Getting paid faster — what AI actually does for accounts receivable

Most businesses are owed more money than they realise, later than they should be, for a reason that has nothing to do with bad customers: nobody chased.

The invoice went out. It was not paid on time. And then the follow-up — the reminder, the second reminder, the phone call, the statement — did not happen, because the person who would do it was busy doing something that felt more urgent. It is almost never a decision not to chase. It is chasing losing, every day, to whatever is on fire.

That is the actual accounts-receivable problem in most small businesses, and it is a strikingly good fit for automation, because the work is defined, repetitive, and only skipped for lack of time.

Why late payment is a follow-up problem

Think about what actually collects a debt. It is not skill. It is persistence — a reliable sequence of nudges that escalate. A reminder before it is due. One on the day. One a few days after. A firmer one at two weeks. A call. A statement.

Every business knows this sequence works. Almost none run it reliably, because it requires someone to remember, every day, which of hundreds of invoices has hit which trigger, and to act — while doing their actual job.

So the sequence runs when things are quiet and lapses when they are busy. And things are usually busy. The result is not a business that decided to be lax about collections; it is a business where collections quietly never happens consistently, and the receivables age accordingly.

What AI does here, precisely

It runs the sequence. Reliably. Without getting busy.

• It knows which invoices are approaching due, due, or overdue, and by how much.
• It sends the right nudge at the right trigger — the gentle reminder before due, the firmer one after, escalating on your schedule.
• It personalises tone and channel — a long-standing customer gets a softer touch than a serial late payer.
• It escalates to a human when the automated sequence has run its course and a person needs to make a call.
• It keeps a record of every contact, so when you do pick up the phone, you know exactly what has already been said.

This is an agentic task in the clean sense: a defined goal (get this invoice paid), a sequence of low-risk reversible actions (reminders), and a clear escalation point to a human when judgement is needed. Reminders are about as safe as automated actions get — the worst case of an unnecessary reminder is mild awkwardness, not a lost payment.

Where it genuinely moves the number

The metric is days sales outstanding — how long, on average, your money sits in someone else's account after you have earned it. For most businesses running no consistent follow-up, simply running the sequence reliably pulls this in, because a large share of late payment is not refusal; it is the invoice slipping the customer's mind, and a timely nudge fixes it.

That improvement is close to free money. You already did the work and earned the invoice. Getting it paid two weeks sooner, consistently, improves your cash position with no new sales and no discounting. For a business where cash is tight — which is most of them — that is worth more than it sounds, and it compounds across every invoice you raise.

Where AI stops, and a human takes over

Be clear about the boundary, because the tools that overpromise here cause real damage.

The genuine dispute. "We are not paying because the delivery was short." That is not a follow-up problem; it is a problem, and it needs a person who can investigate and negotiate. An automated reminder sent into a real dispute makes you look like you are not listening, which hardens the customer and delays payment further.

The relationship call. Your biggest customer is late, again, but they are also a third of your revenue. How hard you push is a commercial judgement no model should make. Push too hard and you protect a receivable while losing the account.

The genuine hardship case. A good customer in temporary trouble may need a payment plan, not a dunning sequence. That is a relationship decision with long-term value on the line.

The pattern is the one from what AI still cannot do in ERP: automate the persistence, keep the judgement human. A system that automates chasing but hands genuine disputes and relationship calls to a person — with full context of what has already been said — is doing exactly the right division of labour. One that tries to automate the judgement too will eventually send a legally-toned final notice to your best customer over a delivery dispute that was your fault, and that is a story that ends in a lost account.

Common questions

Why do my customers keep paying late?

Usually because nobody chased. The invoice went out, it was not paid on time, and then the reminder, the second reminder, the phone call and the statement did not happen, because the person who would do it was busy with something that felt more urgent. It is almost never a decision to be lax about collections; it is chasing losing, every day, to whatever is on fire. A large share of late payment is the invoice slipping the customer's mind.

What does AI actually do for accounts receivable?

It runs the follow-up sequence reliably, without getting busy. It knows which invoices are approaching due, due or overdue and by how much, sends the right nudge at each trigger, adjusts tone and channel so a long-standing customer gets a softer touch than a serial late payer, records every contact so you know what has already been said, and escalates to a person once the automated sequence has run its course.

Should automated reminders keep running when a customer disputes an invoice?

No. "We are not paying because the delivery was short" is not a follow-up problem, and a reminder sent into a real dispute makes you look like you are not listening, which hardens the customer and delays payment further. Genuine disputes, the relationship call with a customer who is a large share of your revenue, and the good customer in temporary trouble all belong with a person who has the full history in front of them.

How do I know whether automated chasing is working?

Watch days sales outstanding — how long, on average, your money sits in someone else's account after you have earned it — before and after. This is one of the few AI investments with a clean, fast, measurable payback, so if the number is not moving, something is wrong and you will see it quickly. Getting invoices paid sooner improves your cash position with no new sales and no discounting.

How to start

Not by overhauling collections. By making the routine follow-up reliable.

1. Map your current sequence. Most businesses do not have a written one — that is the first finding. Decide the triggers and the tone at each stage.
2. Automate the reliable, low-risk part — reminders up to a firmness threshold you set.
3. Define the escalation clearly. At what point does a human take over, and who is it?
4. Give the human context. When it escalates, the person should see the full history, not start cold.
5. Watch the number. Track days sales outstanding before and after. This is one of the few AI investments with a clean, fast, measurable payback — if it is not moving the number, something is wrong and you will see it.

The businesses that get value here are not the ones with the cleverest tool. They are the ones that finally run the follow-up sequence every single day — which is precisely the thing humans are bad at and software is good at.

The next step is usually smaller than people expect. Talk to us about one process worth starting with.

Related: AI cash flow forecasting for businesses and automating order to cash.

Also worth reading: handling a business dispute.

Also worth reading: managing cash on delivery.