A useful AI audit hands you four things: one named bottleneck, the system best placed to fix it, a real number for what it would cost, and a straight answer when AI isn't the right fix. If it ends in a product recommendation before anyone has looked at how your work moves, you sat through a sales call. And if nobody can name a specific problem yet, the audit isn't your next step. Writing down how the work runs today is.
TL;DR
- A real audit starts with your process, not a product.
- Leave with a bottleneck, a system, a cost and a timeline, in writing.
- If the rules live in one person's head, write them down first.
What is an AI audit for a business, and what isn't it?
Three different things get called an AI audit, and only one of them is about your operation.
The first is a compliance review of AI systems you already run, checking models and data against rules and regulations. The second is AI used inside a financial audit, which is a software question for accountants. The third, and the one most owners mean, is a review of your own business to find where AI would pay off. Some firms call that an AI readiness assessment, so you'll see both phrases for the same work. If you'd like to try a version yourself first, here's an AI readiness assessment you can run in an afternoon.
This post is about the third one. If you don't run any AI yet and you're being sold the first, you're buying an answer to a question you haven't asked.
What should an AI audit tell you before you spend anything?
Four outputs are worth your time. Everything else is background for them.
- One named bottleneck. Not a list of 12 opportunities. One process, named the way you name it, with a count of how often it runs and how long it takes now.
- The system that fits it. What would get built, what it would do on its own, and where a person still signs off. If the answer is just a tool name, press for the rest.
- What it costs to build and to keep running. Both numbers. A build price with no running cost hides the part that shows up every month.
- A clear "not this." A good auditor will tell you when the fix is a process change, cheaper software, or a conversation with the people doing the work.
Anything less is a worksheet with a sales call attached. For the money side, how to estimate AI ROI before you automate anything shows how to check the numbers you're given.
What should an audit look at in your business?
Where work piles up, not where AI is fashionable.
| What the audit looks at | What you decide from it |
|---|
| One repeatable process and how often it runs | Whether the time it eats is worth a build at all |
| Work rebuilt from scratch each time (quotes, intake, onboarding, reports) | Which steps follow rules you could write down |
| Where answers live (inboxes, folders, one person's head) | Whether the problem is finding knowledge or creating it |
| Calls and leads that go unanswered | Whether the fix is coverage, routing or staffing |
| Which tools already hold the data | Whether a system can be built on tools you already pay for |
| Who runs it after handoff | Whether you'll own it outright or depend on the builder |
That last row is the one owners skip and regret. Ask it during the audit, not after the build.
How do you tell a real audit from a sales call?
Watch the first 10 minutes. A real audit opens with questions about your process: what comes in, who touches it, where it stalls, how often. A sales call opens with what the firm sells, then works backward to a reason you need it.
Then look at what you're left holding. You should get something in writing that another builder could quote from. If the output only makes sense when that firm does the work, you paid for their sales process. The free audit from AI Marketing Labs ends in a one-page write-up that's yours to keep, whether you work with us, hire someone else, or do it yourself.
When is an AI audit the wrong next step?
Sometimes it is, and a firm worth hiring will say so.
When nobody can name a problem. "We should probably do something with AI" isn't a brief, and an audit will produce a tidy document and no decision. RAND's research on why AI projects go wrong, based on interviews with 65 data scientists and engineers, noted that "by some estimates, more than 80 percent of AI projects fail," and found the most common root cause was people misunderstanding or miscommunicating what problem the AI was meant to solve.
When your rules live in one person's head. If your estimator prices from memory, or your dispatcher knows who can do which job without looking, there's nothing to automate yet. Automating a broken or undocumented process is one reason pilots stall. In a survey of 500 senior IT and AI leaders at US companies with at least $50 million in revenue, "13% report that a piloted project has stalled or been abandoned outright." Those companies are bigger than most readers here, so treat it as a rough guide. Writing the rules down helps even if you never buy anything.
When cash is tight. A build competing with payroll this quarter is a bad bet.
Some decisions stay with you no matter what. What you charge, who you hire and how much risk you'll carry are judgment calls. An audit can show you where the time goes. It can't tell you what kind of company you want to run, and sometimes the right answer is to skip AI for now.
How do you get ready for an audit in a week?
Do this before the call and the audit gets far more useful. None of it requires buying anything. For a longer list, see what to bring to your first AI strategy session.
- Pick the one process that annoys you most: quoting, intake, scheduling, reporting or follow-up.
- Count it for a week: how often it runs, how long each one takes, and who does it.
- Write down the rules you follow today, exceptions included, and mark the ones that live only in someone's head.
- List the tools that already hold the data: CRM, email, spreadsheets, scheduling app, accounting.
- Name who would own the finished system after handoff, and check they have a couple of hours a week for it.
- Decide what a win looks like in a number you already track, before anyone proposes a build.
Related articles
Frequently asked questions
How long should an AI audit take?
Mapping one process and naming the bottleneck can be done in a single working session of about an hour. Larger paid assessments run for weeks. Longer isn't automatically better, so ask what the extra time produces that you'd actually use.
What should you get in writing at the end of an AI audit?
The named bottleneck, the system proposed to fix it, what it would cost to build and to run, and a timeline. It should be clear enough that another builder could quote from it. A deck about AI in general is marketing.
Should you pay for an AI audit?
Only if the output is yours to take anywhere, the scope is fixed, and the price is small next to the build it's scoping. Get the fee and the scope in writing before you agree.
Who should be in the room for an AI audit?
The person who makes the budget call, and the person who actually does the work being discussed. The owner knows what it costs the business; the person doing the work knows the exceptions nobody wrote down.
