Chasing late invoices,
without chasing them yourself.
A working system that follows every invoice from arrival: decides who to chase today, writes each email in the right tone, and understands what the client writes back. Nothing sends without your approval.
I built this to prove a specific thing: that a routine admin job runs better when AI handles the language, plain rules handle the scheduling, and a person keeps the final say.
Status
Running end to end on test data. Everything shown on this page is live and verified.
The problem
Small firms chase late invoices from memory. Larger ones chase through spreadsheets
and inconsistent follow-ups. Either way, the invoices nobody notices in time just
slip.
Among the 22% of businesses that spend staff time chasing, UK government research puts it at 86 hours a year.
Stack
n8n, Airtable, Claude, Gmail.
Role
Built end to end. Architecture, workflows, the decision logic, the interface.
Watch it run, ninety seconds.
From arrival to reply, four steps.
An invoice arrives by email
The system reads the PDF, pulls out the client, the amount, and the due date, and files it. No typing. In this demo invoices come in as email attachments; the same front door can be wired to whatever you already invoice from: Xero, QuickBooks, Sage, FreeAgent.
Every morning, it decides who to chase
Not a reminder rule. The decision logic looks at each invoice: how overdue, how many chases already sent, whether the client has replied, whether they promised a date. It picks one of three answers: chase today, wait until a date, or escalate.
It writes the email, then waits for you
A long-standing client a few days late gets a friendly nudge. One that has ignored three reminders gets a firm letter citing the UK Late Payment Act. No templates. The draft sits in a queue until you read it, edit it if you want, and click send. It goes from your inbox, in your name, with the invoice attached.
It reads the reply
A client answers "payment is scheduled for Friday." The system understands that as a promise, logs it, and stops chasing. Friday passes with no payment, so the next morning it drafts a firmer follow-up. It follows the whole conversation, not just the first email.
One worklist. One approval queue.
Airtable holds every client, invoice, chase, and reply. n8n runs the four workflows between them: reading incoming invoices, deciding and drafting each morning, sending on your approval, and understanding what comes back.
Your day is one screen: every invoice you're owed, with who, how much, when, and what happens next on each row. The morning's drafts wait in a queue. You click through. The system handles the remembering. You keep the final say.
What it does. What it deliberately doesn't.
What it does
- · Keeps every unpaid invoice on one screen, with a next action on each row.
- · Decides who to chase and when, and records why.
- · Writes each chase in a tone matched to the relationship.
- · Reads incoming replies and sorts them into five kinds: payment signal, promise to pay, dispute, question, unrelated.
- · Pauses chasing when a client promises a date, and resumes when the date passes.
- · Logs every decision, so you can see what it did and why.
What it deliberately doesn't
- · Never sends anything on its own. Every email waits for a person.
- · Never marks an invoice paid. A client saying "I paid" flags you to check the bank. It does not close the invoice.
- · Doesn't take payments or handle payment links.
- · Doesn't predict when someone will pay, or score anyone's risk.
The value isn't the email. A free chatbot writes an email.
The value is that nothing gets forgotten, every chase knows the history behind it, and
the system understands what comes back.
What slips when things get busy?
Late invoices are one example. Most businesses have a routine job that gets done properly in a calm week and quietly slips in a busy one. Tell me what yours is and I'll be honest about whether it's worth automating.