One service business we worked with lost the best part of a week every month to chasing payments by hand. Every customer got roughly the same reminder, whether they always paid on day thirty, needed a purchase order number before anything moved, or were in the middle of a dispute. The good payers were being nagged. The difficult accounts were getting a template.

The fix wasn’t a cleverer reminder. It was deciding who should get one at all.

Group customers by how they actually pay

Start with the ledger, not the terms. Payment terms say thirty days; behaviour says something more useful. In that engagement, customers fell into a handful of groups: those who pay on issue, those who pay on terms, those on a purchase-order cycle, and those who are historically slow. Each group gets its own timing. Some get no reminder at all, because they don’t need one.

Add the ways of working the team has already agreed with particular customers. If someone always rings the finance director rather than emailing accounts, that belongs in the plan.

What to automate

  • Routine reminders, timed to each group’s own habits.
  • Checking what’s been paid before anything goes out, so nobody is chased for an invoice they settled yesterday.
  • Flagging anything unusual for a person: a disputed invoice, a part payment, a changed contact.

What to keep with a person

  • Accounts marked sensitive. A key customer, a relationship under strain, someone going through a hard time. These never touch the automation.
  • Disputes and part payments. These need judgement about the relationship as much as the money.
  • Anything that changes terms. Agreeing a payment plan or writing off a balance is a commercial decision, not a reminder.

That second list is short, and it matters more than the first. Writing it down is the same discipline as deciding what not to automate anywhere else.

Start with approvals

At first, every reminder the system drafts should go to a person to approve, edit or reject. That record shows whether it’s getting the tone and timing right, and it lets the system earn more freedom on evidence rather than hope.

In the engagement above, routine chasing ended up running itself, and the team’s time went to the accounts that needed judgement. Read the case study.

Not sure where your business stands with AI? Find your AI profile: ten questions, about four minutes, no sign-up.