Build an AI Agent That Chases Your Late Payments For You

You did the work. You sent the invoice. Now you are doing unpaid emotional labor, rehearsing the same awkward email in your head every Sunday night.

It is 7:40 a.m. in Manchester. Priya Shah unlocks the door of her nine-person interior design studio, puts the kettle on, and opens her laptop to the same three names she has been avoiding for weeks. Three clients. Eleven thousand, four hundred pounds. All of it for work finished before midsummer.

She does not chase it today. Today there is a client presentation, two supplier calls, and a hiring interview. The invoice can wait. It always can, until the day it cannot.

This is not a Priya problem. It is the default state of small business in 2026, and the data says so plainly.

Why "I'll follow up next week" is quietly bankrupting small businesses

Chasing a client for money you already earned is one of the most uncomfortable parts of running a business. So owners delay it. The delay has a price tag.

Roughly half of all B2B invoices, across the US, UK, and Western Europe, get paid after their due date most years, per the Atradius Payment Practices Barometer. That is not a rare bad client. That is the market. Small businesses with fewer than 50 people report late payment on close to one in three invoices they issue. Construction and professional-services firms, the trades and studios that bill on milestones and retainers, run the worst numbers: Days Sales Outstanding of 45 to 65 days is common where the invoice terms say 30.

None of this means your clients are dishonest. It means nobody in the chain has a system, and a missing system always gets filled in by whoever feels the discomfort first. That is usually you, the owner, at 11 p.m., trying to word a reminder so it does not sound rude.

A chatbot writes an email. An agent chases the money.

Ask ChatGPT or Claude to "write a polite payment reminder" and it will, in about four seconds. That is not the hard part and never was. The hard part is remembering which of your forty open invoices crossed day seven this morning, which one is now at day thirty and needs a different tone entirely, and which one quietly slid past day ninety while you were busy running the business.

A chatbot answers a question you remembered to ask. An agent does not wait to be asked. It reads your invoice list on a schedule, works out how overdue each one actually is, decides which message fits that specific day, drafts it in your voice, and puts it in front of you for one click. The shift is not "AI can write emails now." It is "AI can hold the calendar you keep forgetting to open."

What changed: why this is buildable this month, not someday

Two builders have already shipped versions of exactly this agent in public. Rahul Joshi, who teaches business automation on YouTube, built an accounts-receivable recovery agent in n8n that reads unpaid invoices from Google Sheets every day, scores each one on a 1-to-4 urgency scale based on days overdue and reminders already sent, drafts a personalised email with GPT-4o-mini, sends it through Gmail, escalates the riskiest accounts to a legal or collections contact automatically, and logs every action to a recovery sheet with a Slack alert if anything breaks. A separate builder publishing as Automated LLC wired the same idea with a human-in-the-loop checkpoint: the agent drafts the follow-up, but nothing goes out until the owner approves it with one click.

What made this suddenly accessible is not a single model release. It is that agent tools like Claude's Cowork mode, and no-code builders like n8n, can now sit inside a folder of your actual business files, read a spreadsheet, check a condition, and produce a finished draft rather than a generic answer. You are not hiring a developer to write custom software. You are pointing an agent at the sheet you already keep and telling it what outcome you want.

Isn't this just automated spam that annoys good clients?

It is the opposite, if you build it with a ladder instead of a loop. A bad system emails everyone the same "just following up!" note every three days until someone snaps. A good system matches the tone and the channel to how overdue the invoice actually is, and it never fires without your say-so. That second point matters most: every credible build of this agent keeps a human approving the send, especially once a message escalates past a friendly reminder. The agent's job is memory and drafting. Your job stays judgment.

How to build your own invoice-chasing agent, step by step

1. Put every unpaid invoice in one place. A Google Sheet with client, amount, due date, and contact is enough. This single sheet is what your agent will read every morning, so it needs to be the real source of truth, not one of three spreadsheets you half-update.
2. Define your ladder before you automate anything. Business attorney and YouTube creator behind Laws For Wealth lays out a cadence worth copying directly: day seven, a friendly email that assumes the invoice fell through the cracks; day twenty-one, a firmer email restating the terms already agreed; day thirty, it stops being an email and becomes a printed letter, because a physical letter signals this has left the "we'll get to it" pile; day sixty, a stop-work notice if your contract allows it; day ninety, a final demand sent certified, because now you are building a record.
3. Give the agent your contract, not just your invoice list. Before it drafts a day-sixty stop-work notice, it should check whether your contract actually gives you that right. If it does not, the agent should say so plainly instead of drafting a threat you cannot back up.
4. Run it daily and let it score urgency. Following Rahul Joshi's build, have the agent calculate days overdue for every invoice each morning and assign an urgency level, so a seven-day invoice and a seventy-day invoice never get the same tone by accident.
5. Keep a human-in-the-loop checkpoint. The agent drafts. It does not send. Automated LLC's version holds every follow-up for one-click approval, which keeps you in control of the relationship without making you responsible for remembering the schedule.
6. Auto-escalate the accounts that need a human anyway. Past day sixty or ninety, route the account to whoever handles legal or collections in your business, and log every message sent to a simple audit sheet so you can prove, if it ever comes to that, exactly what was said and when.
7. Add a Slack or email alert for the exceptions. If the agent finds nothing to chase that day, or if a connection breaks, it should say so, not go silent. Silence is exactly the failure mode you built this to avoid.

The one sentence missing from your contract that makes the whole system toothless

An agent can draft the most polished final demand in the world, and it will still be just a polite email without the right clause underneath it. Most contracts state when payment is due. Almost none state what happens when it is not. The clause worth adding, according to the same attorney-led breakdown, is a simple one: that the client agrees to pay all costs of collection, including attorney's fees, if the invoice goes unpaid, whether or not a claim is ever formally filed. Without that sentence, winning a dispute can still mean losing money, because you eat your own legal costs even after you are proven right. A three-thousand-dollar invoice dispute becomes not worth pursuing. Add the clause once, in every contract going forward, and every stage of your ladder gets real weight behind it. Contract law is set at the state or national level, so have a local attorney check the wording before you roll it into your own agreements.

What changes when the machine remembers what you forget

A week in, nothing looks different. The agent is quietly reading your sheet every morning, drafting the day-seven notes you would have sent late anyway.

A month in, the pile of invoices sitting past thirty days is smaller than it has been in a year, because nothing slipped through while you were focused on the actual work.

A quarter in, the businesses using structured early follow-up recover more than eighty percent of what they are owed inside the first thirty days, according to commercial credit-industry benchmarks, against recovery rates below thirty percent once an invoice drifts past ninety. The agent did not change how much your clients owe you. It changed how much of it you actually collect.

Same invoices. Same clients. A completely different bank balance.