Can AI Chase Our Unpaid Invoices? What Automated Collections Actually Does

Can AI chase our unpaid invoices?

Yes, AI can chase your unpaid invoices, and the useful framing is that it chases on time rather than chases better. A reminder sent the morning an invoice goes past due does more work than a carefully written one sent three weeks later, and sending on time is exactly what a busy office stops doing when the month gets heavy.

An automated receivables chase does four things. It watches due dates across every open invoice without being asked. It sends a sequence you wrote, escalating in tone and recipient as the days pass. It stops the moment payment arrives, which sounds obvious and is the single most common failure in hand-run chasing. And it tells you each week which accounts moved, which did not, and which need a person.

The judgment stays with you. Whether to put an account on hold, whether to accept a payment plan, whether a customer who always pays at 60 days on 30 day terms is a problem or just a fact of your industry: those are commercial decisions, and the system's job is to present them cleanly and early rather than to make them.

There is a second, quieter benefit that owners tend to name first after a few months. Chasing money is uncomfortable, so people avoid it, and the avoidance is expensive. GoCardless's Pursuing Payments report, a YouGov survey of 500 Australian small business owners run in the second half of 2025, found almost one in four would write off 6% or more of annual turnover rather than have the awkward conversation. Australian data, and a small sample, but the instinct is not national. When the first three reminders are automatic and polite, the only conversations left are the ones that genuinely need a human voice, and there are far fewer of them.

How much money is actually sitting in our overdue invoices?

The money sitting in your overdue invoices is more than the ageing report suggests, because the ageing report does not show what the delay costs. The published figures are a reasonable sanity check on your own.

Atradius, in its Payment Practices Barometer for North America in 2026, found overdue invoices accounting for an average of 23% of business to business receivables, and late payment affecting seven in ten companies in the region. Its US report for the same year puts the figure at 22% of US receivables, with about 45% of business to business sales made on credit and 30 day terms the most common.

The small business picture is sharper because the buffer is thinner. Bluevine's survey of 1,052 US small business owners, published in March 2026, found 29% had delayed their own pay because customers failed to pay on time, and 17% had missed or nearly missed payroll for the same reason. Bluevine is a small business lender and commissioned the work, which is worth knowing, and the direction matches Intuit's: the QuickBooks 2026 Small Business Late Payments Report puts the average amount owed to an affected US small business at about $17,700, and found 59% carrying invoices 30 or more days overdue, up from 47% a year earlier.

Your own number takes twenty minutes. Pull the ageing report. Add the invoices past due. Then add the second number nobody tracks: hours per month your team spends on reminders, statements and payment matching, multiplied by their loaded hourly cost. The GoCardless survey put the Australian average at 1.5 hours a week, about 78 hours a year, among the businesses that chase at all.

Put together, those two numbers are the case. One says how much cash is sitting still, the other says what you are already spending to go and get it. If the spending is small and the cash is large, your problem is timing rather than effort, and timing is the part a system fixes.

Which parts of accounts receivable can run without a person?

Six parts of accounts receivable run without a person, and they cover most of the calendar time in a collections cycle.

Invoice delivery and confirmation. Sent, opened, and in the customer's portal if they use one. Some late invoices were never received by the person who approves them, and that is a tracking problem rather than a payment problem.

The reminder sequence. Before due, on due, and at the intervals you set afterwards, with the tone and the recipient escalating. The schedule is yours. The sending is not a decision.

Payment matching. Incoming payments reconciled against open invoices, including part payments and payments that cover several invoices with no reference. This is where most hand-run processes lose their accuracy, and an unmatched payment means a reminder sent to somebody who already paid, which is the one failure that genuinely annoys customers.

Statements and copies on request. A customer asking for a copy invoice or a statement is asking a question with a lookup for an answer.

Segmentation. Sorting the ledger into who is reliably slow, who has broken a pattern, and who has gone quiet. The second group is where the money is, and it is invisible on an ageing report sorted by days.

Escalation packaging. When an account crosses your threshold, a person should receive the account history, every reminder sent, the payment pattern over the last year and the contact details in one view, rather than piecing it together before they can make the call.

What stays human: negotiating terms, agreeing a payment plan, deciding to stop supplying, and the call to a customer you value who has gone quiet. Those are relationship decisions with money attached.

What should an automated payment reminder actually say?

An automated payment reminder should be short, specific, and send the invoice with it. The tone that collects is factual rather than apologetic, and it escalates in content before it escalates in temperature.

A sequence that works for most businesses looks like this. Three days before due: a note that the invoice is coming up, with the amount and the payment link. On the due date: a reminder with the same details. Seven days past: a direct note asking whether there is a problem with the invoice, which gives a dispute somewhere to surface. Fourteen days past: the same, copied to a second contact, usually whoever approves payments rather than whoever received the work. Thirty days past: a person.

