Ideas · October 22, 2026

How an Accounting Firm Should Choose Its First AI Product

Most accounting firms use AI to automate bookkeeping. A step-by-step guide to finding the one client-facing AI product worth building, scoring the options, and testing it with a prototype.

A popular format in AI content is the business makeover: take a company, rebuild its operations with AI in a few days, and count the hours saved. It is satisfying to watch. It is also almost always about the back office.

This guide asks a different question for accounting firms. Not "how much work can we automate?" but "what could we offer clients that we could not offer a year ago?"

It is the same process we use with any firm, laid out step by step so you can run it yourself.

Why accounting firms need this question

Most accounting firms have already adopted AI in some form: receipt capture, bank reconciliation, drafted emails, document summaries. Staff are faster than they were two years ago.

But clients still choose and leave accountants mostly on price, because from the outside, the service looks the same everywhere. Automation made the work cheaper for every firm at once. It did not make any one firm different. We explain the pattern in everyone uses AI, almost nobody profits.

Step 1: Finish the floor

Before building anything new, make sure the basics are done. Automation is the floor, not the edge, but you should be standing on it:

• Document collection and reminders that do not depend on someone remembering.
• Bookkeeping categorisation that people review rather than do by hand.
• Standard client emails and summaries drafted by AI and checked by staff.

None of this wins new clients. All of it frees time and budget for what comes next. If your firm has not done it yet, start there, probably with an automation specialist. Our comparison with automation agencies explains when that is the right call.

Step 2: Ask the people who talk to clients

The best product ideas rarely come from the partners' office. They come from the staff who deal with clients every day. We share the exact questions we ask frontline teams. For an accounting firm, focus on these:

• Which questions do clients ask over and over?
• Where do clients get stuck, frustrated or late?
• When do clients only call after something has already gone wrong?
• Which clients would pay for advice, if it were easier to get?

Write the answers down word for word. Each recurring answer is a possible product.

Step 3: Turn answers into product ideas

In accounting firms, the answers usually point to three kinds of client-facing products. We describe them in more detail on our page for accounting firms.

1. An "ask your numbers" portal. Clients log in and ask plain questions about their own figures: why margin dropped, how much tax to set aside, what happens if they hire. Answers come from their real books, and the firm reviews anything that matters.

2. A document and deadline assistant. Every client sees exactly what the firm needs from them, in plain language, with reminders tied to their own deadlines, and can ask what a form means outside office hours.

3. A "what if" planner. Clients test decisions before they make them, such as hiring, raising prices or taking a loan, and see the effect on cash and tax using their own numbers.

Your own frontline answers may point somewhere else entirely. Follow them.

Step 4: Score the ideas and pick one

Three ideas is two too many for a first product. Score each one on four questions:

1. Do clients feel it regularly, or only once a year?
2. Is it built on what makes your firm different, such as your advice and judgement?
3. Could a competitor buy the same thing off the shelf? If yes, it is floor, not edge.
4. Could it lead to new revenue, or does it mostly save time?

Here is how the three ideas above typically compare:

On these questions, the document assistant is the most practical but mostly saves time, and similar tools already exist. The what-if planner is valuable but used rarely. The "ask your numbers" portal usually scores best: clients use it often, it is built on the firm's judgement, and it opens advisory work clients pay more for.

Your scores may differ. That is the point of scoring. Pick the winner, and keep the others as possible next steps.

Step 5: Build a prototype before anything else

Do not start with a specification. Start with something people can click. A good first prototype for a client portal would:

• Use a few anonymised client files, never live data.
• Let a test client log in and ask the real questions from Step 2.
• Show where an accountant reviews an answer before the client sees it.
• Show partners what they would see: which clients ask what, and where advice opportunities appear.

You can build a rough version yourself with tools like Lovable. The goal is to learn, cheaply, whether clients would use it and whether the partners believe in it. We explain why in prototype instead of scope.

Step 6: Know what the real build involves

If the prototype works, the real work starts, and most of it is not the AI:

• Connecting to your accounting software so answers use live figures.
• Making sure each client only ever sees their own data. This is where self-built tools most often go wrong. See our security checklist.
• Review flows for accountants that are faster than answering by email.
• Clear limits, so the product explains and prepares, and advice stays with people.

This is the 70% problem: the demo is the easy part.

What changes if it works

1. Clients have a reason to log in regularly, not only at year end.
2. Advisory conversations start from the client's own questions, which makes them easier to sell.
3. The firm owns a product, which matters if the partners ever sell. See how a software product changes what your firm is worth.

And when a client compares you with the firm down the road, there is finally something to compare.

Where we come in

This is the work we do. We run these steps with your firm, decide with you which product would actually move your position, and build a working prototype before any money changes hands. If it lands, one fixed fee covers the build, and after that we are paid on what the product earns.