"Can AI help me collect late payments?" Yes, in a narrow and useful way. A tool can keep the list of who owes you, draft the polite reminder, and tell you which accounts have gone quiet. What it should not do is decide how hard to push a good customer, agree a payment plan, or say anything on your behalf that you would not say yourself. This article is for the owner of a company with ten to two hundred people who is tired of watching cash sit in other people's accounts, and who wants an honest account of what to hand over, what to keep, and how to start without damaging a relationship that took years to build. Before the chasing comes the invoice itself: can AI do my invoicing covers what a tool prepares and what a person signs off.
I was COO and then President of a public company that closed more than forty acquisitions, and I ran the integration on most of them. In every one of those businesses, the speed of getting paid said something about how the company was run. I am a lawyer by training. Now I write fixed-fee AI plans for CEOs of private-equity-backed and owner-led mid-market companies through 013 Labs. This is not legal advice. Rules about contacting people over money owed, about interest and fees, and about what you may say differ by place and by who the customer is, so where the law could apply, talk to your own counsel.
What a tool does well
Most of collections is not negotiation. It is bookkeeping and follow-up, done on time and in the same way every time. That is where a tool helps. Three jobs stand out.
Drafting the routine reminder. Give a tool your usual wording and a few examples of reminders that worked, and it can write a courteous first draft for each overdue account: the invoice, the date, what was agreed, and a clear line about what to do next. The key phrase is your own wording. A tool writing from general knowledge produces a stiff letter that reads like a threat or like a form, and neither fits the customer who has paid you for ten years.
Sorting the list. A list of unpaid invoices is easy to keep and hard to act on. A tool can group accounts by how late they are, flag the ones that were never chased, point out a customer who used to pay on time and now does not, and note an invoice that looks like a duplicate or a dispute rather than a late payment. It is a way of deciding where an hour of a person's time should go first.
Keeping the follow-up on schedule. Many late payments are not refusals. They are an invoice sitting in somebody's inbox. A tool can remind your team that a reminder is due, draft it, and note when a promised payment date has passed. The steady, boring follow-up that busy people skip is exactly the work a tool does without getting tired.
Each of these makes the work faster without changing who is responsible for it. That is the test of a good first use.
What a person must keep
Money owed is a relationship as well as a balance. A person with authority keeps these, every time.
The tone for each customer. A long-standing customer in a hard month, a new customer who has never paid on time, and a customer who is simply careless are three different conversations. A tool does not know the history. A person who does can decide who gets a gentle note and who gets a call today.
The conversation itself. When an account is seriously late, a person calls. People explain things on the phone that they never write in an email: a dispute, a cash problem, a mistake on the invoice. A tool cannot hear that, and a draft cannot respond to it.
Any deal. A payment plan, a discount for paying now, a waived fee, a partial settlement. These are decisions about your money and your customer, and they belong to the owner or someone the owner has named.
Disputes. If the customer says the work was wrong, the delivery was short or the invoice is incorrect, the matter stops being a collections task. Somebody has to look into what happened before another reminder goes out. A cheerful automated nudge to a customer who has a real complaint makes everything worse.
The decision to stop supplying. Putting an account on hold, changing terms, or sending a matter to a lawyer or an outside firm changes the relationship, and may carry legal consequences. These go to a person, and to counsel where the stakes call for it.
Anything that sounds like a threat. A tool is happy to write a firm sentence and may write a sharper one than you meant. Rules about what you may say, and to whom, can be strict. Keep every message that mentions consequences with a person, and have your counsel look at your standard wording.
Sending. A person reads each message against the customer's account and sends it under their own name.
If the list feels long, notice what it leaves out: the drafting, the sorting, the schedule and the tracking, which is where most of the hours go.
The data rule for customer accounts
A receivables list is full of other people's information: names, contacts, amounts, payment history and sometimes notes about their financial trouble. Before any of it goes into an outside tool, you need one sentence about what may and may not. Something in this spirit: "No customer account detail goes into any tool that is not on the approved list, on a company account, and personal details about an individual customer stay out altogether."
In practice that means four things.
Check your agreements and your obligations first. Some customer contracts restrict what you may share about the account, and some rules limit what you may do with personal information. If you are unsure what applies to you, ask your counsel before the list goes in.
Use a company account, not a personal one. Check what the tool says about keeping your data and whether it learns from what you give it. Is ChatGPT safe for business use explains what a business plan changes about your data and what it does not.
Leave out what the draft does not need. A reminder needs an invoice reference, a date and an amount. It does not need a customer's private notes, their bank details or a story about their health or family. Your internal view of how risky an account is stays in your own records.
