity, dependency and the consequence of delivery failure.	Delivery carries risks that are not yet mitigated.
Operational sustainability	Whether the capability can be run, monitored and maintained.	There is no credible plan to operate this after go-live.
Strategic alignment	Fit with stated objectives and existing commitments.	The initiative does not serve a priority the organisation has set.

Gates are evaluated separately from scores

Scoring answers “how good is this case?”. Gates answer “is this permissible at all?”. Mixing them would let a strong commercial case dilute a legal obstacle, which is precisely how organisations end up defending decisions they should never have made.

Blocking

Stops the decision

No lawful basis for the processing, a prohibited practice, safety-critical autonomy without meaningful human control, or a failure consequence the organisation cannot accept. Overrides every score.

Conditional

Attaches a condition

The use case may proceed once a named control, approval or piece of evidence is in place. The condition travels with the initiative into delivery.

Advisory

Records a caution

Does not change the outcome, but is recorded so the reader knows it was considered rather than overlooked.

Every blocking gate must state a remediation path or an explicit statement that none exists. A gate that stops a decision without explaining what would change it is a dead end, not governance.

Thirteen alternatives, scored on the same basis

A recommendation is only meaningful relative to what was rejected. The comparison is mandatory: an assessment cannot produce a verdict without it, and each option carries a stated reason it was not selected.

The full comparison set considered for every use case.
Alternative	Typically preferred when
Do nothing	The cost of change exceeds the cost of the problem.
Process redesign	The process itself is the problem; automating it would encode the fault.
Conventional automation	The task is deterministic and rules are cheaper, faster and auditable.
Existing platform capability	Software already licensed provides this without new investment.
Additional staffing	The constraint is capacity rather than capability.
Outsource the process	The activity is not differentiating and a provider can absorb it.
Buy an AI product	The capability is commoditised and a supplier already sells it.
Adopt an AI platform feature	A vendor you already use offers the capability within existing terms.
Build a custom AI solution	The requirement is genuinely specific and can be sustained internally.
Hybrid human and AI	The value is in assisting people, not replacing the judgement.
Data foundation investment	No AI approach can succeed until the data is fit to support it.
Pilot before commitment	The uncertainty is real but resolvable by a bounded, measurable trial.
Defer the decision	Regulatory, market or organisational conditions are about to change.

Capability fit is not negotiable against price

An option that cannot deliver a capability the use case requires has its capability fit capped, and cannot be recommended below a minimum threshold however cheap or fast it appears. Otherwise “do nothing” would win almost every comparison on cost alone.

Financial modelling

Every financial output is a range built from named assumptions. Each assumption records its basis and whether it is evidenced or estimated, and the report states which one the result is most sensitive to.

Rules that are enforced in code

Money is held in integer minor units. No floating-point drift.
No currency conversion anywhere. Mixing currencies through an invented rate would make a comparison look sounder than it is.
Results are rounded to significant figures, so an estimate never displays a precision its inputs cannot support.
Where inputs are insufficient, no figures are produced at all and the missing inputs are named.

What a financial section contains

Cost range: build, run, change and exit
Benefit range by value pattern
Net benefit and net present value as ranges
Payback period as a range, or a statement that it is not reached
Sensitivity to the assumptions that matter most
A count of assumptions that remain unevidenced

Confidence, and the refusal to conclude

Confidence is calculated from how much of the applicable question set was answered, how much of it was evidenced, how consistent the answers were, and how mature the underlying content is. It is not an impression.

How confidence changes what the platform is willing to say.
Band	What is reported
High	A verdict with financial ranges, conditions and a review date. The remaining gaps are still listed.
Moderate	A verdict, with the specific weaknesses in the evidence base stated alongside it.
Low	Usually a request for further evidence rather than a recommendation, naming exactly what is missing and what answering it would change.

On the maturity of the built-in content

The industry and regional content shipped with the platform was authored from general domain knowledge. It has not been reviewed by named sector specialists or legal counsel, and every pack carries its validation status inside the product and in the reports it produces. Regulatory content should be confirmed with your own advisers before it informs a decision. See the published limitations.

Reproducibility

An assessment pins the exact version of every content pack it used. Re-running the same assessment produces the same verdict, and a report carries an integrity hash so a reader can confirm the document has not been altered since release.

When content is updated, existing assessments are unaffected. A decision is evaluated against the rules that existed when it was made, not the rules that exist when someone later opens the file.