Three rules matter more than the wording. Always include the invoice number, the amount and a way to pay in the message itself, because a reminder that requires the recipient to go and find something is a reminder that waits. Always ask whether there is a problem with the invoice rather than only asking for payment, since late invoices are sometimes disputes nobody reported. And never send a reminder to somebody who has paid, which is a matching problem rather than a writing problem.

What not to do: threaten early, apologise for asking, or send five messages in a week. Volume is what does the damage. Five messages in a week reads as a machine with nobody behind it, and one direct note on the due date does not.

One more thing worth building in from the start. Every reminder should offer the shortest possible route to paying, and for most businesses that means a payment link rather than bank details in a signature. The gap between intending to pay and paying is where a week goes.

When should a person pick up the phone instead?

A person should pick up the phone when the pattern breaks, when the amount is material, or when the customer matters. Those three tests catch almost every account where a human call outperforms another message.

A customer who always pays on day 45 and has just paid on day 46 is not a collections problem. A customer who has paid on day 30 for two years and is now at day 55 with no explanation is one, and the sequence should surface that on the first missed pattern rather than at an arbitrary day count. Pattern breaks are the highest value signal in a receivables ledger, and almost nobody watches for them by hand.

Material amounts deserve a voice regardless of the pattern. Set the threshold by what a bad month looks like for you, not by a percentage of the ledger.

Relationships deserve a voice too. A long-standing customer who has gone quiet is either in trouble or unhappy, and in both cases you want to know before the next invoice goes out. An automated sequence handles this badly because it cannot hear the pause before the answer.

There is a fourth case: when the customer has already said something. Any reply to an automated reminder should stop the sequence and route to a person. A system that keeps sending after a customer has written back is the worst version of this, because it tells the customer there is nobody home.

Where BLACKSIG builds a receivables chase, the sequence, the matching and the escalation packaging run on our infrastructure and we operate them from there, and the calls, the terms and the judgment calls stay with your people. For a business whose receivables sit inside job management or an industry system rather than an accounting package, the connection work is usually the real project, and does AI automation work with my software covers what that depends on.

Does automating the chase damage the customer relationship?

Automating the chase does not damage the customer relationship if the sequence stops when it should and a person is reachable. The evidence on what annoys customers is more specific than the fear usually is.

The damage in collections comes from three things: chasing somebody who already paid, chasing without the invoice details so the recipient has to go digging, and chasing with no way to reply to a human. All three are failures of plumbing rather than failures of automation, and all three are more common in hand-run processes, because a person working from a spreadsheet does not know a payment landed this morning.

What customers mind far less than owners expect is being reminded. A reminder on the due date is ordinary business, and most accounts payable departments rely on them, since they are managing their own queue and a clear reminder with an invoice number attached makes their job easier.

The relationship risk sits with the small number of accounts where something is genuinely wrong, and the point of automating the routine chase is to find those accounts earlier, while the problem is still a conversation rather than a write-off.

One caution on tone. Reminders that sound like legal threats in week two do real damage, and they do it to the customers who were always going to pay anyway. Keep the escalation about who is copied and how direct the ask is, not about consequences, until you are genuinely at the point of consequences.

How do we measure whether it is working?

Measure whether the chase is working with four numbers, taken before anything is built and then monthly.

Days sales outstanding, which is your average collection time, calculated the same way each month so the trend means something. Watch the trend rather than the absolute number, because the absolute number is mostly a function of your industry and your terms.

The percentage of receivables past due, which the Atradius survey above puts at 23% across North America, so you have a benchmark to argue with.

Hours spent on collections per month. This is the number that tells you whether the work actually went away or quietly moved to somebody else. A chase that halves your overdue balance and doubles the time your bookkeeper spends reconciling has not paid for itself.

Disputes raised as a share of invoices. This one usually goes up when a chase is working properly, and that is good news. A dispute surfaced at day seven is collectible. The same dispute discovered at day ninety is a negotiation.

Set the baseline before launch, because without a before there is nothing to compare to, and the argument about whether it worked becomes a matter of opinion. The general method, including how to avoid counting time that moved rather than disappeared, is in how to measure whether AI automation is saving money. Where receivables sit alongside the rest of a professional firm's admin load, the consulting and advisory firms board shows what else is in the same sweep, and for trades businesses invoicing off completed jobs it is the home and trade services board.

Frequently asked questions

Related resources

• How do you automate invoice processing with AI? (the payable side)
• Can AI do our bookkeeping?
• How do you automate quote follow-up?
• Does AI automation work with the software I already use?
• Is our data good enough for AI automation?
• How do I measure whether AI automation is actually saving us money?
• What AI can do in a consulting or advisory firm