What does an AI audit of my business include, and what should I pay?
Short answer: A useful AI audit gives you a written, ranked list of where AI could save time or make money in your business, what each item would take to build, what could go wrong, and exactly what to do first. An AI audit is a short review of how your business runs, done by someone who knows AI tools, to decide where AI is worth using and where it is not. Price depends on scope, so ask for a fixed price in writing. The test of a good one: could your own team start work from the document alone, with the consultant gone?
You want to know what a person would actually hand you after an AI audit, and whether it is worth paying for before you build anything.
I'm Ranjit, and I run Blur Studio: 15+ years building software, 120 apps taken live, and clients who have gone on to raise $10M+. I build AI systems and processes that help businesses do a lot more with a lot less time. I also work differently from most of what else is out there, more on that further down this page. Say hello.
If you would rather skip the shopping, I run a paid 3-day AI Strategy Sprint that does this job, and it starts with a free 15-minute call.
What an AI audit is, and what it is not
The phrase gets used for three different things, and that is why prices and results are all over the place. If you want to know what the work after an audit costs, see what AI implementation costs a small business.
- An opportunity audit. This is the one most owners want. Someone looks at how work flows through your business and tells you which jobs AI could help with, in order of payoff. It is also called an AI opportunity assessment.
- A risk and governance review. This checks what data staff already paste into AI tools, what your vendors do with it, and what rules you need. The US standards body NIST publishes a voluntary framework for this called the AI Risk Management Framework, built around four functions: Govern, Map, Measure and Manage. Most small firms do not need the whole framework, but the questions in it are a good guide.
- A technical audit. This is a review of an AI system you already built, such as a chatbot or an automation, for errors and security holes. It is a different job.
This article is about the first one, with a few risk checks built in. If you only need rules for staff, my one-page AI usage policy template covers that without any audit.
What you should get, page by page
Here is the shape of a document worth paying for. Ask any provider to show you a sample with the client details removed. If they cannot, that tells you something.
Page 1: A map of how work really moves
One page that lists your main repeating jobs: taking enquiries, quoting, scheduling, invoicing, chasing payment, onboarding, reporting. For each one: who does it, which tool they use, roughly how many times a week it happens, and how long each time takes. This page comes from talking to the people who do the work, not from the owner's view alone.
Page 2: The ranked table
This is the heart of it. Each opportunity gets a row. Below is a made-up example for a 12-person plumbing and heating firm. The figures are invented to show the format. Wondering whether a consultant is the cheaper route? See what an AI consultant costs.
| Opportunity | Hours a week now | Hours saved (estimate) | Effort to build | Risk | Rank |
|---|---|---|---|---|---|
| Draft replies to new enquiries | 6 | 4 | Low | Low, person approves | 1 |
| Chase quotes with no answer | 3 | 2.5 | Low | Low | 2 |
| Turn job notes into invoices | 5 | 3 | Medium | Medium, money involved | 3 |
| Weekly owner report | 2 | 1.5 | Medium | Low | 4 |
| AI phone agent for after-hours calls | 1 | 1 | High | High, customer-facing | 5, not yet |
Notice two things. The estimates are written as estimates, with the guess behind them stated. And the last row is ranked low on purpose. A good audit tells you what not to do yet. If every row says "high impact", you are reading a sales page.
Page 3: Risk notes
For each top item, what data does it touch, who sees the output, and what happens if the AI is wrong. For example: a drafted reply goes through a person, so a mistake is caught. An invoice built from notes touches money, so it needs a check before it sends. This is where a human has to stay in the loop, and the audit should say so in plain words. NIST's framework asks the same: map where a system is used and what harm a failure could cause before you rely on it. The OWASP project, which publishes security guidance for AI language models, lists excessive agency, meaning an AI allowed to act with too little oversight, as a named risk.
Page 4: A spec for the first build
One item, described well enough that a developer or a no-code builder could start Monday. It names the trigger (a new enquiry arrives), the inputs (the email and your price list), the output (a draft reply in your inbox), who approves, and how you will know it works (average reply time falls from four hours to under one). Without this page you have a report, not a plan.
Page 5: What to leave alone
Tasks where the person's judgment or relationship is the product. A good audit has this page.
