"Would I Trust It? No. That's My Job." — What It Would Actually Take for a Bookkeeper to Trust AI

August 7, 2026

---

This week we asked a veteran bookkeeper — decades in practice, every client on QuickBooks Desktop — whether she would trust software that auto-categorizes transactions. Her answer arrived in under a second: no. We build AI bookkeeping software, so we want to take her answer seriously instead of arguing with it. It turns out she wasn't rejecting automation. She was pricing it.

Why don't bookkeepers trust auto-categorization?

Because it has failed them, personally and repeatedly. Her words: it does it right now, "and it's almost always wrong." That's not a prejudice — it's a track record. Every bookkeeper using bank feeds has watched software confidently file a loan payment as an expense or a transfer as income. Distrust earned from experience doesn't yield to a marketing claim. It yields only to a different experience.

Would higher accuracy change her mind?

We asked exactly that: what about 95, even 99 percent accurate? Still no — "that's my job." And she's right. Her name is on the books; an accuracy statistic doesn't transfer her liability. A tool that's right 99 times in 100 but can't tell her which time it was wrong hasn't reduced her review burden at all. Every row still needs her eyes, so the promised accuracy saved her nothing.

So what's the real test?

Time. Her most damning observation wasn't about correctness — it was about the review flow: going through the software's pending suggestions and fixing them takes "almost as much time as if I had manually entered it myself." That's the whole ballgame. Automation that requires full re-verification is a net zero. The only automation worth adopting is the kind whose output can be checked faster than it could be redone — which means every suggestion has to arrive carrying its own evidence.

What would software have to prove before she'd switch?

She named three things, in order. Security first: bookkeepers are covered by federal safeguards rules, and a data breach through a vendor they chose is a liability with their name on it — a tool must prove it meets that bar before features matter. Flow second: pages that reset, windows that vanish, and back buttons that overshoot cost more than any feature saves. Price third: if it isn't cheaper than the incumbent, "people are going to stick with what they know."

Why does this matter right now?

Because the ground is shifting under the tools bookkeepers actually prefer. The affordable desktop software many practices run on is being retired on a known clock, and the replacement paths cost several times more per year. A generation of careful practitioners is going to be forced to choose new software whether they want to or not — and the incumbent's own answer is the product they already tried and disliked.

What did we take away, building the thing she doubts?

That her "no" is the specification. Don't ask for trust — make it unnecessary: show the evidence behind every decision, route anything uncertain to a human instead of guessing, and let her measure the tool against her own judgment before it touches anything. The bookkeeper who says "that's my job" is not the obstacle to good bookkeeping software. She's the quality bar for it.

Where Nalo fits, stated plainly

Nalo is built for her test: every categorization shows its reasoning and evidence, anything uncertain goes to review instead of the books, nothing posts without confirmation, and a month's close states exactly what was verified and what wasn't. Two-factor authentication and encrypted, isolated client data are already live — the security bar comes first there too. See how it works at nalo.app.