How to Segment Customers by Payment Behavior for Adaptive Dunning

Every AR team eventually notices the same thing: a handful of customers pay like clockwork and a handful drag every invoice out. Then they send the same three-touch cadence to both, because the system does not know the difference.

Adaptive dunning is the discipline of teaching your process the difference, so the reliable payers get lighter service and the erratic ones get earlier pressure. It is one of the two or three highest-leverage changes an AR lead can make.

What are the four customer segments in adaptive dunning?

The clean framework uses two axes: average days to pay past due date, and variance in that number.

The insight most teams miss is that predictably slow is not the same as erratic. A customer who consistently pays on day 42 is not a collections problem; they are a term-mismatch problem. A customer who pays anywhere from day 5 to day 55 is a collections problem, even if the average looks fine.

How do you actually classify each customer?

For each customer, compute the last 12 months.

• Days past due. For each closed invoice, days between due date and payment received date. Negatives (paid early) count as zero for this purpose.
• Standard deviation. Of the days past due values across all closed invoices in the period.
• Volume. Number of invoices in the period. Fewer than 6 invoices is not enough data; classify as "insufficient history" and use a default cadence.

Then apply the thresholds.

• Mean under 10 and std dev under 8: Reliable
• Mean 10 to 30 and std dev under 12: Predictably slow
• Std dev above 15, regardless of mean: Erratic
• Mean above 30: Chronic late

Do this quarterly. Segment membership is stable enough that monthly reclassification is noise, and slow enough that annual is too late to catch a customer whose behavior has changed.

What does the sequence look like for each segment?

The cadences differ in three dimensions: number of touches, timing of first touch, and tone at each stage.

Reliable segment.

• Touch 1: pre-due confirmation, day minus 5, light and administrative
• Touch 2: only if invoice is 3 days past due, gentle nudge, one line
• Escalation: never in the automated flow. If a reliable account goes past 15 days, the AR lead makes a personal call, not a systematic escalation.

Predictably slow segment.

• Touch 1: pre-due confirmation, day minus 5
• Touch 2: day 5 past due, neutral reminder
• Touch 3: day 20 past due, request dated commitment
• Touch 4: day 40 past due, firmer, name a plan
• Touch 5: day 55, escalate to buyer

The shifts here mirror the customer's actual payment cycle. A firm day-15 email to a customer who has always paid at day 30 will just annoy them.

Erratic segment.

• Touch 1: pre-due confirmation, day minus 7 (earlier)
• Touch 2: day 1 past due, neutral reminder, request payment date
• Touch 3: day 10 past due, request dated commitment
• Touch 4: day 25 past due, escalate to buyer, log promise-to-pay
• Touch 5: day 40, account-manager loop

Erratic customers respond to structure. Getting them to commit to a dated remit early is what shifts their behavior over time.

Chronic late segment.

• Touch 1: pre-due confirmation, day minus 7, with a phone check-in
• Touch 2: day 1 past due, phone plus email, request commitment
• Touch 3: day 15, account-manager involvement, credit review flag
• Touch 4: day 30, CFO or VP finance conversation, plan or hold
• Touch 5: day 45, escalation, credit hold, or new-order block

The strategic question with chronic late payers is not how to collect one invoice. It is whether you continue to extend credit. Cadence alone will not fix a customer who is genuinely a credit risk.

How does tone shift by segment?

Cadence is the timing. Tone is the register. Both need to differ.

• Reliable. Assume competence. "Just confirming this is in your queue" sounds like a workflow check, not a chase. Never use "please" more than once in an email to this segment.
• Predictably slow. Neutral, clear, and non-emotional. State the fact, request the commitment, thank them.
• Erratic. Firmer and more specific. "Can you confirm the exact date this will remit" rather than "any update." Do not accept vague answers; log every conversation.
• Chronic late. Direct, without hostility. Name the pattern, name the consequence, offer a plan. A payment plan agreed at day 20 collects more than a lawsuit filed at day 90.

Tone that does not match cadence undoes the segmentation. A firm day-1 email to a reliable account, or a light day-30 email to a chronic account, tells the customer your system is not paying attention.

How do you handle segment changes mid-cycle?

Behavior shifts, sometimes sharply. An acquisition, a new AP lead, or a leadership change on the customer side can move an account from reliable to erratic in one cycle.

• Detect. Any account whose most recent 3 invoices sit more than 2 std devs above their historical mean should trigger a review flag, not an automatic reclassification.
• Investigate. Before moving them to a firmer segment, understand why. A one-time delay from an internal transition is different from a persistent shift.
• Communicate. For strategic accounts, escalate to the account owner before changing the segment. "We are seeing longer cycles on your invoices, is there anything on your side we should know about" is a conversation, not a system change.
• Reclassify only after 3 to 4 more invoices. Sample size matters. One late payment is noise, three is signal.

What tooling do you actually need?

Adaptive dunning is not implementable in a spreadsheet past 30 or so customers, because the segmentation math and the branching cadences require state per invoice.

The minimum viable stack:

• Payment history at invoice grain, not customer grain
• Segment classification refreshed monthly
• Sequence engine with branching by segment
• Promise-to-pay logging tied to the invoice
• Dispute state that pauses the automated cadence
• Manual override on any account, one click

Most AR-specific software handles this out of the box, on top of your accounting system. General email automation does not, because it does not have the invoice grain to segment properly.

The mistake to avoid

The seductive mistake is over-personalizing. Adaptive dunning is not about writing bespoke emails per customer. It is about running four cleanly differentiated sequences against four segments defined by objective payment behavior. The other seductive mistake is under-differentiating, running the same cadence against everyone because "our customers are all similar." They are not, and the aging report will prove it. Segment on data, not intuition, and let the cadence do the work.