Test Concept Page
AI is a practical tool in our work — applied only where it improves insight, speeds decision-making, or reduces effort, always in service of the outcomes our clients care about.
We do not introduce AI for its own sake. Every application is judged against a simple test: does this measurably help the client understand performance, redesign work, or deliver better results?
AI helps us analyse large volumes of operational data, documents, and performance metrics quickly. This shortens the time needed to form a clear picture of current performance and surface patterns that would otherwise take weeks to identify.
We use AI to synthesise findings, test hypotheses, and explore alternative operating models. The output is always reviewed and refined by experienced consultants so that recommendations remain practical and grounded in the client's context.
When redesigning how work gets done, AI can model process variations, estimate effort, and highlight bottlenecks. This allows us to present clearer options and trade-offs to decision-makers.
AI assists with drafting structured reports, business cases, and presentation material. Human expertise remains in control of content, judgment, and final quality — resulting in faster turnaround without compromising standards.
Worked Example
Most organisations already know their technology portfolio is a problem. The evidence for it sits in a spreadsheet one person maintains, a slide pack that was accurate in March, and a hand-built page nobody can update. The question is rarely do we need a view of this. It is can we trust the view we are given.
So we built one, to find out what a trustworthy version actually looks like. The National Regulatory Services Agency is a synthetic organisation we constructed for the purpose — a realistic Crown agency estate, built entirely from invented data, so the tool could be designed, tested and attacked without touching a single real record.
See the example NRSA IT Portfolio and Roadmap Dashboard
Synthetic figures. No client data of any kind was used to build, test or demonstrate this application.
Everything below reads from the same governed source. Nobody is reconciling two versions of the truth in a meeting.
The decisions currently waiting on someone, ranked — drawn only from what your role is permitted to see.
Every system scored by domain, with the number always printed beside the colour, and the drivers behind the score one click away.
Initiatives laid across financial years, with the fixed, dated commitments marked separately from the intentions.
Select a system and trace what goes down with it. The blast radius is ranked, and the sharpest dependencies are flagged.
The full initiative list — sortable, filterable, exportable. Initiatives not mapped to any system are called out rather than buried.
Delay one initiative and watch everything coupled to it slip. Labelled assumption throughout — because that is what it is.
A portfolio dashboard is easy to build and easy to make dangerous. These three constraints were fixed before any code was written, and they are enforced in the application rather than promised in a policy.
Rule 01
Every figure on every screen carries a confidence tier — fact, indicative, or assumption. The tier is shown as a written label and a distinct border pattern, never as colour alone, so an assumption can never be mistaken for a fact, and a reader with colour vision deficiency loses nothing.
The tier labels as they appear in the application.
Rule 02
Stale, absent, withheld and failed-to-load are all first-class states with their own visible treatment. The dashboard will tell you when it does not know something. It will never quietly render last quarter's number as though it were today's — which is the single most common way a reporting tool misleads the people relying on it.
Rule 03
Sensitive fields — risk detail, named owners, commercial position — are stripped on the server before anything is sent. Your role is read from a signed session the browser cannot forge, and a role asserted by the browser is rejected outright. Opening developer tools reveals nothing, because nothing was sent. Hiding a field in the interface is not a control; removing it at the source is.
AI did the volume work: the data model, the migrations, the interface, the endpoints, and the test suite. It did not get the final word. Each stage was gated, and at each gate an independent adversarial review ran with one instruction — attack this work and prove the claims false.
That review found defects in code that had already been declared finished. Two medium-severity issues in an early pass, one of them in the sign-in path. Four more in a later pass, two of them high severity. All were fixed before the milestone was allowed to close. The application now carries a verification record: 69 back-end tests, a front-end test suite, a clean typecheck, and two documented adversarial passes with every finding closed.
This is the part that matters for client work. AI compresses the build. The review is what makes the result defensible. A named person remains accountable for what is handed over — and the evidence that the work was tested is part of the deliverable, not a claim about it.
Organisation-agnostic by design. The estate, its domains and the financial years are data, not code — a different agency is a different data set, not a rebuild.
Runs either as a hosted application against a live database, or as a single self-contained file that opens in a browser with no backend at all — useful for a board session on a locked-down device.
Sign-in is deliberately an adapter. The demonstration uses a role switch; it is built so an agency's own identity provider can take its place without the permission logic changing.
Read-only today. Owner editing with a full audit trail is the next stage of work.
We can walk you through the working application on a call. Arrange a demonstration →
We do not replace professional judgment with automated recommendations.
We do not apply AI where the cost, complexity, or risk outweighs the benefit.
We do not use client data in ways that compromise confidentiality or security.
We do not present AI-generated content as finished advice without rigorous human review.
We do not present a projection as a fact. Every figure we hand you is labelled with how much confidence it deserves.
Clients receive clearer insight earlier, more options to consider, and high-quality deliverables in less time — while retaining full confidence that experienced practitioners are accountable for the advice and recommendations.
Discuss your objectives