How it works
From an unexamined idea to a decision you can defend
An assessment takes a proposed use case and puts it through the same sequence every time. Nothing is skipped because the answer looks obvious, and the sequence is visible to everyone who reads the report.
Stage 1
Establish the problem
The assessment opens on the business problem, who it affects, the current process and what would count as success. No technology is discussed until this is on record.
Where this stage can stop the processIf success cannot be described in measurable terms, the assessment reports that no technology decision should be made yet. That is a legitimate result.
What is produced
- A stated problem and measurable success criterion
- The current process and its known failure points
- Who is accountable for the outcome
Stage 2
Determine whether AI is required at all
Before any AI-specific question, the assessment establishes whether the task is deterministic, whether existing licensed software already covers it, and whether the problem is really a process design problem wearing a technology costume.
What is produced
- A necessity finding, with reasoning
- Identification of existing capability that may already apply
- An early route to a no-AI outcome where one is warranted
Stage 3
Gather the applicable evidence
Questions are selected by industry, region, deployment context and prior answers. Sensitive questions explain why they are asked. Some answers require evidence before they can support a financial claim.
What is produced
- A recorded answer set with sources where evidence was required
- Named gaps where information was not available
- Flagged contradictions between answers
Stage 4
Score and gate
Eleven dimensions are scored, each with a rationale and its own confidence. Separately, legal, safety, ethical, data and operational gates are evaluated. The two are deliberately independent.
Where this stage can stop the processA blocking gate overrides the scores entirely. A high-scoring use case that crosses a legal or safety gate is refused, and the report names the gate.
What is produced
- Eleven dimension scores with stated reasoning
- Triggered gates with severity and remediation
- Risks with likelihood, impact and required controls
Stage 5
Compare the alternatives
Thirteen alternatives are scored on the same basis: capability fit, cost, time to value, risk, reversibility and organisational fit. An option that cannot deliver the required capability cannot win on price.
What is produced
- A ranked comparison across all thirteen options
- A stated reason each option was not selected
- Explicit marking of options that are not feasible here
Stage 6
Model the value
Costs and benefits are built from named assumptions, each marked as evidenced or estimated. Outputs are ranges, tested for sensitivity, in a single currency with no conversion.
Where this stage can stop the processWhere the inputs do not support a credible financial model, no figures are produced. The report states which inputs are missing instead of estimating them.
What is produced
- Cost and benefit ranges with the assumption list
- Net benefit, net present value and payback as ranges
- The assumption the result is most sensitive to
Stage 7
Conclude, then record
Confidence is calculated from coverage, evidence quality and internal consistency. The verdict follows deterministically. A decision-maker then accepts or overrides it, and an override must carry a written reason.
Where this stage can stop the processWhere confidence is too low to conclude responsibly, the verdict is a request for further evidence, naming precisely what is missing.
What is produced
- One verdict from the closed set of fifteen
- A confidence band and what would raise it
- A sealed decision record with an integrity hash
The assessment adapts to the answers given
There is no fixed questionnaire. The applicable set is derived from the industry, the region, the deployment context and the answers already recorded — so an internal document-summarisation tool and an automated clinical triage system do not receive the same questions.
Every question explains itself
Each question carries the reason it is being asked, what it affects, and where the answer is likely to be found in your organisation.
“I don’t know” is a valid answer
Recorded honestly, it lowers confidence and appears in the report as a named gap. Guessing to make a form complete produces a confident wrong answer.
Contradictions are surfaced
When two answers cannot both be true, the assessment says so and explains which pair conflicts, rather than quietly averaging them.
After the decision
A verdict is not the end of the process. Approved initiatives carry their conditions, review dates and expiry forward, and the forecast that justified the approval is frozen so that later performance is measured against what was actually promised.
| Carried forward | Why it matters |
|---|---|
| Conditions of approval | Each has a description and a date. They remain visible for the life of the initiative rather than living in the minutes of one meeting. |
| Frozen financial baseline | The forecast at the point of approval is sealed. A later, quieter revision of the forecast cannot become the benchmark for success. |
| Review date and expiry | A decision made on stale evidence should not remain in force indefinitely. Overdue reviews and expired decisions are listed. |
| Required controls | The controls the decision depended on are named, so implementation cannot silently drop them. |