Collections Automation: Manual vs Software vs AI Agents

Quick Summary

• Everyone already knows what great collections looks like. The constraint is hands, not knowledge
• A collector can work a few dozen to a few hundred accounts. Past that, judgment gets dropped in a predictable order
• Personalization and skip-tracing die first, followed by team coordination and then record accuracy
• First-generation AR software automated the compromise rather than removing it. Throughput rose, treatment stayed one-size-fits-all, and response rates kept falling
• The dividing line between tools is replying. Most automation handles outreach. The valuable half is handling what comes back
• Automate the skeleton, keep the judgment human. Routine nudges and the long tail on one side, strategic accounts and sensitive disputes on the other

Why great collections is so hard to sustain

Ask an experienced collector what excellent looks like and the answer is consistent.

You work the account rather than the invoice. You find the right person even after they have moved on. You check with the CSM before escalating, meet the customer's AP process on its terms, pull disputes out of the cadence instantly, write like a person from a real address, and keep records accurate enough that the next action reflects what is true today.

None of that's controversial, and most AR teams could describe it unprompted.

The problem is hands, not knowledge.

A collector can properly attend to a few dozen accounts, perhaps into the low hundreds. Past that something gives, and what gives is predictable:

• Axes of judgment get dropped first. Fifteen signals per account collapses into big versus small plus a couple of risk buckets, because nobody can weigh fifteen things across a thousand accounts. Every dropped axis is signal sacrificed
• The long tail gets abandoned. Everything below the top accounts falls to a fixed sequence from a no-reply address, or gets nothing at all. Recoverable money sits in aging buckets because no human had time to work it
• Personalization dies. Bespoke, on-thread messages don't scale by hand, so they become templates and blasts. Customers learn to ignore those, response rates fall, and the treatment becomes the opposite of the standard
• Skip-tracing stops. Finding a hidden right-contact is slow manual work. At volume it simply doesn't happen, so thousands of accounts quietly degrade to dead email addresses
• Coordination breaks. Looping in the CSM and arranging warm introductions go first under load, and the team defaults to firing notices in isolation
• Records rot. Promises go unpinned and disputed invoices get chased anyway. Broken records create false past-dues, and false past-dues destroy trust faster than any late notice

First-generation AR software was supposed to fix this. Mostly it automated the compromise instead of removing it.

Those tools were built around invoice-centric workflows and fixed calendar cadences, so they made a crude process faster without making it individual. Throughput went up. The one-size-fits-all treatment stayed, and response rates kept falling, because the customer could still tell it was a machine.

The standard was never wrong and it was never unknown. The only thing standing between every account and that level of attention is human bandwidth.

Remove the ceiling and the long tail can be worked to a standard that used to be reserved for the top few accounts.

Manual vs legacy AR software vs AI agents

Three broad approaches, and they aren't equivalent.

1. Manual collections

A person runs everything. The spreadsheet of who owes what, the reminders, the replies, the promises, and the escalation.

The upside is judgment. A good collector reads each account and adapts, which is exactly what the standard demands.

The downside is bandwidth, and it bites early. One person can't weigh every signal across hundreds of accounts, so the long tail goes unworked and follow-ups get dropped whenever the week gets busy.

2. Legacy AR software

The common next step automates the sending. Reminders fire on a schedule, aging gets tracked, and throughput improves.

Then a customer replies, and the system has nothing.

Legacy tools can't process a promise to pay or recognize a dispute. The calendar keeps firing while a human scrambles to handle the responses. It automates the easy half and leaves the hard half on your desk.

There's a second cost. A fixed calendar sending from a no-reply address is precisely the pattern customers learn to tune out, so the tool can lower response rates while raising activity.

3. AI collections agents

The newer category targets the half legacy tools skip.

An AI agent can read a customer's reply, recognize that "we'll pay next week" is a promise, schedule the follow-up for the day after, route a dispute out of the cadence, and adjust its tone to fit the account.

The aim is to work the way a good collector works rather than to send faster.

That's the real dividing line between the three. Most automation automates outreach. The harder and more valuable job is handling what comes back.

What to automate and what to keep human

Automate the skeleton and keep the judgment human.

Worth automating:

• The pre-due confirmation on every invoice, which is the highest-value touch and the one teams skip
• Routine nudges through the early days-past-due bands
• Follow-up scheduling, particularly the day-after-a-promise check
• The long tail nobody currently reaches
• Logging of every message, promise, and contact update

Worth keeping human:

• Strategic accounts where the relationship outweighs the invoice
• Sensitive disputes, and anything involving a real service failure
• The decision to apply a hold, involve legal, or write off a balance
• Customers in real financial distress, where the conversation is about survival rather than process

Good automation clears the busywork so the collector spends time where judgment actually pays.

The bar: no signal on any account gets ignored

Here is the standard worth holding any approach to, human or automated.

A great collector wants to weigh everything they know about an account, every time, and can't do it at scale. That's the compromise every AR team is making, whether or not anyone has named it.

The bar for a good system is that nothing about a customer gets ignored. Every available signal feeds the next decision, on every account, every time.

Anything less is a bandwidth compromise, and the job is to make the smallest one you can.

This is also where automation earns back the long tail. Those low-value, high-volume accounts go unworked under a manual model purely because there were never enough hands for that much ground. Worked consistently, they're real cash sitting in your aging report.

Where Lunos fits

Most of the collections process repeats. Reminders, follow-ups, promise tracking, and escalation run the same shape across hundreds of accounts, with occasional exceptions that need a person.

The volume behind that work keeps growing. The 2026 Intuit QuickBooks Small Business Late Payments Report found that 59% of US small businesses now carry invoices overdue by 30 days or more, up from 47% a year earlier, with an average of $17.7K outstanding. An earlier QuickBooks survey of mid-sized businesses put the time cost at around 14 hours a week on administrative work tied to collecting payments, reported by 65% of respondents.

Lunos is an AI coworker for accounts receivable rather than a rules engine. It runs on top of the process you already have, handling the collections lifecycle the way a skilled collector would.

In practice, it:

• Runs your full cadence across every account, in email and in Slack
• Reads customer replies, recognizing intent, tracking promises, and following up on the right day
• Logs every message and promise, so the audit trail stays clean and records stay live
• Routes disputes and exceptions to your team, and handles the routine chasing on its own

You stay in control throughout. Lunos offers Monitor, Suggest, and Act modes, so you decide how much autonomy the agent has and can widen it as trust builds.

The result is a consistent collections process that runs at scale without your team copy-pasting reminders all day. For a wider view of the category, see our roundup of AR automation software, or read more on AI in accounts receivable.

Ready to see it on your own ledger? Request a demo.