Cure Rates: What Actually Brings a Delinquent Account Back

A collections operation reports a 78% cure rate on early-stage delinquency and treats it as evidence the process works. It's evidence of almost nothing. Most accounts one cycle past due cure without any collections activity at all — the card expired, the payment date fell during a trip, the transfer didn't arrive, and the customer noticed and paid. An operation working those accounts is taking credit for outcomes that would have happened regardless. The number that measures whether collections works is incremental cure — treated minus untreated — and calculating it requires deliberately not contacting some accounts, which almost nobody does.

The natural cure rate

The concept most collections reporting doesn't contain: the share of accounts that would return to current with no contact whatsoever.

It's high early and falls sharply. The shape, illustratively:

Stage	Natural cure rate	What's in the population
1–15 days past due	Very high	Almost entirely oversight
30 days	High	Oversight plus short timing gaps
60 days	Moderate	Genuine difficulty entering
90 days	Low	Mostly inability
120+ days	Very low	Structural difficulty

Why the early population cures itself: most of it never had a payment problem. A card on file expired. A payment date landed while someone was away. An account had insufficient funds for two days. The customer has the money and simply didn't transfer it — and once they notice, they do.

Which produces the finding that reorganizes the whole metric: lift from collections effort is smallest exactly where cure rates look best. The stage with the most impressive gross numbers is the stage where your contribution is closest to zero, and the stage with the worst gross numbers is where treatment does the most.

Why gross cure misleads

Three specific errors it produces.

It rewards working easy accounts. An operation measured on gross cure has every incentive to concentrate on early-stage accounts, where the number will look excellent regardless of what they do. The metric rewards the allocation that adds least value — which compounds the misallocation our recovery frontier analysis identifies.

It makes mix changes look like performance. An operation whose delinquency population shifts toward early stage will report improving cure rates having changed nothing — the same composition effect our benchmarking analysis describes for recovery rates.

It makes every intervention look successful. Add a new treatment to early-stage accounts, observe a high cure rate, conclude it worked. The cure rate would have been high anyway, so the test can't fail — which means an operation evaluating changes this way will accumulate treatments that do nothing and believe all of them work.

That last point is the expensive one. It's the same structural failure our fraud analysis identifies with fraud rate: a metric that always moves in the favourable direction cannot evaluate anything.

Measuring incremental cure

The fix is a holdout, and it's cheaper here than almost anywhere else in credit.

Incremental cure = Cure rate (treated) − Cure rate (untreated)

Design:

1. Randomly withhold treatment from a small share of accounts at each stage — 5% is usually sufficient at reasonable volume.
2. Genuinely withhold it. No calls, no messages, no letters. Statements and any legally required notices continue.
3. Measure cure at a fixed horizon for both groups.
4. Compute the difference — that's your lift.
5. Repeat by stage and segment, because lift varies enormously across both.

Worked illustration. An operation runs holdouts across stages:

Stage	Treated cure	Untreated cure	Lift
15 days	91%	88%	3 pts
30 days	78%	66%	12 pts
60 days	54%	31%	23 pts
90 days	32%	17%	15 pts
120 days	19%	11%	8 pts

Read the lift column rather than the treated column. The 15-day stage reports the best cure rate and produces the least lift. The 60-day stage reports a mediocre 54% and is doing nearly eight times as much work.

The allocation implication is direct: an operation spending heavily at 15 days and lightly at 60 has it backwards, and it will never discover this from gross cure rates, because those point the opposite way.

Note also the shape — lift peaks in the middle and declines at both ends. Early, the accounts cure anyway. Late, they mostly don't cure regardless. Which is the general result and it generalizes beyond this example.

Where the lift actually is

Generalizing: treatment matters where the outcome is genuinely uncertain, and nowhere else.

Three bands:

• Near-certain cure. Contact adds almost nothing. Treat cheaply — an automated reminder — or not at all.
• Genuinely uncertain. All the value is here. A reminder, a payment path, or an affordable arrangement changes the result.
• Near-certain non-cure. Treatment rarely changes the outcome, though it may still be worth low-cost contact given the option value of circumstances changing.

