Where should a small bookkeeping firm start with AI?

Why start with a picture instead of a purchase?

The most common AI failure in a bookkeeping firm makes no noise: a tool bought on the enthusiasm of a demo, half-configured on a Friday afternoon, then left in a browser tab while the subscription keeps billing. The tool was not the problem; the missing picture was. A small firm that does not yet know where its non-billable hours go cannot know which tool will bring them back, and the market will happily sell it anything in the meantime. So the picture comes before the spending: where the time goes, what is already paid for, what deserves automating first. That work requires no software and no credit card; it requires an hour of honesty about a normal production week, and it completely changes the quality of the decision that follows it.

How do you inventory what eats the hours?

The inventory is taken on a normal week, not an ideal one. The questions are simple: how much time goes to document follow-ups with clients who do not answer? To entering documents that always travel the same path? To sorting and filing? To deadline tracking held in someone’s head or in a fragile spreadsheet? Non-billable hours are the thread to pull: that is where the work hides that requires no professional judgment yet still exhausts the team. Tax season is worth including in the exercise: whatever overflows during the busiest weeks is almost always the thing that was already fragile the rest of the year, only larger. At the end, the firm holds a short, ranked list; the first item on that list is, most of the time, the answer to the question “where do we start”.

What does the accounting software already do?

The accounting software a firm already pays for every month does more than most firms use. QuickBooks Online and Xero include receipt recognition, bank rules and basic reminders; turning those on costs nothing and sometimes solves part of the problem before any purchase. The curation rule applies here in full: when the capability already included in your accounting software does the job well, use it, and pay no one. Checking this before spending avoids the market’s most embarrassing scenario: paying a second subscription for a feature already included in the first. What remains after that check is AI’s real territory: drafting, summarizing and explaining with an assistant like Claude, and automations wired into the firm’s actual workflows, like the follow-ups that chase client documents on their own.

How do you pick the first quick win?

The first quick win is chosen with a double criterion: frequent and mechanical. Work that returns every week and always follows the same rules lends itself to automation; work that is rare or full of judgment is not a starting point. In most small firms that intersection points at document follow-ups, data entry or sorting. Discipline matters more than ambition: one workflow, set up completely, measured, proven on real client files, before anything gets extended. One proven workflow builds the team’s confidence and justifies the next one; three open projects at once reproduce exactly the failure of the tool left in a tab, at a higher price. Once the first win has held for a few months, the firm adds one automation at a time, at the rhythm of its own season, never during the busiest weeks.

What is the 30-minute call for?

The picture described above is exactly what Agentica delivers in a free 30-minute video call, with no system access and no obligation: the firm’s real numbers, a neutral verdict on what is worth doing, what the software in place already covers, what to ignore, including what not to buy. The firm leaves with a written plan it keeps, whether or not it ever hires us; if the next step makes sense, a one-page proposal arrives within 48 hours, with a fixed price quoted before any work starts. Installations run from May to November, never during tax season, so the system is broken in before the next busy stretch. Starting with the picture costs half an hour; starting with a purchase usually costs one more dead subscription.