Every month, the firm rebuilt a report for each client from scratch, out of whatever had landed in email and WhatsApp: PDFs, photographed receipts, spreadsheets and forwarded documents. It was a monthly scramble, and the senior people spent it collating instead of advising.
The value the firm sold was judgement. What its month actually looked like was collation: chasing documents across inboxes and chat threads, then keying and formatting them into a report, one client at a time, before anyone could even start thinking.
The source material arrived in every shape at once. A receipt as a photo, a statement as a PDF, a figure buried in a forwarded email, a spreadsheet with the columns in the wrong order. Pulling a clean monthly picture out of that by hand was slow and easy to get wrong, and it did not scale as the client list grew.
So the people best placed to give advice spent the month preparing the thing they were meant to advise on, and the work that clients actually paid for waited until the assembly was done.
An AI pipeline wired into the channels the material already arrives on, with a professional signing off.
We wired an AI pipeline into the firm's own email and WhatsApp, so the material is read the moment it lands, in whatever format it lands in, instead of waiting to be gathered up at month end.
The pipeline pulls the relevant figures and details out of PDFs, receipt photos, spreadsheets and forwarded documents, and normalises them into a consistent structure.
It assembles a structured monthly report per client in the firm's own format, so what lands on the reviewer's desk already looks like the firm's work, not a raw data dump.
Because the assembly is automatic, the whole client list is prepared in parallel rather than sequentially by hand, so the month stops being a bottleneck.
Every report is reviewed and signed off by a professional before it goes out. The AI does the collation; the judgement, and the accountability, stay with a person.
The pipeline is tuned for the high accuracy accounting requires, so review means checking sound work, not hunting for the machine's mistakes.
The month freed up, and the seniors went back to advising.
It runs in production today, across dozens of clients, at the high accuracy the work demands. The monthly report is prepared by the pipeline instead of by hand, so the deadline stopped being a scramble.
The team now reviews the reporting instead of assembling it line by line, so the accountants spend their time on judgement and advice, the work clients actually pay for, rather than on gathering and formatting documents.
The point was never to replace the accountant, only the hours of gathering and formatting in front of the accountant. Reading the material where it already arrives, extracting it accurately, and keeping a professional on the sign-off is what makes it safe to run on real client work. See the wider picture on our AI for accounting firms page, or the same intake-and-extraction discipline pointed at a different industry in our workflow-automation work.
A retailer was burning budget on scattered AI projects. We ranked every idea by value and effort, cut dozens to three, and delivered a 90-day roadmap, with a quick win that proved a return in about eight weeks.
Read the case → Workflow automation · PropertyA large Israeli property manager had 8 people re-keying data by hand. We automated it end to end around a 15-year-old ERP. Average handling time fell from 4 hours to 30 seconds.
Read the case →
Reef TRH