Eight questions that separate organisations prepared to run AI in production from those prepared to talk about it, covering use cases, data, governance, risk, oversight, ownership, vendors and adoption.
AI initiatives rarely fail at the model. They fail at the unglamorous surroundings: data that cannot support the use case, governance nobody defined, and processes that simply ignore the output. Eight questions, fixed scoring, nothing uploaded.
Last reviewed 3 July 2026 · Free and ungated
Put the AI initiative to selected senior operators who have run these systems in production, and receive a confidential report before committing budget.
Get independent perspectivesAnswer for the specific AI initiative closest to funding, not for your most polished pilot.
These answers describe an organisation preparing an announcement, not a deployment. Committing serious budget now means paying to discover in production the gaps in data, ownership and oversight that could have been found on paper.
There is genuine substance here (probably a working pilot and real intent) but the scaffolding for production is incomplete. This is the most dangerous band for AI specifically, because pilots create pressure to scale at exactly the moment governance and data readiness have not caught up.
The fundamentals for responsible deployment are largely present, and the remaining weaknesses are specific rather than structural. The risk now is uneven depth: governance that exists on paper but has never handled a live incident, or adoption that holds only while the project team is watching.
The organisation has done the unfashionable work: named owners, tested controls, adoption designed rather than hoped for. What a strong score cannot certify is judgement: whether this use case deserves AI at all, and whether the business case survives honest attribution.
Not model quality, and not enthusiasm. It checks the eight conditions that decide whether an AI initiative survives contact with production: a defined use case, data that holds up outside the pilot, accountable governance, real risk controls, designed oversight, an owner for the long run, managed vendor exposure and users who will actually adopt the output.
Three answers per question, worth 0, 5 or 10 points; the total is normalised to a score out of 100 and mapped to four readiness bands. No free text, no interpretation layer and (deliberately, given the subject) no AI: the assessment is fixed arithmetic in your browser.
Finding these gaps before commitment is cheap; finding them in production is not. An honest 35 today, acted on, beats a generous 70 that unravels in front of a regulator or a customer. The assessment exists to move the discovery earlier, when changing course costs a meeting rather than a write-down.
Score the one closest to a funding or deployment decision, because that is where readiness gaps convert into losses. Scoring an average across the portfolio blurs exactly the differences that matter.
Only indirectly, through the governance and risk-control questions. It cannot substitute for legal review under the regulations that apply to your sector and territory; treat compliance as a parallel workstream, not a question to average in.
Platform controls are useful, but accountability cannot be outsourced: when the system is wrong, the consequences land on your organisation, not the vendor. Score the governance question on what your organisation owns, exercises and can evidence.