How to Audit Accounting Workflows Before Deploying AI

An audit doesn’t mean hiring an external consultant or filling out a form. It means observing and documenting what your team actually does, from start to finish. Not what the manual says they should do. What they really do. This matters because the gap between “how we’re supposed to do it” and “how we actually do it” is often where the real problems hide. Here’s how to audit accounting workflows before you deploy AI.

This blog is part of our series, How Accountants Get Real Value from AI.

Which workflow should you audit first?

Pick one common client type or service line, and make sure it’s something you do regularly and are comfortable documenting in detail. For example, VAT, year-end accounts, or payroll.

How do you track the complete workflow cycle?

From the moment the work comes in (or the client relationship starts on that service) to the moment it’s delivered, invoiced, and archived. Follow the actual work. Write down every step, be specific and don’t generalise.

What questions reveal the real bottlenecks?

Who does it? Is it always the same person, or does it rotate? If it rotates, do they do it the same way each time? Do different people approach it differently? If so, which approach is better? Why?

How long does it actually take? Not how long you think it should take, but how long it really takes? Time a few cycles and average them. Where do they wait for something else? For approval from a partner? For a client to send information? For another system to complete a task?
Where do they switch between tools? Every switch is a friction point. Every switch is an opportunity for error, for miscommunication, for data to get lost or mistyped. How many times does your team switch tools in a single workflow? How much of that switching could be eliminated with better process design or software integration?
Where do they make decisions? Are those decisions documented? Do different team members make them differently? Are there rules for the decisions, or does each person interpret them based on their own judgment? Inconsistent decision-making is a sign that the workflow isn’t clear.
Where do they verify their work? Do they spot-check every client? Every tenth client? Do they wait for an error to surface in the wild? Do they have a formal review step, or do people just check their own work? Verification points tell you where the risks are. They also tell you where there’s duplicate effort.

Most practices discover that the time they think they’re spending on execution is actually spent on waiting, switching tools, finding information, and verifying work.

What patterns emerge from your workflow map?

Tasks taking longer than expected due to waiting or tool-switching, inconsistent processes, and unclear handoffs. One week of mapping saves months of wasted effort.

This is tedious work, but do it anyway. One week of tedious mapping saves months of wasted effort later. Patterns will emerge. For example, you might notice that a task labelled “20 minutes” actually takes an hour because someone waits for approval or for information. You could find that what should be a standardised process has six different versions depending on who does it. You’ll see where the software doesn’t talk to the software, forcing manual re-entry. You’ll spot where people are doing the same verification twice because there’s no trust in the first verification.

Where are your real workflow blockages?

Once you have the map, you can see the real blockages. In most practices, 80% of the time is consumed by 20% of the tasks. But not always because those tasks are intrinsically hard, but because the task was never properly designed or it’s been done the same inefficient way for ten years and nobody questioned it.

Now you’re ready to ask which of these problems can AI actually solve? Sometimes it’s one or two specific steps where AI can add value. Sometimes it’s most of them but only if you fix the underlying workflow first. For further information, check out our blog on why using AI on broken processes creates faster failure.

How does the audit define what success looks like?

The audit also tells you what success looks like. If you’re spending two hours manually collating data from five different sources into a report, success means “one-click reporting from consolidated data.” That’s your target. Now you can evaluate whether AI, software integration, or process redesign gets you there.

This content may have been developed with the assistance of AI tools. Content is reviewed by a member of the Bright team before publication. Guidance on legislation can be obtained from your industry body.