How US Small Businesses Use AI to Collect Unpaid Invoices (and Where to Keep a Human)

AI can draft invoices, check billing data and send reminders in seconds. Here is how US small businesses use it to get paid, and the guardrails that stop an AI from giving your money away.

Maria runs a pump-and-valve supplier in Austin, Texas. Her customers love her parts. Paying for them is another story: $186,000 outstanding, some of it three months late.

So Maria did what a growing number of owners are trying. She asked AI for help. It drafted clean invoices, kept her records straight, checked every billing line and wrote polite reminders faster than she could read them. Customers in Dallas, Houston and El Paso paid.

Then she let it run the whole collection process unsupervised and went fishing. Her biggest customer, a distributor in Monterrey, had an AI assistant too. The two machines got along beautifully. Forty-eight friendly emails later, Maria's assistant had agreed to a 20% "quality discount", 180-day payment terms and free shipping on the next order. Nobody had complained about quality. Not once in four months.

Maria is fictional. The risk is not. This guide explains how small businesses in the United States can use AI to collect unpaid invoices faster, what public data says about how AI is actually used for billing work, and the guardrails that keep a human on the money.

This article is general information, not legal advice.

What AI is genuinely good at in invoicing and collections

Most unpaid invoices are not fraud. They are friction: a wrong PO number, an invoice sent to the wrong inbox, a statement nobody reconciled, a reminder nobody sent. That friction is exactly where AI assistants shine.

Anthropic publishes the Anthropic Economic Index, an open dataset that matches anonymised Claude conversations to the job tasks they resemble. It measures observed AI usage by task, not who the users are, and it cannot show trends over time. Read with that framing, the May 2026 data is still revealing about billing work. Looking at tasks commonly done by billing and posting clerks:

• Preparing invoices and itemised statements: in 53% of these conversations the person stays actively involved, shaping and correcting the draft.
• Keeping records of invoices: 77% of these conversations hand the task over to the AI.
• Verifying billing data and fixing errors: 62% of these conversations hand the task over.
• Emails and messages: conversations in the United States produce an email or message 5.9% of the time, against 4.6% worldwide.

The pattern is intuitive. People are comfortable letting AI do the clerical work — logging, checking, formatting — and prefer to stay involved when the output goes to a customer. That instinct is the right one.

In practice, US small businesses use AI assistants to:

1. Draft and check invoices against orders, delivery notes and contract terms before they go out.
2. Reconcile payments with open invoices and flag short payments or duplicates.
3. Write reminder sequences in the right tone for each customer, from a friendly nudge to a firm final notice.
4. Prioritise the aged receivables list by amount, age and customer history, so the owner calls the right three accounts first.
5. Summarise disputes by pulling the relevant emails, delivery records and contract clauses into one page.
6. Draft in other languages when customers are in Mexico, Canada or overseas.

Done well, this turns collections from a monthly panic into a steady weekly routine.

The new risk: when your AI negotiates with their AI

Here is what Maria learned the hard way. AI assistants are built to be helpful and agreeable. That is a virtue when drafting an email. It is a liability when the other side asks for money.

More and more accounts-payable departments use AI assistants to answer supplier emails. When your assistant meets theirs, you can get a perfectly courteous conversation in which every concession flows one way:

• a "quality discount" granted without any evidence of a quality problem;
• payment terms stretched from 30 to 90 or 180 days "given the circumstances";
• late fees waived "as a goodwill gesture";
• credits against future orders that nobody approved.

Every one of those messages can look reasonable in isolation. Together, they quietly rewrite your contract. The fix is not to stop using AI. It is to decide, in advance, what the AI may never do on its own.

Seven guardrails for using AI in collections

1. No concessions without a human. Tell your assistant, in writing, that it may never agree to discounts, credits, extended terms, waived fees or write-offs. It can acknowledge a request and pass it on; it cannot say yes.
2. Put an approval gate on anything that changes money. Amounts, due dates, payment plans and settlement offers go to a named person before they leave the building.
3. Evidence before credits. A quality claim raised months after delivery deserves a close look. Ask the AI to pull the signed delivery receipts, inspection dates and the complaint history before anyone discusses a credit.
4. Set escalation triggers. Hand the account to a person, and often to a professional collector, when an invoice passes 60 days, a customer goes silent, a dispute appears, the debtor is abroad, or the amount is large enough to hurt.
5. Keep an audit trail. Every message your assistant sends should be logged and reviewable. If the conversation ever reaches a lawyer or a court, you will want the full thread.
6. Know what interest you can charge. Start with your contract. If it sets a late-payment rate, that normally governs, subject to state usury limits. If it is silent, state law may still help: in Texas, for example, Finance Code §302.002 allows legal interest of 6% a year starting on the 30th day after the amount is due. Rules differ by state, so check before your AI adds a rate to a reminder.
7. Protect customer data. Use tools approved by your business, and do not paste bank details, contracts or personal information into assistants that are not covered by your data policies.

