Every Thursday, bank lines the bookkeeping software could not code are grouped per client, and AI suggests a likely account from each client's history. Clients answer one WhatsApp message with tap-to-answer options, and their replies post straight back to the ledger.
Typical results
≈6 hrs/wk
Bookkeeper time saved chasing explanations
≈2 days
Typical client reply time
≈80%
Answers posted without bookkeeper review
Figures reflect typical results for this kind of automation.
The challenge
Each month, hundreds of bank feed lines arrive with vague descriptions, such as a card payment or a transfer with no reference. Bookkeepers email clients long spreadsheets of questions, answers trickle back weeks later, and month-end closes with a pile of suspense items that someone has to chase by phone before the VAT return is due.
Before and after
Open bank items at month-end, average per client
−60%Days to get a client's answer
−86%Bookkeeper hours per week chasing explanations
−75%The automation
Every Thursday, a Make scenario pulls the bank lines that each client's coding rules left uncoded. OpenAI compares each line with that client's past coding and suggests the most likely account. The client receives one WhatsApp message listing the open items, each with buttons such as stock purchase, owner drawing or other. Replies post the coding back to the ledger, while free-text answers and suggestions below 80% confidence go to the bookkeeper's review queue.
How it works
- 01
Thursday sweep of the bank feed
Lines that the client's coding rules could not match are pulled per client from the bookkeeping software.
- 02
AI suggests an account
Each line is compared with that client's coding history, and a likely account is attached with a confidence score.
- 03
One WhatsApp message per client
The client sees all open items in one message, each with tap-to-answer options and a free-text choice.
- 04
Answers post to the ledger
Confident answers code the line in the ledger automatically and clear it from the open list.
- 05
Bookkeeper reviews the rest
Free-text replies and answers under 80% confidence land in the review queue with the suggestion attached.
Starts from
- Uncoded bank feed lines
- Client coding history
Delivers
- WhatsApp question list
- Coding posted to ledger
- Bookkeeper review queue
The impact
Open bank items at month-end fall by about 60%, because clients answer a short message within days instead of a spreadsheet weeks later. Bookkeepers spend about 6 fewer hours a week chasing explanations and review only the low-confidence answers before the month is closed.
Questions about this automation
Which bookkeeping systems can it work with?
It works with bookkeeping software that exposes bank feed lines and accepts coding through an API, such as QuickBooks or Xero, and with local systems through export and import files. Messages go out from the office's own WhatsApp business number, and Make runs the weekly sweep.
What if a client does not reply?
Unanswered items are repeated in the next Thursday message, and after two weeks without a reply the bookkeeper gets a list for that client to follow up by phone. Items stay on the open list until someone codes them, so nothing quietly drops out of the ledger.
How does the AI learn each client's coding habits?
Before suggesting an account, the AI looks at how similar lines for the same client were coded over the past year, by payee, amount range and description. It suggests only accounts that client already uses, and uncertain answers go to a person. It is a multi-step build, usually rolled out in stages.