Who buys an audit, and who should not
The audit is for a business that suspects AI could help but cannot yet say where, or has three competing ideas and no way to choose between them, or needs a costed case before a budget holder will sign anything.
It is not for you if you already know what to build and the data is in place. In that case an audit is a two-week delay wearing a suit. Book a call, tell us the project, and we will quote the build instead.
What the two weeks look like
Week one is spent with the people doing the work, not only the people managing it. Four to six interviews of about an hour, each one asking the same unglamorous questions: what do you do all day, which parts are repetitive, where do you wait for someone else, and what goes wrong. Then a session on the data, which is where most AI plans quietly die.
Week two is ours. We score what we found, check whether the data behind each idea actually supports it, cost the most promising one properly, write it up, and walk your team through it.
How we score an opportunity
Every idea gets scored on two axes, and both matter.
| We ask | Because |
|---|---|
| How much time or money does this save, per week? | An idea that saves twenty minutes a month is a hobby |
| Does the data it needs exist, and is it usable? | This is the most common reason AI projects fail, and it is knowable in advance |
| What does a wrong answer cost? | A wrong product recommendation is a shrug; a wrong invoice is a legal problem |
| Can a person check the output? | If nobody can tell whether it is right, you cannot run it |
| Is there an off-the-shelf tool for this already? | Often there is, and buying it beats building it |
| How would you measure it? | If we cannot design a test for it, we cannot tell you it worked |
An idea that scores well on value and badly on data readiness is not a no. It is a sequencing problem, and the plan says what to fix first.
The things we regularly say no to
Saying no is most of the value in the first week. It is also why we sell the audit as a fixed fee rather than free: an assessment that only ever recommends a build is a sales call with a spreadsheet attached.
The data readiness check
Most AI plans do not fail on the model. They fail because the data an idea needed turned out to live in three systems that disagree, or in a field people fill in inconsistently, or nowhere at all.
So we check, per idea: does the data exist, who owns it, how far back does it go, how clean is it, is it allowed to leave your systems, and what would it take to make it usable. That answer changes the cost and the order of the plan more than any other single factor.
What you leave with
A written report you own, covering the scored opportunities, a build, buy or wait recommendation for each, the data readiness findings, a costed plan for whichever one you pick first, the measurement plan, and the risks. Then a session walking your team through it.
If you take it to another agency, it will work there too. That is deliberate. A plan that only functions while we are holding it is not a plan.
For the version of this written as an explainer rather than an offer, including what to expect from other firms selling the same thing, see what is an AI readiness audit.
Why we can do this quickly
We build AI products of our own and run them, including Crodo, a macOS voice assistant, and Decornoa, which generates interior redesigns from a photo. Both taught us the things that do not show up in a demo: what inference actually costs at volume, where quality drifts after a model update, and how much of the work is the boring plumbing around the model.
That is the difference between an assessment written from a survey of the market and one written by people who have paid the bills.
What happens after
Most audits end in one of three ways. You build the first project with us, you build it with your own team using the plan, or you do something cheaper than AI that we found along the way. All three are fine outcomes, and we would rather be the firm you call again than the one that sold you a project you did not need.
If the first project is an agent, that is AI agents and automation. If it is a feature inside software you already have, that is AI integration. If it is a new product, that is AI MVP development.