Corporate finance teams have spent decades building systems to record what customers owe. Very few of those systems tell them what customers are about to do.
That gap is what artificial intelligence vendors and banks are now targeting. Accounts receivable has long been run as a back-office chore, something organizations assumed would sort itself out as long as invoices were sent on time. Economic pressure and rising delinquencies are forcing a rethink.
Suppliers of goods and services are chasing delinquency on buyers more frequently, driven by tariffs, economic pressures and fluctuating interest rates, Billtrust Vice President of Software Products Dave Ruda told PYMNTS. The cost of that inaction has become quantifiable.
The Hackett Group’s U.S. Working Capital Survey, based on an analysis of the top 1,000 U.S. publicly traded nonfinancial companies, found that $1.7 trillion remains trapped in excess working capital. Accounts receivable accounts for the largest share of that total, an opportunity valued at $600 billion, as DSO saw its second straight year of degradation driven by customer bargaining power and extended payment terms.
That scale of trapped capital is what’s drawing investment at speed.
From aging buckets to behavioral intelligence
The change in AR isn’t just about speed. It’s about what the systems now know.
Advertisement: Scroll to Continue
Traditional AR reporting works as a lagging indicator: Teams compile aging reports at month’s end, categorize overdue invoices by days past due, and assess risk using historical averages or static credit scores, according to PYMNTS Intelligence. AI models embedded in ERP systems can now forecast the likelihood of a specific invoice being paid late before it is sent, drawing on structured data like payment history and invoice size alongside unstructured data like sentiment from customer emails and dispute frequency.
Purpose-built AR platforms take this further by applying a behavioral layer on top of ERP data. Rather than treating each transaction as an isolated event, these systems incorporate historical payment patterns to guide next steps automatically. A short payment is typically recorded as a variance in an ERP, an exception to be investigated.
A purpose-built AR platform applies contextual intelligence: It knows the customer’s behavior and keeps the cash moving, Billtrust Chief Product Officer Lee An Schommer told PYMNTS.
The metrics follow. Companies using purpose-built AR platforms that combine invoicing, payments and collections see a 23% reduction in days sales outstanding and a 25% reduction in days to pay, with an additional 34% reduction when all three functions operate together, Schommer said.
Banks move to capture the AR automation market
Institutional players are now moving to own a larger share of the AR stack.
Truist Financial launched an AI-enabled integrated receivables platform in February, using machine learning to automate the matching of payments to invoices for commercial and corporate clients, the company said. The platform scans both traditional checks and electronic payment rails, applies business rules to auto-match payments, and includes a smart remittance capture feature that extracts data directly from emails to minimize exceptions.
The launch comes as North American mid-market firms show widening performance gaps based on how aggressively they’ve modernized AR. Firms integrating digital tools into their working capital strategy are realizing materially stronger bottom-line results than those that do not, with AI most commonly applied in customer onboarding, identity verification and financial planning, according to the 2025–2026 Growth Corporates Working Capital Index produced by PYMNTS Intelligence in collaboration with Visa.
Where adoption still lags
The gains aren’t reaching most companies yet.
Eighty-three percent of firms have yet to fully automate their AR operations, according to PYMNTS Intelligence data from the “From Friction to Flow: AR Automation in 2025” report. Data fragmentation is a primary constraint.
Companies manage an average of three ERP systems, creating data silos that make it difficult to build a unified view of customer behavior, payment history and dispute patterns, Schommer said. Without consolidated data, predictive models produce unreliable outputs. The intelligence layer is only as useful as the infrastructure beneath it.
Among North American mid-market firms, U.S. companies continue to rely more heavily on traditional payment methods than their Canadian counterparts, and where card acceptance is more widely used as a receivables strategy, revenue losses are lower, the PYMNTS-Visa index found. The gap is driven not by customer behavior but by firm-level choices about payment infrastructure.
