How to Automate Collections and Past-Due Follow-Up

Two different companies, four months apart, same answer.

I asked an accounts receivable specialist what took the most time in her week. Collections: four hours. Pull the aging report, work down it customer by customer, decide who gets a notice, generate each notice by hand, match it against the invoice, raise a credit note where one is needed, send.

At another company, a different person doing the same job described the same four hours. Fifteen- to thirty-day payment terms and no automated follow-up of any kind. Every reminder that went out went out because a human remembered.

Neither of them was complaining. Both of them said it the way you describe weather.

Why AR keeps coming out on top

When we finish discovery, the workflows get ranked. Collections lands near the top more often than anything else, by a distance. There are four reasons.

The baseline is already true. Nobody has to construct it. Days sales outstanding is a number your accounting package already produces, your controller already watches, and your bank already asks about. You are not inventing a metric to prove a project worked. You are moving one that already exists on a report someone reads monthly.

The process is genuinely rule-shaped. Invoice is thirty days past due, customer is not on a payment plan, no dispute is open, so a notice goes out. The judgment calls are real but rare, and they sit at the edges: the customer you do not chase because of a relationship, the disputed line, the one going through a restructuring. Those go to a person. The rest is a rule your controller can read.

It has one owner. Most workflows cross four departments and take a month of meetings to change. AR is usually one person and their manager. You can build it, show it to two people, and have it running.

The output is cash. Every other efficiency argument ends with "so we could redeploy that time." This one ends with money arriving sooner. That is a conversation an executive team finishes in one meeting.

What we actually build

Not a chatbot. Not a model that decides who to chase.

A system that reads the aging data on a schedule, applies the rules your controller wrote, drafts each notice with the invoice attached and the right contact on it, and puts the whole batch in front of a person once a day. They scan it, pull the two they want to hold, and release the rest.

That last part matters, and it is the part people try to remove. Keep the human release for at least the first several weeks. Collections touches customer relationships, and the cost of one wrong notice to your largest account is larger than the cost of a person spending five minutes a day looking at a list. The review gate is part of the design, not training wheels.

What comes off the four hours is the assembling. The pulling, the matching, the generating, the copying between the spreadsheet and the invoice and the notice. What stays is the deciding, which took about fifteen minutes of the four hours and was the only part that needed a human at all.

The part that surprises people

The first thing this produces is a list of everything wrong with your data. Faster payment comes after.

Customers with two records. Invoices with no contact on them. Payment terms in the system that do not match the ones in the contract. Credit notes raised against the wrong period.

Every one of those was already costing you. They were just being absorbed by a person who worked around them silently, several times a week, for years. When the process becomes a system, the workarounds stop being invisible and become a list you can actually fix.

That is worth saying to a nervous executive team before the build. The first month will surface problems. Those problems are not new, and the system did not cause them. It found them.

If you only do one thing

Ask whoever owns receivables how long the past due run takes them, and how they decide who gets chased.

If the answer to the second question is a set of rules, you can build it, and the number on the first question is your baseline.

If the answer is "it depends, I just know," spend an afternoon writing down what they know. That afternoon is the project. Everything after it is typing.