Which means the modelling task is different from the one most operations run. Predicting who will default is a different question from predicting where contact will change the answer — and the second is what should drive allocation. An account with a 92% chance of curing and one with a 6% chance are both poor targets despite being at opposite ends of any risk score.

What predicts membership in the uncertain middle:

• First-time versus repeat delinquency. A customer with a clean history who has just gone past due is very likely oversight; one with a pattern is not.
• Payment method failure type — an expired card is different from insufficient funds, and the return codes in our bank payments guide distinguish them.
• Prior cure behaviour, which is the strongest single signal.
• Balance relative to typical payment.
• Contactability, since an unreachable account gets no treatment effect by definition.
• Whether the customer initiated contact — a strong indicator of engagement.

Forgetting versus inability

The distinction that determines which treatment works, and it maps onto the temporary-versus-structural framing in our forbearance analysis.

	Forgetting	Inability
Has the money	Yes	No
Typical stage	Early	Deepening
Responds to	A reminder and a payment link	An affordable arrangement
Cost to resolve	Near zero	Real effort
What a call achieves	Little more than a message would	Establishes capacity
Wrong treatment	Escalation, which damages the relationship	Reminders, which annoy without helping

The bottom row is where operations lose money in both directions. Escalating a forgetter damages a good customer over an oversight. Reminding someone who can't pay accomplishes nothing and makes them stop answering — which destroys the contactability needed later, when an arrangement might have worked.

The diagnostic is cheap: ask. The finding from our forbearance analysis applies here in miniature — the determining variable is something the customer knows and no model does, which makes the quality of one conversation worth more than the sophistication of the routing that produced it.

Matching treatment to cause

For likely forgetting:

• A message with a working payment link — the cheapest effective treatment in collections, and the one most operations under-deploy relative to calling.
• Card updater services for expired credentials, which prevent the delinquency rather than curing it.
• Pre-due reminders, which move the intervention before the event and cost less than curing after.
• A due date change where the timing mismatches income, which permanently fixes a recurring problem.
• Retry logic timed to likely funding, per our payment failure guide.

For likely inability:

• A conversation establishing capacity, not a demand.
• An affordable arrangement sized to what they can sustain, per our plan design guide.
• Partial payment acceptance, which establishes a pattern and is worth more than a full promise.
• Hardship options where the difficulty is temporary.
• Self-service arrangement setup, so a customer avoiding a conversation can still resolve.

The point that ties them together: the treatment for forgetting is almost free and the treatment for inability is not. Which means correctly identifying which you're dealing with is where the cost efficiency lives — an operation calling every early-stage account is paying conversation prices for reminder problems.

Cures that don't last

The failure mode gross cure rate conceals entirely: an account that cures and re-delinquents within two cycles was not resolved.

Why it matters:

• It consumed treatment cost twice or more.
• It counted as a cure each time, inflating the metric.
• The underlying problem — affordability, timing, a broken payment method — was never addressed.
• Repeat cyclers are a distinct population requiring a structural fix rather than repeated collection.

What to track: cure durability at 60 and 90 days after cure. An operation with a high cure rate and poor durability is spending repeatedly on the same accounts and reporting each cycle as a success.

The intervention for repeat cyclers is almost never more collections. It's a due date change, a payment method fix, or an arrangement sized to actual capacity — a one-time structural change that ends a recurring cost, and identifying these accounts is one of the higher-return analyses available.

What to report

• Incremental cure by stage, from holdouts. The headline number, replacing gross cure.
• Cost per incremental cure, which is the efficiency measure and the one that supports allocation.
• Natural cure rate by segment, maintained as a standing baseline.
• Cure durability at 60 and 90 days.
• Repeat cycler population and what's being done structurally about it.
• Roll rates alongside cure, since they capture the flow cure rates miss — per our metrics guide.
• Treatment cost by stage, so allocation can be compared against lift.

And the deletion: remove gross cure rate from any dashboard where it could drive decisions. Keep it as a diagnostic if you like. Reporting it as performance is what causes operations to concentrate effort where it does least and to conclude that every intervention works.

Frequently asked questions

---

This guide is educational and does not constitute legal or compliance advice. Worked figures are stylized illustrations rather than benchmarks. Communication requirements and required notices continue to apply to accounts in any holdout group; consult qualified counsel before withholding treatment.