Summary: Portfolio averages are improving while borrower-level stress concentrates, so the average account no longer describes anyone. A collection strategy answers five questions per stage and per customer: who to engage, when, through which channel, with what offer, and how success is measured. The stage sets the objective, and behavioral signals set the treatment inside it.

What should you send to a customer who is 45 days late, and how is that different from what you send at day 5? A collections strategy answers that question for every account in the book. It decides which past-due accounts to contact, when, on which channel, what to offer them, and how you will know it worked. The work splits into five stages: pre-collections before anyone misses, 1 to 30 days, 31 to 60, 61 to 90, and the handoff to recovery or write-off.

Lenders and telecom operators run the same five stages with very different leverage inside them.

A collection strategy built on portfolio averages serves nobody

At first glance, the numbers look good. The Philadelphia Fed’s Q2 2026 Insights Report found that late payments and write-offs at large banks have been falling for over a year. More cardholders are clearing their balance in full than at any point since 2021, at 37%.

The same quarter tells a different story underneath. The Fed’s Consumer Credit Explorer found that nearly 14% of credit card borrowers are now seriously behind, and those borrowers cluster in particular cities rather than spreading evenly. Both figures cover US credit cards, though the pattern shows up elsewhere too.

The Fed explains the gap itself. Some of the improvement comes from lenders approving safer borrowers rather than from existing customers doing better, and plenty of households remain under real financial pressure.

A collection strategy gives each stage its own objective

Most programs carry one objective from the first reminder to the final notice: get paid now. That is why messaging escalates in tone while results stay flat.

Stage Objective Who it mostly reaches Channel that fits Offer on the table Metric that proves it worked
Pre-collections Prevent the miss Accounts with a risk signal last cycle SMS or in-app Due-date change, autopay repair % of flagged accounts paying on time
1–30 days Resolve a one-off Customers who forgot or had a card fail SMS, email, in-app One-tap payment, short extension Cure rate within the bucket
31–60 days Stop the pattern forming Customers prioritizing other bills SMS plus outbound voice Split payment, revised due date Promise-to-pay kept rate
61–90 days Prevent the roll Customers who have disengaged Voice plus one written channel Payment plan or hardship route Roll rate to the next bucket
Handoff Decide the exit Accounts unlikely to cure in-house Formal notice Recovery, placement, sale or write-off Recovery value net of cost

The design test is simple. If the objective in two different rows reads the same, the program is a schedule rather than a strategy, and collection design is what closes that gap.

Behavioral signals separate customers inside the same stage

Two customers sit at 61 days past due. One had a card expire and has not noticed. The other is choosing between rent and the bill. They share a status code, a queue, and a message sequence. The first needs 30 seconds of friction removed. The second needs an affordable arrangement.

Four signals already in most servicing data do separate them:

  • A partial payment last cycle, which signals prioritizing rather than refusing.
  • A broken promise to pay, the strongest predictor that this cycle slips too.
  • Channel response history: who opens, who replies, who has gone quiet.
  • Whether past arrangements were completed or abandoned.

The concentration finding above is the operational point. Where severe delinquency clusters in particular places, a book contains pockets where the standard path performs far worse than the average implies, and the average hides them.

Pre-collections is the cheapest stage in any collection strategy

The cheapest account to work is one that has not missed a payment yet. There are no arrears, no hardened avoidance, and the customer still thinks of themselves as someone who pays on time. Pre-due contact also generally sits outside the rules governing collections contact, though consent and frequency obligations continue to apply.

A designed pre collections programme contains four things:

  • Due date alignment. Fixed due dates imply fixed pay cycles and it's often cheaper to just move the date than to send any kind of message.
  • Autopay and card failure recovery. Many first misses are mechanical, and treating them as a fix rather than a collection resolves them faster.
  • Risk-signal triggers. A partial payment or a broken arrangement in the last cycle gets you a pre-due contact this cycle.
  • Hardship awareness. Where a plan or discounted tariff exists, eligible customers need to know about it before they miss.

The lender version of this is a payment plan or a date change. The telecom version is a social tariff or bill smoothing. Either way, pre-collections rewards design attention more than any later stage.

