Accounts Receivable: What AI Can Chase for You

Ask a small business owner why an invoice is 60 days late and the honest answer is usually that nobody chased it. Not a dispute, not a customer refusing, not a cash flow crisis at the other end. The invoice sat in an inbox, the approver was away, and the one person who might have followed up had a busier week. That is a process failure, and it is the kind automation is genuinely good at.

Sort your late invoices first

Before automating anything, spend twenty minutes categorising what is currently overdue. Most owners have never done this and are surprised by the split.

Reason it is late	What actually fixes it
Nobody followed up	A reminder sequence that runs without anyone deciding to
Missing PO number or wrong contact	Fixing the invoice template and the intake step
Disputed work or an unresolved complaint	A person, quickly, and no automated reminders
Customer cannot pay	A conversation about a payment plan

The first two rows are usually the majority, and both have fixes that are cheaper than chasing. The second row in particular tends to be a template problem repeating on every invoice, which means one afternoon removes a recurring cause rather than a single instance.

What to automate

The reminder before it is due. Three days ahead, friendly, with the invoice attached and the payment link in the message. This one is underused and it converts, because it reaches the approver while there is still time.

The reminder the day after. Short, neutral, no implication of bad faith. Most payments recovered by a sequence come from these first two messages.

Payment matching. Reconciling incoming payments against open invoices is repetitive, high volume and verifiable, which makes it a good candidate. Keep a person checking the exceptions, since the mismatches are where the errors concentrate. The wider extraction question is in extracting data from PDFs with AI.

The weekly ranked list. Rather than an aged receivables report nobody reads, a short list each Monday of the five accounts most worth a call, ordered by amount and by how far past their usual pattern they are. That turns a report into a task.

What to keep human

The third message onward. Once a reminder starts carrying an edge, the relationship risk rises and the wording matters. Draft it with AI if you like, and have a person read it before it sends.

Anyone with an open complaint. An automated payment reminder landing during a service dispute converts a recoverable situation into a lost customer, and it happens constantly because the collections system and the support inbox do not talk to each other. Build the suppression rule before you build the sequence.

Your largest accounts. If one customer is 30% of revenue, they get a call from a person, always. The efficiency saving is irrelevant next to the relationship.

Payment plans and stop-work decisions. Judgment, consequence and discretion, which is the combination that stays with people.

On predicting who pays late

With two years of payment history, a model can rank customers by how they have behaved. Some pay around day 45 whatever the terms say. Some pay on time until something changes, and the change is the signal.

That ranking is useful for deciding who gets the first phone call on Monday. It is not a credit policy, and it should not quietly become one. A score built on past behaviour can end up standing in for something you would not want to decide on if it were named out loud, which is the mechanism described in AI bias in a small business.

Keep it as a prioritisation aid with a person deciding, and revisit the list occasionally to check it is not simply reproducing a pattern about who your customers are rather than how they pay.

A two-week version

Categorise the current overdue list. Fix whatever the invoice template is getting wrong. Turn on a before-due and a day-after reminder in the accounting package you already pay for, since most of them have this built in and switched off. Add a rule that suppresses reminders for any account with an open complaint. Then produce one ranked list each Monday and actually call the top three.

Measure days sales outstanding before you start and again after eight weeks. That number moving is worth more to a small business than most efficiency gains, because it is cash you already earned. The related bookkeeping automation is in AI bookkeeping, and the invoice side in GST and HST invoice automation.

Frequently Asked Questions

What is accounts receivable automation?

Software handling the repeatable parts of getting paid: issuing the invoice, sending reminders on a schedule, matching incoming payments to open invoices, and flagging what needs attention. AI adds two things on top. It drafts the follow-up in language suited to the specific customer and situation, and it can rank which overdue accounts are most likely to need a call rather than another email.

Why are most invoices paid late?

In small businesses, usually because nobody followed up rather than because the customer refused. The invoice sat in an inbox, the person who approves it was away, or it needed a purchase order number nobody supplied. Those are process failures on both sides and they respond well to consistent reminders. Genuine disputes and genuine inability to pay are a smaller share, and they need a person, not a sequence.

Should reminder emails about invoices be automated?

The first two, generally yes. A polite reminder three days before due and another the day after due is exactly the kind of consistent, unemotional follow-up that people find hard to sustain manually and that recovers a meaningful share of late payments. Anything after that carries relationship risk and should be reviewed before it sends, because the customer receiving it may be the one you most want to keep.

Can AI predict which customers will pay late?

It can rank accounts by past behaviour, which is useful and is not prediction in any strong sense. If you have two years of payment history, a model can tell you which customers historically pay around day 45 regardless of terms and which pay on time until something changes. Use that to decide who gets a phone call first, not to set credit policy, and be careful that the ranking is not standing in for something you would not want to decide on.

What should stay human in collections?

Anything where the relationship or the money at stake is significant. The call to a long-standing customer whose payments have slipped. Any negotiation of a payment plan. The decision to stop work or refer an account. And any message going to a customer who has already raised a complaint, because an automated reminder landing in the middle of a service dispute turns a recoverable situation into a lost one.