AI

AI Workflows for Accounting Firms: Stop Turning Yourself Into a Developer  

One clever AI prompt saves you an hour.

So, you improve it.

Then you build another one for a different job. Then a template. Then another workflow because one client is slightly different.

Before long, you have six AI-assisted jobs running at once, prompts scattered across different places, multiple versions of what should be the same process, and a growing feeling that every new job requires you to teach the AI all over again.

One accounting firm owner summed up the frustration perfectly:
“I’ve turned myself into an AI Agent developer, which wasn’t really the initial aim.”

That line gets to the heart of the problem.

You went looking for leverage. Instead, AI has quietly given you another role.

AI workflows for accounting firms should reduce the amount of work that depends on the owner. They should make recurring tasks easier to complete, easier to repeat and easier for other people in the firm to use.

If you are constantly rebuilding, fixing, and remembering how every AI workflow works, you may be creating more complexity than capacity.

When the AI that saves time also needs maintaining 

 The early stages of AI adoption can be exciting because you can see opportunities everywhere.

A prompt works well, so you improve it. Another task looks suitable for AI, so you build something for that too. Then somebody finds a better way to structure the instructions, so another version appears.

None of those decisions is necessarily wrong.

The issue is what happens when they keep accumulating.

Someone has to remember which version is current.
Someone has to update the instructions when the workflow changes.
Someone has to explain to the team why one prompt works for one type of job but not another.
Someone has to work out what went wrong when the AI produces something unexpected.

And in many small and mid-sized firms, all of that comes back to the owner.

That is the warning sign.

A useful AI workflow should reduce the number of decisions and repetitive tasks that require your involvement. If every efficiency gain creates another thing only you know how to operate, the business has not reduced its dependency on you.

It has simply moved that dependency somewhere new.

Your firm probably needs fewer AI workflows than you think

 The goal is not to collect as many prompts, assistants, and AI tools as possible.

The goal is to identify the recurring work where AI can make a meaningful difference.

Think about the jobs that consume time every week:

• client communications
• meeting summaries and follow-up
• onboarding
• information requests
• procedure writing
• first-draft analysis
• workpaper support
• review preparation

Now ask a commercial question:
Which of these tasks happens often enough that improving it will genuinely create capacity?

Those are the places worth concentrating on.

Build a reliable starting point for each recurring task. Give the AI the context it needs. Be clear about the outcome. Define the expected format. Include the right source material and make the boundaries obvious.

Then use the workflow repeatedly.

Do not rebuild the whole system every time an unusual client or exception appears.

Professional services will always have exceptions. The problem starts when every exception becomes another permanent layer of instructions.

Over time, the workflow becomes harder to understand, harder to teach and harder to trust.

Simplifying it is not going backwards.

It is often how an experiment becomes something the business can actually use. 

Stop spending time improving AI that does not improve the business 

 AI can be interesting enough that improving the tool starts to become the activity.

That is where owners need to keep coming back to ROI.

You can spend a significant amount of time improving a prompt for a task that rarely happens.

The prompt may become technically better.

But has the business become better?

Every AI workflow should earn its place by improving something that matters.

• Does it save meaningful time?
• Does it reduce repetitive work?
• Does it improve consistency?
• Does it speed up turnaround?
• Does it free capacity for higher-value work?
• Does it contribute to a better commercial outcome?

If the answer is no, it may not deserve more of your attention.

There will always be another AI feature, tool, or technique to explore. That does not mean your firm needs all of them.

AI-first does not mean AI everywhere.

It means deliberately looking for the places where AI creates a worthwhile return, then making those uses part of how the firm works.

Sometimes the right decision is to improve an AI workflow.
Sometimes the right decision is to stop using it.

Six AI jobs running at once can create a new kind of overload 

AI gives owners the ability to start and progress work much faster.

But your own attention has not suddenly multiplied.

So, one job is waiting for client information. Another has produced a draft. A third needs a decision. Another has gone in the wrong direction. Two more are sitting in separate conversations you vaguely remember starting yesterday.

You are producing more, but you may feel less in control.

That is where the operating system around the work matters.

Every AI-assisted job should make four things visible:

• where the job is now
• who owns the next action
• what it is waiting for
• what needs to happen before it is complete

This does not automatically require another piece of software.

Use the workflow or practice management system you already have if it can do the job.

What matters is that AI-assisted work does not live only inside private chats, personal prompt libraries or the owner's memory.

If nobody can see where the work is, the firm cannot manage it properly.

Faster output is only useful when the workflow around it can keep up.

Build AI workflows the firm can use without you standing beside them

A useful question for any AI workflow is:
Could another person in the firm use this without the owner explaining it every time?

If not, you may have created personal productivity rather than business capability.

There is a big difference.

Personal productivity helps you get through more work.
Business capability helps the firm get through more work without everything coming back to you.

That means the useful workflows need to be accessible.

Give them clear names. Remove duplicates. Keep one agreed version where possible. Show the team what the workflow is designed to do. Make it clear when it should be used and when it should not.

When an improvement genuinely makes the workflow better, update the shared version rather than creating another private copy.

That is how the value starts moving beyond the person who originally built it.

And that is where AI becomes much more useful to a growing firm. 

The real measure of AI leverage is less dependence on you

The most impressive AI workflow is not necessarily the most valuable one.

A complicated system that only the owner understands can still leave the business dependent on the owner.

A simpler system that the team can use repeatedly may create far more value.

Before adding another AI workflow, ask:

• Does this remove repetitive work?
• Will we use it often enough to justify maintaining it?
• Can another person use it?
• Can we see where the work sits after AI becomes involved?
• Is it creating a measurable business return?

If those answers are weak, adding more AI is unlikely to solve the problem.

There is enormous opportunity for accounting and professional-services firms that learn to use AI well.

But the objective is not to become brilliant at managing AI. The objective is to build a better business.

That means creating capacity, improving consistency, pushing suitable work away from the owner and freeing experienced people to spend more time where their judgement creates the greatest value.

You do not need dozens of clever AI systems to achieve that.

You need a smaller number of useful workflows that solve recurring problems and can operate without requiring your constant attention.

That is when AI starts giving you leverage instead of another job.

PREVIOUS POST

How to Review AI Work Without Rechecking Every Line