AI Shortlist · Guides
"AI bookkeeping" is sold as one thing. It's really three, and they have very different payback.
1. Capture. Photograph a receipt, get structured data: vendor, date, amount, tax. This is a solved problem and it is usually free — most business cards and banks now bundle it, and standalone receipt apps start at a few dollars a month.
2. Categorisation. Mapping that data to your chart of accounts. Cheap to buy, but only pays back if you review it monthly; unreviewed auto-categorisation is just a faster way to be wrong.
3. Reconciliation. Matching invoices to payments and chasing the mismatches. This is where the real hours live — and the only layer worth a real subscription.
| Task | Typical monthly cost of doing it by hand | AI payback |
|---|---|---|
| Typing receipts | 1–3 h | Small — and mostly free already |
| Categorising + coding | 2–5 h | Medium, only with monthly review |
| Chasing mismatches, late invoices | 3–8 h | High — this is the real target |
• Does it connect to your bank and your accounting software? If you still export CSVs by hand, you bought a prettier spreadsheet.
• Who owns the data and can you export it? Your books must survive a switch.
• How does it handle tax and VAT lines? This is where generic tools quietly fail.
• What happens on an unusual transaction? Good tools flag it instead of guessing.
If the tool cannot tell you, in one number, how many hours it saved last month — and you cannot reconstruct that number yourself — cancel it. In this category, "we'll figure out the value later" almost always means zero.
Prices and bundles move fast in this category. Treat every range above as a shape, and check the vendor's current terms before you pay.