AI collections software: what to look for beyond automated dunning

AI collections software is a tool that uses artificial intelligence to run collections: deciding what each account needs, shaping the outreach, and, in the strongest systems, taking the action and recording the result. The catch is that most products sold under this name are rule-based dunning with an AI label. The real value comes from an agent that reasons about each account and acts on it, not a template engine on a timer.

If you are evaluating tools, the marketing will not tell the two apart. Every vendor says "AI". The difference shows up in what the software does on a live account when no one is clicking approve. This page is about how to find that difference and what genuinely capable AI collections software should do.

What AI collections software promises vs delivers

The promise is consistent across vendors: less manual chasing, faster cash, lower DSO, a smaller team covering a larger book. The delivery varies enormously.

At the shallow end, "AI" means a slightly smarter mail merge. The software fires reminder one, two, and three on a fixed schedule, maybe with a payment-date prediction bolted on the side. It does not read replies. It does not change course when a customer disputes. A human still works every exception by hand. The DSO gain is modest, because the bottleneck, people reading and deciding, never moved.

At the capable end, the software does the reading and deciding itself. It behaves like a collector, working each account on its merits, and it owns the result. That is the gap between a tool you still have to drive and one that runs the function. The same distinction defines agentic AI for collections: an agent reasons and acts, where a scheduler only fires.

Rules and templates vs agentic decisioning

The clearest way to grade AI collections software is to ask what drives the next action.

In a rules-and-templates system, the trigger is elapsed time. Day three sends template A, day ten sends template B. The customer's reply, history, and dispute status do not change the path. It is predictable and it is brittle, because real accounts do not behave on a schedule.

In an agentic system, the trigger is the state of the account. The software reads the latest reply, checks the payment history and the aging, and chooses: push, pause, escalate, or wait. If the customer wrote back to dispute a line, it pauses collection on that line and routes the dispute rather than sending the next scheduled nudge. The behavior adapts because the system understands the situation, not because the calendar moved.

When you demo a tool, push on this directly. Send a disputing reply mid-sequence and watch what happens. A scheduler keeps dunning. An agent stops and handles it.

Core capabilities that actually move cash

Past the label, a few capabilities separate software that moves DSO from software that just sends mail.

• Per-account decisioning. The system chooses the next action for each account from its current state, not from a fixed timeline.
• Reading and acting on replies. It parses free-text customer email, classifies intent, dispute, promise, question, copy-invoice request, and takes the matching action automatically.
• Cash application tied to collections. It matches incoming payments, including partials and short pays, so the aging stays accurate and the agent never chases money that already arrived.
• Dispute and deduction handling. It catches short pays early, classifies the reason, routes to an owner, and pauses collection on the disputed amount.
• A real audit trail. Every decision and action is logged with its reason, so finance can see what the software did and why.

Capabilities you can ignore: vanity dashboards, sentiment scores no one acts on, and "AI insights" that produce a chart but never an action.

Integration with ERP and accounting systems

Collections software is only as good as its connection to the system of record. The ledger lives in your ERP or accounting platform, and shallow integration leaves the real work stranded in a separate tool.

Look for two-way integration. The software should read open invoices, payments, credit memos, and customer data from the ERP, and write its actions and results back: logged promises, applied cash, dispute status, updated notes. When the write-back is real, the ledger reflects reality without anyone retyping it, and DSO and aging numbers stay current on their own. When it is not, you get a parallel tool that someone has to reconcile by hand, which adds work instead of removing it. Confirm the depth of the NetSuite, SAP, QuickBooks, or other connector before you buy.

Buyer checklist for evaluating AI claims

Use these questions to test any "AI" claim in a demo.

• What happens with no human click? Ask for a live account walked end to end, unattended. If every step waits on approval, it is assistive, not autonomous.
• Does it read and act on a free-text reply? Send a real disputing email and watch whether the system understands and changes course.
• How deep is the ERP write-back? Confirm it writes actions and results to the system of record, not just exports a worklist.
• What is it measured on? Push for cash recovered, DSO, and share of cases resolved without escalation. Be wary if the answer is emails sent.
• Where is the audit trail? Every action should carry a recorded reason a person can review.

For a fuller framework, see how to evaluate AI AR vendors. It is also worth understanding AI versus RPA in accounts receivable, since a lot of "AI collections software" is really scripted automation that breaks the moment an account does something unexpected.

Why outcome accountability matters

The single best filter is accountability for the result. A tool that owns an outcome, cash recovered, DSO down, has to read accounts, decide well, and adapt, because activity alone will not hit the target. A tool measured on activity can send thousands of messages and recover nothing, and still report success. Buy on the outcome, and the difference between real AI and a relabeled scheduler resolves itself.

How Rex approaches AI collections

Rex is an agentic AI accounts receivable agent, not dunning software with an AI label. It works each account on its current state, reads every customer reply, decides whether to push, pause, or escalate, and writes the result back to your ERP. It is measured on the outcome that matters, cash recovered and DSO down, and it escalates only the cases that need a human decision. Your team stops driving a tool and starts overseeing a function.

See how Rex runs collections on each account, autonomously, from first reminder to applied cash.