Cure and roll rates prove a collection strategy is working

The most commonly reported collections metric is also the least informative. Contact rate counts attempts rather than results, and a team judged on it will contact customers more often, which generates complaints.

Metric Plain definition What it tells you Common mistake
Cure rate Share of delinquent accounts returned to current within a set period Whether the treatment worked Counting an arrangement agreed rather than an arrangement completed
Roll rate Share of accounts moving from one delinquency bucket to the next Whether treatment is holding the line or delaying the slide Reading it monthly without seasonal adjustment
Promise-to-pay kept rate Share of agreed promises that result in payment How reliable your arrangements actually are Treating a promise as recovered revenue
Collection effectiveness index Amount collected as a share of amount available to collect Overall efficiency across a period Comparing across portfolios with different age profiles
Time to resolution Days from first delinquency to cure or exit How quickly the program works Measuring to first contact rather than to outcome
Repeat delinquency after cure Share of cured accounts delinquent again within 90 days Whether the cure held Not measuring it at all

Outcome metrics show whether money came back. Guardrails show what it cost in customer relationships.

If cure rate rises while guardrails turn red, the strategy has borrowed rather than worked. The same risk appears when evaluating an AI collections agent, since a system measured on the wrong signal gets better at the wrong thing. Fuller definitions sit in the collections KPIs glossary.

Payment optimization makes stage decisions per account

Payment optimization is the layer that makes them per account instead. KredosAi sits on top of a collections management system rather than replacing it, deciding for each customer who to contact, when on this channel, and with what message and offer, adjusting as that customer responds.

Across KredosAi’s enterprise portfolio, write-off rates fell 11.5% while customer lifetime value rose 13.6%. The outcome metric and the guardrail metric moved together, which is the test the previous section describes. Whether to buy that layer or build it internally depends on integration cost and internal capacity, which build versus buy addresses.

Balance size decides how each industry runs the same stages

Dimension Telecom Lender
Typical balance Tens to low hundreds Hundreds to tens of thousands
What’s at stake for the customer Losing service they use daily Credit file damage, asset repossession
Mid-stage leverage Service suspension or restriction Reporting, fees, collections escalation
Regulatory frame Communications regulator, consumer law Financial conduct rules, FDCPA, Regulation F
What cure means Account current and service restored Account current, arrangement completed
What follows the worst case Disconnection, then win-back Charge-off, placement or sale

Balance economics decide channel, because a small telecom balance cannot support the per-account cost that a large loan can. Suspension is leverage lenders do not have, and it is the same action that converts a payment problem into a churn event. Telecom also has a second act in win-back that post-charge-off lending mostly lacks.

Five questions reveal whether you have a collection strategy

  1. Does the objective change between stages, or is it get paid now throughout?
  2. Can you report cure rate by segment rather than only in aggregate?
  3. Is there a documented reason for your contact frequency, or was it inherited?
  4. What does your program do when a customer genuinely cannot pay?
  5. Do you track what happens to cured accounts 90 days later?

Three or more answers in the negative means the program is a schedule.

The strongest programs decide stage by stage and signal by signal

The portfolio underneath the averages is separating into groups with different problems. The collection strategies that work now decide stage by stage and signal by signal what each customer actually needs.

FAQ

What is a collections strategy?

The set of decisions around which accounts to engage at each stage, via which channel, with what offer and how success is measured. A dunning schedule applies a single fixed sequence to everyone, that's the difference.

What are the stages of collections?

Pre-collections, early stage (1 to 30 days), mid stage (31 to 60 days), late stage (61 to 90 days) and handoff to recovery/sale/write-off.

What is a cure rate?

The share of delinquent accounts returned to current within a set period. It is the most direct measure of whether a strategy is working, provided it counts completed arrangements rather than agreed ones.

What is roll rate?

The share of accounts moving from one delinquency bucket to the next. Tells you whether treatment is actually holding the line or just delaying the slide.

How is telecom collections different from lending?

Smaller balances, service suspension as a form of leverage, a different regulatory regime and a genuine win back opportunity post disconnection.

Should collections be segmented by days past due or by behavior?

Both. The stage sets the objective, and behavioral signals set the treatment within it.