A useful rule of thumb from the Index's own vocabulary: let AI take over the paperwork, but stay involved in anything a customer will read and anything that changes the amount you are owed.

A simple workflow: what AI does, what people do

Cross-border invoices need a different playbook

AI writes excellent Spanish, French or German. Language is no longer the obstacle when your customer is abroad. Leverage is.

Take Mexico, the most common cross-border market for many Texas businesses. Mexican commercial law gives creditors a powerful tool: a properly drawn pagaré (promissory note) allows a fast-track juicio ejecutivo mercantil, in which the debtor's assets can be attached at the start of the case rather than at the end. The note must meet formal requirements, including the word "pagaré" in the document, an unconditional promise to pay a set amount and an original signature, and the fast-track action must be brought within three years of the note's due date. Without a pagaré, claims go through ordinary or oral commercial proceedings, which take longer.

Two other points matter for US creditors. A judgment from a US court is not automatically enforceable in Mexico; it must go through a recognition process first, which is why pursuing the debt locally is often faster. And evidence wins: signed delivery receipts, purchase orders and a clear email trail are what turn a stalled account into a paid one, exactly as they did for Maria.

A note on US compliance: the Fair Debt Collection Practices Act protects consumers, and business-to-business debts generally fall outside it. That does not make aggressive messages safe. Reminders that threaten actions you cannot or will not take can still create legal risk, so keep your AI's tone firm, factual and honest. Time limits also vary by state; in Texas, most debt claims must be brought within four years.

When to bring in a collection agency

AI will keep your small, friendly, domestic invoices moving. Bring in a professional when:

• the customer has stopped replying, or replies only with new questions;
• a dispute appears late, conveniently, or without evidence;
• the debtor is in another country;
• the amount is material to your cash flow;
• your own AI, or theirs, has started negotiating.

How INTERCOL helps US businesses get paid

INTERCOL is an in-house international collections team for businesses owed money across borders, with no brokers and no subcontracted networks. For US companies we act as the human on the money: we read the whole thread, investigate the debtor, gather the delivery evidence, withdraw concessions nobody authorised, and speak to your customer's finance director directly, in Spanish, English and other languages. When a debtor needs more than a conversation, we move to legal recovery.

We work on a no cure, no pay basis: if we do not recover, you do not pay a commission. For wider strategy, see our guide to collecting from overseas business clients or our United States page. Tell us about your case and we will assess it free of charge.

Frequently asked questions

Can AI collect unpaid invoices on its own?

AI can handle much of the routine work, such as drafting invoices, checking billing data, reconciling payments and sending reminders. It should not negotiate on its own: discounts, credits, extended terms and settlements need a human decision, because AI assistants are designed to be agreeable.

What does the Anthropic Economic Index say about AI and billing work?

Looking at tasks commonly done by billing and posting clerks (May 2026 data), 77% of conversations about keeping invoice records and 62% about verifying billing data hand the task over to the AI, while people stay actively involved in 53% of conversations about preparing invoices. The Index measures observed Claude usage by task, not who the users are.

What should I do if a customer's AI asks for a discount?

Treat it like any other request: acknowledge it, ask for evidence, and route it to a person with authority. Check delivery receipts, inspection dates and the complaint history before agreeing to anything. Late, unsupported quality claims are a common delaying tactic.

Can I charge interest on overdue business invoices in the US?

Usually, yes. Your contract's late-payment clause normally applies, within state usury limits. Where the contract is silent, some states provide a default rate; Texas, for example, allows 6% a year from the 30th day after the due date. Rules vary, so check your state's law.

How do I collect from a customer in Mexico?

Start with evidence: signed delivery receipts, purchase orders and the email trail. If you hold a valid pagaré, a fast-track commercial proceeding may be available, with attachment of assets at the outset, if brought within three years of the note's due date. A US judgment must be recognised in Mexico before enforcement, so local action is often faster. A collector with in-house Spanish-speaking staff can usually resolve the account before court is needed.

---

Marcus Ellerton writes for INTERCOL on cross-border commercial recovery. Data: Anthropic Economic Index (May 2026 release), observed Claude usage by task, CC BY 4.0 — anthropic.com/economic-index. All characters in the film are fictional.