How to Put AI to Work in Accounts Receivable (Without Torching Your Customer Relationships)

Atradius surveyed US businesses and found that 43% of B2B invoiced sales are currently overdue, with bad debts eating about 5% of long-overdue invoices. Meanwhile, across the 10,000+ live AR, billing, collections, and credit roles in Audit Friendly's database of accounting job postings, current through June 2026, fewer than 3% mention AI at all. Sit with that pairing for a second - nearly half the invoices in the economy are late, the function responsible for chasing them is hiring at scale, and almost none of those job descriptions have noticed the technology reshaping the rest of the finance stack.

I wrote a while back that AP is the easiest place to put AI to work, and I stand by it, but AR is the more valuable one, because AR is where the cash is. So I want to work through why receivables has lagged, what AI is actually good at on the collections side, and how you'd wire it up without torching the customer relationships your revenue depends on - because that fear is the real blocker, and it's only half wrong.

Why is AR so far behind on AI?

The numbers are stark even by accounting's standards. In our postings data, about half of AR-family roles explicitly name collections work as a core duty, under 7% mention automation of any kind, and barely 1 in 40 mentions AI. For comparison, automation mentions run around 16% across all 80,000+ recent postings we track, so AR sits behind a profession that is itself behind. And only about 3% of AR postings even mention DSO, the one metric the whole function exists to move, which tells you how many of these roles are still scoped as data entry with a phone attached.

I get why, honestly. AP automation is internal - if the bot mis-codes an invoice, you annoy your own controller and fix it Tuesday. AR automation touches customers - if the bot fires a snotty dunning email at your biggest account, you've created a real problem for actual revenue. So finance leaders kept collections manual out of caution, and that caution made sense right up until the tooling got good enough to split the work properly.

What AI is genuinely good at in receivables

Cash application is the cleanest win, and it's not close. Matching payments to invoices, the remittance-advice scavenger hunt, the partial payment that ties to three invoices and a credit memo - that's pattern matching at scale, machines have gotten shockingly good at it, and nobody's customer relationship was ever damaged by getting cash applied correctly and fast.

Payment prediction and aging triage come next. A model that has watched a customer's behavior for two years knows whether the invoice sitting at 15 days past due is normal lag or a real risk, and it can sort your aging report by who actually needs attention instead of by who happens to be biggest or oldest. That triage is the difference between a collections desk that works the right ten accounts today and one that blasts reminder emails alphabetically and hopes.

And drafting. Reminder sequences, escalation notes, the summary of a disputed balance with every referenced invoice pulled together - AI drafts all of it well. Draft is the operative word, and it's what the rollout below hinges on.

What stays human, and why that's the point

The call to your biggest customer's AP manager stays human. The judgment about whether a struggling-but-loyal account gets thirty more days stays human. The negotiation on a payment plan, the read on whether a dispute is a stall tactic or a real grievance - human, human, human. Collections is a relationship job wearing a process-job costume, and the people who are great at it earn their keep ten times over. AI's job is to clear the shit out of the way so those people can spend the day on conversations instead of on copy-pasting remittance data, and the AR specialist who understands that trade is set up for a much better career than the one bracing against it.

A rollout that won't blow up your customers

Start with cash application, because it's invisible to customers and the accuracy is measurable from day one - you'll know within a month whether the match rate justifies trusting it further. Then point AI at the aging report and let it rank the queue each morning, and sanity-check its ranking against your own instincts for a few weeks before you let it drive. Then dunning, under the one rule that makes all of this safe: AI drafts, a human sends. Every reminder gets a ten-second review, your name stays on the email, and the bot never freelances with a customer. Measure DSO the whole way, because per Atradius nearly half of invoiced B2B sales are sitting overdue, and pulling even a few days out of that cycle is real cash at any scale. When you're ready to evaluate tools, we keep the AR and revenue automation category of our software directory current for exactly this.

What I'd do

If I ran an AR desk, I'd pick our messiest paying customer and run one month of their cash application through an AI tool next to the manual process, and I'd let the results argue for themselves. If I worked on one, I'd learn this stuff before my employer asks for it, because the postings will catch up the way they have everywhere else, and right now the candidate who can talk credibly about AI-assisted collections is interviewing against ten who can't - with a median posted salary around $70,000 USD across these roles, that edge is worth real money on the job board today. The same logic I laid out for rolling reconciliations applies here: the grunt work is exactly the part AI handles, so the best practice finally became the easy practice.

Frequently asked questions

Can AI replace collections staff?

No. AI handles cash application, payment prediction, queue triage, and drafting, but the negotiations, judgment calls, and customer conversations that actually resolve hard balances still need a person. The roles shift toward relationship work rather than disappearing.

Where should AI start in accounts receivable?

Cash application. It's customer-invisible, the accuracy is easy to measure, and it frees the most hours fastest. Aging triage and drafted-not-sent dunning come after you trust the tooling.

Does AI in AR damage customer relationships?

Not if a human stays on send. The safe pattern is AI drafts plus human review on anything customer-facing, while the fully automated wins happen in matching and prioritization where customers never see the machine.

What's the business case for AI in collections?

Atradius's 2025 US data shows 43% of B2B invoiced sales overdue and bad debts at about 5% of long-overdue invoices. Faster application, smarter triage, and consistent follow-up pull days out of DSO, and that's working capital you stop borrowing.

Are employers asking for AI skills in AR roles?

Barely - fewer than 3% of the 10,000+ live AR-family postings in our database mention AI, and under 7% mention automation, per Audit Friendly data, June 2026. That gap is exactly the opportunity for anyone in the function right now.

Every first-party figure in this piece comes from our database of 80,000+ accounting and finance job postings, updated daily, as of June 2026.