Payment behavior leaves signals weeks before an invoice goes overdue. We analyzed 18 months of B2B payment data to map the early warning patterns.
The Case for Looking Earlier
The standard AR workflow is reactive by design. An invoice goes out. The due date passes. The aging report turns amber. The AR specialist sends a reminder. That sequence is so embedded in how we think about collections that questioning the starting point — "what if you acted before the due date?" — feels almost counterintuitive. You can't chase an invoice that isn't overdue yet.
But "acting before the due date" doesn't mean sending dunning emails to invoices that haven't missed payment. It means identifying, before the due date, which invoices in your open AR balance have a meaningfully elevated probability of going 60+ days past due — and routing those accounts to higher-priority outreach earlier in the cycle.
This is a different kind of problem than detecting late payment. It's predicting it. And the question we wanted to answer when we started building Cashvyne was: is there actually enough signal in historical payment data to make useful predictions at the invoice level, before the due date?
The answer is yes — with important caveats about what "useful" means in practice.
What the Payment Data Contains
When we look at 18 months of B2B payment transaction data across mid-market portfolios — companies with roughly $10M-$100M in annual revenue and 50-300 active customer accounts — several patterns emerge consistently.
The most predictive feature for any given invoice is the recent payment trend of that specific customer. Not their average days-to-pay over all time, but the direction of change over the last 3-4 invoices. A customer who paid at net+28 for two years but whose last three invoices came in at net+38, net+44, and net+51 is on a trajectory that a static average completely obscures. Their historical average might still show net+31 — technically "acceptable" — while the recent trend is pointing toward serious aging.
The second strong signal is invoice size relative to the customer's typical transaction volume. An invoice that's 2.5x or more above a customer's median invoice size shows elevated late-payment probability, independent of the customer's general payment behavior. This makes intuitive sense: larger-than-usual invoices often require additional approval steps, budget allocation decisions, or simply more internal scrutiny before payment is authorized. That friction shows up in payment timing.
The third signal is invoice seasonality in context. Some customer payment delays are seasonal — a manufacturing customer may consistently slow payments in Q4 as they close their own books, or a distributor may tighten cash management during inventory build cycles. These patterns are visible in multi-year payment history and produce false positives in systems that don't account for them.
The Signals That Don't Hold Up
We want to be precise about this, because overstating predictive accuracy is a real problem in how AR intelligence gets marketed.
Days Sales Outstanding at the customer level — the most commonly cited AR health metric — is a poor predictor of individual invoice outcomes when used in isolation. DSO is an averaging mechanism. It smooths out the variance that actually matters for prediction. A customer with a DSO of 38 might be composed of 80% invoices that paid at net+28 and 20% that went to net+72 — but you don't see that distribution in the aggregate number.
Invoice age alone is also weak as a forward signal for 60+ DPD risk. The fact that an invoice is currently at net+12 tells you almost nothing about where it will be at net+60, unless you combine that age signal with the behavioral history of the customer.
Accounts that are brand-new — first invoice, no prior payment history — present a genuine predictive blind spot. Without a behavioral baseline, you're working from industry base rates and whatever you know about the customer's company characteristics. This isn't nothing, but it's materially less reliable than the predictions you can make on customers with 6+ months of payment history.
How Early Is "Early Enough"?
In practical terms, identifying a high-risk invoice with 8-12 days before its due date gives an AR team meaningful room to act. That window allows for a proactive outreach call — not a dunning email, but a relationship-forward check-in — that often surfaces information the invoice is heading toward a payment delay for reasons that could be resolved early: a disputed line item, a missing purchase order number, an internal approver on leave.
Many of the 60+ DPD invoices we see in mid-market portfolios got there not because of customer inability or unwillingness to pay, but because of process friction that nobody addressed while the invoice was still in net-30. An incorrect billing address. A quantity discrepancy that the customer noted internally but never escalated back to the supplier. A change in the customer's AP team that created a temporary gap in invoice processing. These are recoverable situations — but only if you surface them before they're 45 days old.
Consider a scenario: a food-service equipment distributor in the upper Midwest manages roughly $5M in open AR at any given time across about 80 active accounts. Their average invoice is $12,000-$15,000. When we look at their historical data, invoices that went 60+ DPD in the previous 12 months showed an average days-to-pay trend shift of +14 days across their prior 3 invoices — visible well before the current invoice's due date. In at least 60% of those cases, a proactive outreach in the net-30 window would have found a resolvable issue. Instead, those conversations happened at net+45 or later, after the friction had solidified into a more adversarial collections dynamic.
Predictive Score vs. Aging Bucket: Different Information
An aging report tells you what has already happened. An invoice risk score tells you what is likely to happen. These are not redundant views — they answer different questions.
Your aging report says: "You have $340,000 in the 30-60 day bucket." A risk-scored view of your open AR says: "Of your $2.1M in open invoices that haven't missed payment yet, approximately $180,000 is in accounts showing elevated 60 DPD risk over the next 30 days." The second piece of information is actionable in a way the first isn't — it tells you where to put AR effort now, not where the damage already is.
This distinction matters especially for CFO-level cash flow forecasting. An aging report says "$340K is currently at risk." A predictive view says "$340K is currently overdue and $180K more is likely to join it." The difference in cash position implication is significant for a company trying to manage working capital against a credit line or planned expenditure.
What This Doesn't Solve
We're not saying predictive scoring replaces the judgment of an experienced AR manager. A customer going through a bankruptcy proceeding, a dispute that has escalated to legal review, or a business relationship where the sales team is actively managing a large upsell — these situations require human context that no payment trend model captures adequately.
Prediction also doesn't eliminate write-offs. A customer who genuinely can't pay won't be saved by earlier outreach. What prediction changes is the distribution of effort: more early attention on the accounts where intervention is still possible, less reactive chasing on accounts that needed a different kind of conversation from the start.
The goal of building predictive intelligence into the AR workflow isn't to automate away judgment — it's to surface the right accounts for judgment earlier, when options are still open. That's the gap between an AR team that's always fighting fires that started six weeks ago and one that occasionally gets to prevent them.