Anyone who’s spent time around collections knows that an aging report only tells part of the story. Three overdue invoices from the same customer can mean three different things: someone’s waiting for proof of delivery, someone’s arguing about the price, and someone paid last week but the cash is still sitting unapplied. By the time a collector has worked through the emails and figured out who knows what, half the morning is gone and the customer has received another reminder about a bill they think they’ve already settled.

That’s a useful place to start with AI in working capital. Give a collector the invoice history, the relevant correspondence, and a clear explanation of what’s holding up payment, and they can spend more of the day resolving problems. The technology can help assemble that picture, but the financial result depends on what happens next: whether the missing paperwork gets sent, the dispute gets resolved, and the customer actually pays sooner.

There’s plenty of reason to examine that process. On October 5, The Hackett Group reported a $773 billion accounts receivable opportunity among North America’s largest nonfinancial companies, up 29% from the prior year. Its analysis of FY2025 results from 1,000 companies also found that days sales outstanding worsened by 2.1 days. Those figures describe performance against working capital benchmarks, so a corporate CFO or PE operating team still needs to work through its own balances before estimating what better collections could release. The $773 billion isn’t a forecast of cash AI can recover.

Microsoft’s June account of its collections redesign gives a practical example of the work involved. Its global team has more than 1,000 collectors, and the system helps case managers prepare cases, prioritize accounts, route inquiries, and draft responses. The company reported a 40% reduction in call preparation time, while noting that its figures are internal and directional; data consolidation, standardized processes, and training changed alongside the technology. The account doesn’t show how much improvement came from AI alone or how much sooner customers paid. Faster cash application, another reported result, helps keep the ledger and collection decisions current, but the customer has already sent that money.

Oracle’s 26C Collector Workspace describes similar assistance, with prioritized worklists, invoice and dispute context, payment histories, promises to pay, and communication drafting. That documentation shows what the workspace is designed to do; a finance team still has to establish the return in its own operation. Dynamics 365 payment predictions uses machine learning to estimate whether an invoice will be paid on time, late, or very late. A team could pair those estimates with a separate generative assistant to summarize correspondence and draft responses, checking each job independently because a fluent summary can make a shaky prediction sound more convincing than it deserves.

For a mid-cap manufacturer or an enterprise finance team, a manageable pilot would cover one entity or customer segment, combining aging, invoice acceptance, disputes, payment history, unapplied cash, and promises to pay. Once those inputs tie out, the queue can give each case a reason, supporting records, a proposed next step, and an owner: Sales for commercial terms, Billing for missing documents, Collections for broken promises. Receipts and disputes need to refresh that view. Collectors review messages, and credit notes, write-offs, and term changes retain existing approvals. The treasury trust boundary applies here too, with permission-scoped access and logged evidence and decisions, including separate customer access for each portfolio company.

The pilot also needs a comparison, whether that’s a similar customer cohort or a phased rollout with a baseline. Track cash receipts, dispute resolution time, promises kept, and cash collected per collector hour, reconciling receipts to bank and ledger records and recording the preceding action. Customers may have paid anyway, so compare payment timing and collection rates while accounting for seasonality, customer mix, terms, write-offs, and factoring. Keep preparation hours visible as well: productivity can improve because collectors spend less time on the work even when cash receipts stay flat. That can support an investment, provided the business case reflects what improved. Record which recommendations collectors accept or reject, along with their reasons, so the next round of work addresses the errors they actually encounter.

For a portfolio company CFO, a receivables release improves liquidity and can reduce borrowing needs, but the cash benefit is generally one-time at the new operating level. Sustaining it takes continued billing and collection discipline, while interest savings and operating cost reductions need separate measurement alongside implementation costs and ongoing fees. A lower receivables balance doesn’t automatically increase EBITDA. Over the next quarter, the pilot should reveal which payment obstacles keep recurring and where collectors need better information. As with the broader accounting use cases, expansion follows reconciled data, resolved exceptions, and validated results. The next entity gets a tested workflow, and Billing gets a useful list of problems to stop creating.