Every small business has this job: invoices arrive by email, someone saves them somewhere, and at month end someone else works out which ones are missing. It is unskilled, unavoidable and nobody wants it. We automated ours, and the same pattern fits most businesses.
The tool we built for our own bookkeeping: it reads supplier invoices out of the inbox, files them in the right place in the accounts drive, and reconciles them against the bank statement export when one is dropped in.
Invoices and receipts are found by what they are, not by someone remembering to forward them.
Straight into the shared accounts drive, in the structure the bookkeeping actually uses.
The useful half: not just filing documents, but knowing which payments have no invoice behind them.
An automation that files things into your accounts needs to fail safe, not fail quietly.
Most of the value in a system like this comes from the systems it talks to, and most of the risk does too. Each connection is scoped to the access it needs for the job.
The path a piece of work takes through the system, start to finish.
The job runs on a timer against the accounts inbox. No dashboard to check and no service to log into. Reconciliation runs against the bank statement export whenever a fresh one is dropped in.
New attachments are pulled from the inbox and classified by sender, subject and document contents into invoices, receipts and everything else.
Each document is fingerprinted so duplicates are recognised regardless of filename, then filed into the accounts drive under a consistent name.
Filed documents are matched against the bank statement export, and anything unmatched in either direction is reported.
The problems that decided how this was built. They are the reason it works the way it does rather than the obvious way.
An automation with write access to the accounts drive can make a mess faster than a person can. A wrong classification rule, run unattended for a month, is a genuine cleanup job.
The tool defaults to a dry run: it reports every action it would take and changes nothing unless explicitly told to act. Classification rules, the storage layer and the full pipeline each have tests, so a rule change is verified before it ever runs against the real drive.
An automation that can be left to run, because every rule change is proved in dry run before the scheduled job is allowed to act.
We don’t publish screenshots, logos or client names for this work. Much of it runs inside a business and handles sensitive data, so we describe what was built rather than who it was built for. We’re happy to walk through the detail on a call.
Every build on this page started as a workflow that off-the-shelf software nearly handled. Tell us where the gap is and we’ll scope what it takes to close it.