What it costs
Prices vary widely, from a short call and report to a multi-week review, so a single average would mislead. Here is how to judge a quote.
- Ask for a fixed price in writing that names the deliverables, the days, and who you will speak to. Open-ended hourly audits drift.
- Price should rise with the number of people interviewed, not with how impressive the slide deck is.
- Free audits are often sales calls. That can be fine, but know that the output is shaped to sell you the next step.
- Cheap does not mean light. In my view, a 3-day review of a 10-person firm can be thorough, and a 6-week review of the same firm is mostly padding.
A worked example with made-up numbers: say your office manager spends 15 hours a week on the top three items in the table above, and a good audit shows 10 of those hours could be saved. At a loaded cost of $30 an hour, that is $300 a week, or about $15,600 a year. If building the first item costs more than the first year of savings on it, rank it lower. If the audit cost is a fraction of one year of savings, it paid for itself on paper. Do not trust the savings number until the first build proves it.
The test of whether the output is usable
Before you pay, ask these five questions. Then check the finished document against them.
- Could my best employee read it and know what to do first, without a call?
- Does each opportunity name the trigger, the data and the person who approves?
- Are hours saved written as estimates with the basis stated?
- Does it say what not to automate?
- Does it tell me which tools or builders could do the work, including doing it myself?
If you answer no to two or more, you have bought a pitch. A real audit works even if you never hire the author.
Red flags when shopping
- A form that ends in a booked demo of a product. The audit is the lead capture.
- A score out of 100. Scores feel precise and tell you nothing about what to do on Monday.
- No interviews with the people who do the work.
- Promised percentages such as "cut costs 40 percent" before anyone has seen your business. Ask what the number is based on, and treat an answer of "our other clients" as no answer.
- No mention of your data. A good audit asks what data staff already put into AI tools. My guide on keeping business data out of AI training shows what to check.
Audit versus just trying it
For a business under about 5 people, you can often skip a paid audit. Spend a week listing your repeating jobs, time them, and try one drafting task with an AI assistant such as Claude or ChatGPT. My guide to AI for small business and five starter workflows are a good start. An audit starts to earn its cost when you have several staff, several tools and no clear view of where time goes.
If you are also deciding whether to hire help at all, read what an AI consultant does and charges and a step-by-step AI implementation plan. If the top item on your table is a repeating admin job, business process automation explains the building blocks.
How I'd approach this
I would not pay for a score or a deck. I would pay for the five pages above, in plain words, and I would check them with the usability test before signing off. I would also insist the audit tells me one thing not to do.
That is how I run the AI Strategy Sprint: a paid 3-day first step where you and I find where AI could transform the business and generate new value, and you get the exact steps written down. It starts with a free 15-minute AI brainstorm call, so you can see whether it fits before you commit to anything.
Questions people ask
What is an AI audit for a small business?
It is a short review of how your business runs that ends with a written list of where AI is worth using, in order of payoff, with risks and a first step. Done well, it is a plan you can act on. Done badly, it is a lead form.
How long does an AI audit take?
For a business of 5 to 50 people, a few days of interviews and review is, in my view, enough to produce a usable plan. Be careful of audits that take many weeks for a small firm, since the extra time is usually padding. My guide to hiring an AI consultant covers what to ask about timing.
Do I need an audit before I build an AI tool?
Not always. If you already know the one job that wastes the most time, build a small version of that first. An audit pays off when you have many possible jobs and no way to rank them.
Is a free AI audit worth taking?
It can be, as long as you understand it is also a sales process. Use it to get a first list, then run the usability test above on whatever you receive.
Should the audit cover data privacy?
Yes, at least the basics: what data staff paste into AI tools, what each vendor does with it, and who can see the output. The US FTC's guide to protecting personal information asks businesses to take stock of what data they hold and keep only what they need, which is a sound starting rule. Ask a lawyer about anything involving health, children or financial data.
Can I do an AI audit myself?
Yes, for a small team. List each repeating job, time it, note who does it, and score each on hours spent, how rule-based it is and how bad a mistake would be. Rank them and try the top one. My AI readiness checklist is a good first pass.
More in this topic: Problem solving: AI strategy and adoption (24 guides) →