Why AI developers need real-world business data, and what it means for accounting firms

If you run an accounting or CPA firm, you may wonder why anyone building artificial intelligence would care about your old engagement letters, checklists and internal procedures. The short answer is that AI tools learn from examples, and the best examples of how professional work is really done are the records of firms that do it. This article explains that idea in plain words, without hype, and what it could mean for a firm like yours.

AI tools learn from examples

An AI model is not programmed with a rulebook. It is shown a very large number of examples and learns patterns from them. If you want a tool that can help draft a client letter, summarise a set of review notes or follow a month-end close routine, it helps to show it examples of those things being done properly.

Developers refer to the material they learn from as training data. They also need material to test a tool afterwards, to see whether it behaves sensibly on work it has not seen before. Both uses depend on examples, and the quality of the examples matters.

Why public material is not always enough

Much of what is publicly available is general. It can tell a tool what a bank reconciliation is. It tells it much less about how a real firm sets one up, who reviews it, what gets queried and how the answer is recorded. As we understand it, that day-to-day detail of professional work lives inside firms, in their internal documents, rather than on public websites.

That is why some AI developers look for real-world business data: records that show how work is genuinely carried out, including the corrections, follow-up questions and exceptions that tidy public examples leave out.

We do not speak for any buyer, and each one has its own needs. Treat this as a general picture, not a statement about a particular company.

Three things real records show that tidy examples do not

When people describe how work is done, they usually describe the clean version. Real records tend to show three things that are easy to leave out:

1. Review and correction. A checklist with a reviewer’s notes shows what was queried, what was changed and who signed off.
2. Exceptions. Real engagements rarely follow the template exactly. Records show where and why the standard approach was adjusted.
3. A firm’s own wording. Standard letters, templates and explanations reflect the plain language a firm actually uses with its clients.

These are the details that make examples realistic, and they are also the details that make careful scoping and privacy review so important.

Why accounting and CPA firms are interesting

Accounting is structured, rule-based and well documented. Over the years a firm builds up material that records how its work runs:

• Engagement letters, templates and standard wording
• Workpaper and review checklists
• Standard operating procedures and an internal knowledge base
• Month-end close and reconciliation procedures
• Tax-season workflows and planning notes
• Training and onboarding material
• Helpdesk or support histories

Work also repeats in cycles, such as the monthly close or the busy season, so the same kinds of task appear many times, handled by different people. Examples like that can be useful for a tool that is meant to help with similar work.

Notice what is not on that list: your clients’ financial detail. The interest is in how the firm operates, not in who the clients are. We say more on that below, and in our article on how records are anonymized and reviewed.

Common misunderstandings

• “They want my clients’ financial statements.” The interest described here is in how a firm operates, such as its procedures, templates and checklists. What your firm includes is your decision, and client-confidential material is your firm’s call.
• “I would be handing over my files.” A licence covers a copy. Your originals stay with you.
• “Only big firms count.” Buyers set their own size guidelines, and they change. We currently look for about 15 or more full-time employees at an accounting firm, and we will tell you plainly if your firm is under that.

Why age matters

Older records are often the most useful. A long retained history shows how procedures changed, how problems were resolved and how the firm’s practice developed. That is one reason buyers tend to ask about firms with several years of retained records, and it is why our fit check asks how many years of usable records you keep.

What this does not mean

It is easy to read too much into a story like this, so here are the limits:

• It does not mean your records have a value. Whether they are of interest depends on the buyer’s own review, and we do not estimate or promise any figure.
• It does not mean a buyer will accept your firm. Every business goes through the buyer’s review, and a buyer may decline.
• It does not mean your clients’ information is on offer. Buyers say personal and sensitive information is removed before their team works with any material, and your firm decides what is in scope. Confidentiality is a serious question for your own advisor.

What a firm actually keeps

When a firm licenses records, it licenses a copy. Your originals stay in your own systems and you carry on using them. The terms are set out in a written agreement, and you should read it closely and show it to your own advisor before you sign.

The arrangement is also voluntary at every step. You can look, ask questions and decide not to apply, and that is a perfectly good outcome.

How a firm can find out where it stands

Buyers set their own criteria, such as firm size, location and the language of the records, and these change over time. A quick, free way to see where your firm stands today is our fit check. It runs in your browser, takes a few minutes and shows whether a free review looks worthwhile. For a fuller list of what we look at, see our guide to how AI training data collection works and what buyers look for.

A note on how we are paid

Thomas & Finch is an independent referral service. Your firm never pays us. If a firm applies through our referral link and the buyer accepts it, the buyer pays us, and if it is not accepted we are not paid. We are not affiliated with any buyer.