AI for accounts receivable and collections

Ask a finance team what falls off the list in a busy month and collections is usually the answer. It is not urgent on any given day, it is uncomfortable, and its absence has no immediate consequence.

The consequence arrives later, as a debtor ledger with a lengthening tail, and by then the conversations are harder because the invoices are old.

This makes receivables an unusually good automation candidate — not because chasing is difficult, but because it is the thing humans reliably fail to do consistently.

What automation handles

Knowing what is due, accurately, today. Sounds trivial and frequently is not. If cash application is behind, the aged debtors report shows invoices as unpaid that were settled last week. Chasing a customer for money they have already sent damages the relationship and destroys the credibility of every future chase. Accurate receivables depend entirely on reconciliation being current.

Sending the reminder on the right day, every time. No judgement, no discomfort, no forgetting. This alone typically moves the average collection period more than any other single change.

Escalating on a defined path. Polite reminder before due date, firmer note after, escalation to a named person, hold on further supply. Each step defined in advance rather than decided in the moment, which is what makes it consistent.

Applying receipts. Matching payments to invoices, including partial payments, multi-invoice settlements and deductions — the same matching problem as bank reconciliation.

Flagging behaviour change. A customer who has always paid within 30 days going to 45 is a signal worth having early. This is where prediction genuinely helps: not forecasting whether an individual invoice will be paid, but noticing that a pattern has shifted.

What should stay human

The relationship call. Your largest customer, 20 days late, on an invoice they have queried. That is a conversation, and an automated escalation into it is actively harmful.

Disputes. Once a customer has raised a genuine query, the chase should stop and a person should take over. A system that keeps sending reminders on a disputed invoice tells the customer nobody is reading their emails.

Supply holds. Stopping delivery to a customer is a commercial decision with consequences beyond the ledger. Propose it, do not execute it.

Anything with history. Special payment terms agreed verbally, an ongoing negotiation, a customer going through difficulty you have decided to accommodate. If the context is not in the system, automation will act as though it does not exist.

The rule that works: automate the routine chase, escalate the human cases to humans early, and make sure a dispute suppresses automation immediately.

The prerequisite everyone underestimates

Automated collections is only as good as your cash application.

If receipts are applied weekly, your ledger is wrong for six days out of seven, and any chase sent in that window risks being wrong. One reminder sent to a customer who paid on time costs more credibility than ten correct reminders build.

So the order matters: get cash application current and automated first, then automate chasing. Teams that do it the other way round generate a burst of embarrassing emails and then switch the whole thing off.

What actually moves the numbers

Three effects, in order of size:

Consistency. Reminders that always go out, on time, regardless of how busy the month is. Most of the improvement comes from here, and it is unglamorous.

Earlier contact. A reminder before the due date — a courtesy note rather than a chase — reliably improves on-time payment and costs nothing in goodwill.

Faster dispute discovery. Many late invoices are late because of a problem nobody surfaced: wrong PO reference, missing delivery note, incorrect price. Chasing promptly finds these while they are cheap to fix instead of at 90 days when the customer has forgotten the detail.

That third one is underrated. A large share of aged debt is not reluctance to pay; it is an unresolved problem that nobody chased early enough to discover.

What to watch

Tone. Automated reminders read as automated unless someone writes them properly. Have a person write the templates and read them aloud.

Frequency. Escalation paths designed without restraint produce four emails a week. That trains customers to ignore you.

Suppression rules. Disputes, agreed payment plans, customers in negotiation — all need to stop automated chasing immediately, and it needs to be easy for anyone in the business to trigger that, not just finance.

Common questions

Can AI chase customers for payment?

It can handle the routine part reliably — knowing what is genuinely due, sending reminders on schedule, escalating along a defined path, and applying receipts including partial and multi-invoice payments. The relationship conversations, disputes, supply holds and anything involving context not held in the system should remain with a person, and a raised dispute should suppress automated chasing immediately.

What has to be in place before automating collections?

Current cash application. If receipts are applied weekly, the ledger is wrong most of the time and automated reminders will go to customers who have already paid, which costs more credibility than correct reminders build. Automating receipt matching and reconciliation first, then chasing, is the order that works.

Does automated chasing damage customer relationships?

It does when it chases disputed invoices, escalates too frequently, or contacts customers who have already paid. Handled properly it usually improves relationships, because problems get surfaced early while they are still easy to resolve rather than at ninety days when the detail has been forgotten by everyone involved.

What actually reduces days sales outstanding?

Consistency more than anything else — reminders that go out on time regardless of how busy the month is — followed by earlier contact, including a courtesy note before the due date, and faster discovery of disputes. A significant share of aged debt is not reluctance to pay but an unresolved problem such as a wrong reference or a missing delivery note that nobody chased early enough to find.

Related: getting paid faster with AI · automating order to cash · automating bank reconciliation with AI