Write it down. If you have nothing in writing, do I need an AI policy for my small business shows how to put the rule on one page.
Starting safely
Start small and keep a person in every loop. Four rules make the start safe.
Draft-only mode. The tool writes a draft and never sends. A person reads each reminder and sends it. If a tool is connected to your email and can send on its own, switch that off or do not use it. A tool that contacts your customers about money without a person looking is the single riskiest setup there is.
One kind of account. Pick the plainest case, such as a first reminder on a modestly late invoice from a customer with a clean history, and leave disputes, large balances and difficult customers entirely to your people. A narrow start tells you more than a broad one.
Two weeks. Run it for two weeks, long enough to meet the odd accounts and the busy days, and short enough that a mistake cannot spread.
A log. Keep a simple record: the draft, what the person changed, which reminders brought a payment, and roughly how long the follow-up took. A shared spreadsheet is enough. After two weeks the log shows what the drafts handle well, what they get wrong, and whether any time or any cash was really gained. Decide on the evidence. If it helps, widen the scope a little. If not, you have lost two weeks and nothing else. Once you can see when customers really pay, can AI forecast cash flow covers turning that into a weekly forecast.
You do not need an IT department for any of this. What to automate first in a small business helps you pick the right first job, and can AI answer customer emails for a small business draws the same line between draft and promise for the inbox. If you are watching your cash closely because of a credit line, what to tell your bank about AI covers what a lender may ask about the tools you use. To judge whether the change paid off, is AI saving my business money gives a plain way to check.
What customers notice
Customers do not mind a courteous reminder. They notice a wrong one. They notice a reminder for an invoice they already paid, a tone that does not match the relationship, the same message arriving three times, and a stiff paragraph that could have come from anyone. They also notice the opposite: a clear, friendly note that arrives the day after the due date and makes it easy to pay.
So aim the tool at timing and consistency, and leave the judgment to your people. Check every draft against the account before it goes, and make sure a payment received yesterday has been recorded first. Be honest if a customer asks whether a tool helped. And keep a person they can call, because a customer with a question about money wants an answer from someone who can stand behind it.
Picture it going wrong
Picture a distributor of seventy people. A busy office manager, short of time at month end, pastes the whole list of unpaid invoices, with contact names and some personal notes about customers, into a free tool on a personal account. She asks it to write reminders and sends them in one batch with a quick look. One goes to a long-standing customer who paid by transfer the day before, and the payment had not yet been matched. Another goes to a customer who had a real dispute over a short delivery, and the wording is sharper than anyone intended. A third mentions a consequence the company has never decided to apply. The first customer calls the owner, offended. The second puts the account on hold. The list, with its personal notes, now sits in a tool nobody approved. The repair is small: draft-only mode, payments matched before any reminder, disputes pulled out for a person, a one-sentence data rule, one kind of account, two weeks and a log. This is a hypothesis, a composite and not a client.
Mistakes that repeat
Letting a tool send reminders on its own. Chasing an invoice that has already been paid. Sending a reminder to a customer with a live dispute. Letting a draft sound like a threat. Pasting the whole customer list into a personal account. Treating every late customer the same way. Offering a payment plan or discount without the owner's say. Starting on every account at once. No log, so nobody knows if it helps. Skipping your own counsel where the law could apply.
Frequently asked questions
Can AI help me collect late payments from customers?
It can draft reminders from your own wording, sort the list of unpaid invoices and keep the follow-up on schedule. A person should handle the tone, the phone call, any deal and any dispute, and should read and send every message.
Should AI contact my customers about overdue invoices on its own?
No. A tool does not know the customer's history, a dispute or a payment that has just arrived. Keep it draft-only, with a person sending each message.
Can I put my list of unpaid invoices into an AI tool?
Only if your agreements allow it and the tool is on your approved list, on a company account. Keep personal details and private notes out. If you are unsure what applies to you, ask your counsel before you put anything in.
How do I start without risk?
Draft-only mode, one plain kind of account, two weeks and a log of what the person changed. A person reads and sends every message, and handles anything serious by phone.
Will customers know a tool helped?
They will notice timing and accuracy, and they will notice a wrong or cold message. Check every draft against the account, and be honest if asked.
Is this legal advice?
No. Rules about contacting people over money owed, about interest and fees, and about what you may say depend on your contracts, your location and who the customer is. Have your counsel look at your standard wording.
Not sure where you stand? Take the 3-minute AI readiness scorecard.