What an AI audit actually looks like

An OpenKit AI audit is an independent, builder-neutral assessment of where AI fits your business, and where it does not. It maps the work your people really do, ranks the candidate workflows, and hands you a report and a costed plan you can act on.

The wrong workflow gets picked because it demos well, a tool gets bought because the vendor was persuasive, and later there is a subscription nobody opens and a team that trusts AI a little less than it did. The audit exists to stop that happening before the money moves.

We start with your people doing their jobs

Before anything gets ranked we sit with the people doing the work: the finance lead walking us through a month-end close, or the person who quietly built the spreadsheet the whole company runs on. Stacks and licences come later, because the systems that matter are the ones people are already working around.

The same questions repeat in every session. Where does volume erode judgement, and what do people search for and fail to find? We also ask which parts of the work a team would never hand to a machine even where it could do them, and that answer shapes the engagement more than anything else we hear.

Alongside the interviews we map where the data lives and who is allowed to see it. We check what your regulators and insurers expect, and what your current systems can support without a rip-and-replace.

The best output of an audit is often the list of things you should not build.

Then we rank what we found

Everything surfaced in the interviews goes into one table, and each candidate is weighed on who would be accountable when it goes wrong and whether they could defend that to a regulator or an insurer. Then we look at what it takes to fit into the systems you already run, and how long the payback really takes.

Most candidate workflows die in that table. What survives is usually two or three where the case is clear enough that it stops being a leap of faith, and the report says what we would do with each one, which is sometimes to build it, sometimes to buy something that already exists, and sometimes to leave the process alone. Where an off-the-shelf tool costs less than a lunch and does the job, we would rather tell you that than build you something.

What you walk away with

The deliverable is a written report a board can read in one sitting, presented rather than emailed over. It carries the ranked workflows with the reasoning shown, and a prioritised implementation roadmap, typically an Excel Gantt chart, priced against what the audit found. When EMQN needed to know whether an AI marking platform was viable at all, that is the shape the work took. We tested six frontier models against representative reports in all six languages the platform would have to mark in. Per-criterion accuracy came out at 93 to 96 percent, and strict accuracy, meaning all seventeen criteria correct on the same report, dropped to 45 percent, which is what settled the question of whether a human stays in the loop. You can see that discovery engagement in the portfolio.

What happens next

The engagement is scoped and priced before it starts. Interviews and analysis come first, then the report, the plan and the costings, so you finish with the roadmap and a cost against each line. What follows is a separate decision: an Embedded AI Lead engagement to carry the roadmap, a bespoke build where nothing off the shelf fits, or private deployment where the data can’t leave your walls.

The AI Audit page sets out the engagement in full. The pricing page has the fees.