What an AI Audit Actually Looks At

The word audit suggests somebody inspecting your systems. That is not what this is, and businesses that expect a technology review are usually surprised by how little of it is about technology.

What is actually being looked for is where time goes and why, which is a question about how work moves rather than about what software you own.

Why it starts with the problem, not the tools

Statistics Canada, in a release dated 16 June 2025, found that 12.2% of Canadian businesses had used AI to produce goods or deliver services in the previous twelve months. In the same survey, 41.2% reported AI investment as not relevant to their operations.

Four in ten saying it does not apply to them is the interesting figure. For some it is genuinely true. For a great many it is a description of what they have been shown, which is usually a chatbot or a writing tool, neither of which helps a contractor whose staff spend six hours a week chasing insurance certificates.

So an audit that starts by asking what AI you could adopt produces nothing useful. It has to start from where your business currently loses hours, and work backwards.

The four things being looked for

Where work waits. Not where people are busy. Where a job sits doing nothing between two stages, because that is where elapsed time comes from and it is almost always longer than anybody estimates. A quote that takes three days usually involves twenty minutes of work.

Where information gets retyped. Anything entered into more than one place. It is the clearest signal available, it is easy to count, and it is almost always the best first project because nobody defends it.

What waits on one person. The approval, the answer, the specification only one person can produce. That is your capacity ceiling and it is usually a single named individual.

What arrives unstructured. Documents, emails, photographs, forms filled in by hand. This is the category that was impossible to automate until recently and where the largest untapped gains sit in most businesses.

The questions that get asked

• What do you type into more than one place. Ask three people and the list will be longer than you expect.

• What waits for you specifically. Directed at the owner, and the answer defines the ceiling on the business.

• What happens when somebody is on holiday. Reveals every undocumented dependency in about four minutes.

• What do you chase. Chasing is pure overhead and every instance is a system that does not exist.

• What do clients ask that you have answered before. Repeated questions are a measurable, removable cost.

• What number would you want on a Monday morning. The answer is almost never what the current reporting shows.

What gets measured before anything is proposed

One number per candidate process, recorded as it stands today. Elapsed time from trigger to done, hours consumed, error rate, proportion handled within a day.

This is the step most reviews skip, and it is why so many projects are later described as having produced no measurable return. Without a baseline the result is a matter of opinion, and opinions do not survive a difficult year.

What a useful audit tells you not to do

An audit that returns a list of everything that could be automated is a sales document. A useful one is shorter and says no to things.

Processes nobody has agreed on, where two experienced people describe the work differently. Fix that first, in a room, in an hour.

Steps that should be deleted rather than automated, which is a fifth to a third of what gets mapped in most businesses.

Anything running monthly rather than daily, where the build outlives the benefit.

And anything where being wrong carries a cost the business cannot absorb without a control that does not exist yet.

What you should have at the end

Three things, and if you have anything longer you have been sold a programme.

A ranked list of processes with an estimate of hours against each. One recommended first project, narrow enough to be running inside a month. And the number it should move, with today’s figure written next to it.

Everything else is detail that can wait until the first thing works.

Who needs to be in the room

The people who actually do the work, and this is where most reviews go wrong before they start.

An audit conducted only with owners and managers produces a map of the official process, which is not the process. The exceptions, the workarounds and the informal steps live with the person doing the job, and none of it appears unless they are asked directly and believe the answer is safe to give.

Say explicitly that nothing here is an assessment of anybody’s performance. Without that, people describe the process as it is supposed to work, and every recommendation that follows is built on a description of something that does not exist.

What to prepare beforehand

Very little, and preparing too much is its own failure.

A list of the software you use, a rough sense of how many enquiries, jobs or invoices you handle a month, and permission for somebody to look at where work currently sits. That is enough.

Do not tidy anything up first. The mess is the information. A business that spends three weeks preparing has, by definition, presented a version of itself that does not match a normal Tuesday.

How long it takes

Days, not weeks. Conversations with three or four people who actually do the work, a look at where things sit in your systems, and a short document.

Be wary of anything scoped as a multi week discovery phase producing a strategy. That shape of engagement is designed for organisations with a hundred stakeholders. You have four, and they can be in a room on a Tuesday.

Sources

• Statistics Canada, Analysis on artificial intelligence use by businesses in Canada, second quarter of 2025, released 16 June 2025.

AI Optimize runs exactly this, and the deliverable includes what we think you should not build. That work sits under Workflow Automation.