Automating bank reconciliation in an accounting practice

By Jérôme Denis, AI & data governance consultant · 7 min read · 12 August 2026

Bank reconciliation means pairing every line on a bank statement with the document that justifies it. Nobody in an accounting practice finds this interesting, and it consumes an unreasonable share of the month, mostly because the documents are missing and getting them requires asking a busy person several times. That last sentence is the whole problem, and it is where automation actually helps.

Why this is worth automating now

If your practice works with French entities there is a deadline attached. Electronic invoicing becomes mandatory to receive on 1 September 2026, and mandatory to issue for small and medium companies on 1 September 2027. Documents that used to arrive as photographs in a WhatsApp message will arrive as structured files with a status. A practice still collecting paper at month end will find the new flow arriving faster than it can be processed manually. Reorganising the reconciliation process at the same time as the invoicing change avoids doing the work twice.

There is a data protection angle too. Bank feeds mean a third-party provider holds credentials or tokens against client accounts. In the EU this runs on the regulated open banking interfaces created by the payment services directive, so access is consent-based, scoped and revocable, rather than someone storing a password. That is a better arrangement than the alternative it replaced, which was often a shared login on a sticky note.

How the mechanism works

Two things run in sequence. First, a direct connection between the client's bank accounts and the practice's production software pulls transactions daily, without anyone downloading a statement. Then matching rules pair each line with a scanned sales or purchase invoice, using amount, date and counterparty name. Where the match is unambiguous, the software reconciles it and moves on. On a well-maintained file this covers most of the volume.

What matters is the residue. Traditionally a member of staff listed the unmatched lines at month end and wrote one email asking for everything. The client read it, found some of the receipts, sent them loose, and the staff member reopened the file three times. Automated chasing changes the shape of that exchange: the client gets a notification about one specific transaction, on their phone, with a link to photograph or upload the document, and the file is matched on arrival. Same total work for the client, spread across the month in ten-second pieces instead of one dreaded hour.

The same file, two ways

Take a services company with roughly 150 bank transactions a month.

Manually: staff wait for the statements and the invoices at month end. About three hours goes on ticking lines, identifying the twenty-five missing documents and writing a summary chase. The client replies over several days with documents in no particular order, so the file is reopened repeatedly before the entries can be finished.

Automated: the bank is connected live. Every Tuesday the software processes the week's transactions and reconciles around 120 lines on its own. For the remaining thirty it messages the owner with the list and a button per item. The owner clears them from their phone across a few minutes. By Friday the file is 95% current, and the monthly handling drops from three hours to about fifteen minutes.

The figures come from a file that behaves. A client with messy records will not produce them in the first quarter, which is a point worth making before you quote the saving to a partner.

Where it breaks

Character recognition is imperfect. A creased receipt photographed at an angle produces a wrong amount, and the matching engine will confidently pair it with something. Two transactions of similar value a few days apart produce a false match that only a person notices. These are not reasons to avoid automation. They are reasons the reviewer role does not disappear, and firms that removed it discovered the cost at the year-end review.

Client tolerance is the other limit. Notifications set to fire aggressively read as nagging, and the relationship pays for it. Tune the frequency down until someone complains it is too quiet, rather than up until someone complains it is too loud.

The broader point is that the tool removes the repetitive part and leaves the judgement. Staff shift from ticking lines to investigating the anomalies that the ticking used to hide, which is both more useful and more interesting, and is the argument that persuades the team rather than the partners.

Rolling it out

1. Audit the tooling and the client base

   Which clients bank where, which use accounting software you can connect to, and which are comfortable with a phone-based flow. This shapes the pilot list.

2. Pilot with a small volunteer group

   Ten to fifteen technically comfortable clients. Use them to validate the bank connectors and tune the matching rules before the rules meet the whole portfolio.

3. Train the team properly

   Staff need to run the tool and to walk a client through it on the phone. The second skill is the one that gets skipped and the one that generates support load when it is missing.

4. Explain the benefit to the client, not the technology

   What the client hears should be that they will not get a stressful year-end chase and that their figures are current enough to steer by. Nobody adopts a tool because it has a good matching engine.