Smart Collections Blossom with Agentic AI

A recent discussion revealed how Agentic AI is helping Accounts Receivable (AR) teams prioritize collections, reduce manual effort, improve cash application, and turn receivables into cash faster.

For many AR teams, deciding what receivables to collect first still begins with an aging report. Just sort the invoices by due date or value, identify what’s overdue, and start working down the list.

That approach can identify risky debt. But it can’t accurately predict which customer is most likely to pay, when to contact them, or what communication method is most likely to produce payment.

This is where agentic AI is essentially changing collections. Rather than simply automating the same collections workflows, AI can bring together payment history, customer behavior, ERP and banking data, with previous interactions to help collections teams determine where human attention will make the biggest difference.

During a recent webinar Bottomline Principal Value Consultant, Naomi Smith explained “The aging balance and the spreadsheet approach for sure indicate the riskiest debt, but it doesn't indicate which invoice has the greatest propensity to pay.”

Identify the Debt Most Likely to Turn into Cash

Collections teams have traditionally had a limited number of practical ways to prioritize large portfolios. Aging, outstanding value, and due date provide simple, measurable criteria. The problem is that customers don’t all behave in the same predictable ways.

Some can’t pay. Some won’t pay without additional prompting. Others may pay readily if they receive the right communication at the right time. Collections teams trying to treat all three the same way expends considerable effort without necessarily improving outcomes or achieving full portfolio coverage.

Panelists agreed that AI allows prioritization to be more dynamic. Historical payment behavior, customer records, previous interactions, deductions, and banking data can help identify which debts to pursue, when to pursue them, and the type of engagement most likely to work.

That might mean acting before an invoice becomes overdue. If past behavior suggests a customer responds well to a reminder three days before the due date, there is little reason to wait until the invoice crosses an arbitrary aging threshold. Another customer might be more likely to respond to a mobile SMS with a secure payment link.

The objective is not to chase more invoices; it is to direct collections resources toward actions most likely to convert receivables into cash.

Look Beyond the Collections Queue

Getting collections right also requires context beyond AR. A customer who short-pays an invoice, for example, may have discussed a discount threshold with sales or customer service. Without visibility into that conversation, collections may chase an amount the customer believes is not owed, accept an invalid deduction, or spend days investigating something another department already knows the answer to.

“The collections team really can't operate as a silo,” Smith said. “To collect cash effectively requires additional context from the wider organization.”

The same principle applies after payment arrives.

Cash application can leave money effectively trapped because the business cannot identify what a payment relates to. Combining bank data with invoice records, historical payments, and customer behavior gives AI more information with which to identify likely matches and reduce unapplied cash.

A customer's promise to pay can become another useful data point for reconciliation.

This is why improving collections can have effects beyond AR. Faster collections and cash application can reduce days sales outstanding (DSO) and improve the cash visibility that treasury needs for forecasting, and that other finance functions need for their own decisions.

Move from Fixed Workflows to Adaptive Collections

Automation itself is hardly new to collections. A conventional workflow might send an email four days after an invoice becomes overdue, trigger a phone call at seven days, and escalate the account at 14 days. It is efficient and repeatable, but the sequence remains essentially the same regardless of the customer's behavior.

Agentic AI introduces a different model: sense, decide, act, and learn.

The agent assesses available information, decides an appropriate next step, acts within established guardrails, and evaluates the outcome. Did the customer respond? Was the invoice paid? Did one channel work better than another? That new information informs the next interaction.

“Traditional workflows execute rules,” Smith said. “Agentic collections continuously learn what drives payments, and [will] adapt accordingly.”

Importantly, adaptive does not mean uncontrolled. In a live demonstration, Bottomline Senior Solutions Consultant Matt Nuccoll showed how organizations can establish guardrails requiring human intervention at specific points.

“Maybe we want to stop all communications if they're over a certain amount, maybe $20,000 or so, and make sure we can review and approve any of that communication before it goes out to a customer,” Nuccoll said. The underlying financial system remains the system of record rather than allowing AI to independently alter financial records.

Start with Where Cash is Getting Stuck

For finance leaders exploring agentic AI, the most useful first step may be surprisingly conventional: identify the biggest problem in the existing collections process.

Is too much collector time spent determining what to work on next? Are disputes delaying otherwise collectible debt? Are current communications failing to produce payment? Is cash arriving but remaining unapplied? All are signs of ineffective approaches.

After determining collection activities, estimate the cost of the problem. Time spent on manual processes can provide one starting point, but the bigger opportunity may be improved recovery rates rather than simply labor savings.

Cost also features in deciding how to build. Webinar host, Bottomline’s Head of Solutions Consulting, Richard Ransom put a common attendee question to Smith: “What is the difference between this solution and building it ourselves using the tools we might have available like Copilot or Claude?” Smith pointed first to cost. “It's largely unknown how much the costs will eventually get to if you build your own, because AI requires tokens and data space,” she said.

As Smith told webinar attendees, AI should not be viewed purely to automate manual processes. “It actually is there to improve the success of those collections as well.”

That may be the more important change-up.

Collections has traditionally focused on managing overdue debt efficiently. Agentic AI creates the possibility of making better decisions earlier, applying the context surrounding each customer and learning from the actions that result in payment.

The question changes from which invoice we should chase next to a much more valuable one: What should we do next to get paid?