The Promise to Pay: Why Collections' Favorite Metric Predicts So Little

Nearly every collections operation tracks promises to pay, reports them to management, and in many cases compensates on them. It's an appealing metric — it's countable, it happens during the call, and it feels like progress. It is also among the weakest predictors of recovery in the entire function, and measuring collectors on it makes recovery worse. The reason is structural: a promise is what ends an uncomfortable conversation. It costs nothing to make, requires no capacity at the moment it's made, and is frequently sincere in a way that has nothing to do with whether the money will exist. This guide covers why kept rates vary so much, what predicts better, and what to measure instead.

Why promises break

The mechanism is worth stating precisely, because the usual explanation — that people are evasive — is mostly wrong and leads to the wrong response.

Consider what a promise costs the person making it. Nothing, at the moment it's made. No money moves. No capacity is required. The obligation arrives later, and the immediate effect is that a stressful conversation ends.

Now consider who's on the call: someone under financial pressure, frequently embarrassed, who wants the interaction over. Agreeing is the cheapest available way to achieve that — and it is usually sincere. The person genuinely intends to pay. The intention is real; the money is what's missing.

Which produces the analytical point: a promise measures the outcome of a conversation, not the state of a household's finances. Those correlate weakly, and the correlation gets weaker the more pressure was applied to obtain the promise — since pressure raises the value of ending the call without changing capacity at all.

This is why the standard interpretation misleads. A high promise rate can indicate collectors who are good at obtaining agreement, which is not the same as identifying accounts that will pay. And a low promise rate can indicate collectors having honest conversations that correctly conclude the person cannot pay this month — which is more useful information than a promise that won't be kept.

Promises are not one thing

The category conceals enormous variation. Sorted roughly by how well each predicts payment:

Type	What it looks like	Predictive strength
Payment taken now	Money moved during the conversation	Certain — not a promise at all
Scheduled and authorized	Amount, date, and authorization captured	Strong
Specific and anchored	"$180 on the 15th, when I get paid"	Moderate
Specific, unanchored	"$180 on the 15th"	Weak
General intention	"I'll take care of it this month"	Very weak
Pressured agreement	Agreement obtained after resistance	Near zero

Two properties separate the top of that table from the bottom.

Whether money moved. Any commitment involving an actual transaction — a payment now, an authorization for later — is categorically different from a statement of intent. It required the person to have funds or an account and to take an action, both of which are evidence.

Whether it's anchored to an income event. "The 15th, when I get paid" is materially stronger than "the 15th," because the person has reasoned about capacity rather than picked a date. Asking "when do you next get paid?" and setting the date to that is one of the cheapest improvements available to any collections script.

The operational implication: reporting a single promise-kept rate across all of these is averaging things with different meanings. An operation that segments its kept rate by promise type immediately learns which conversations produce money — and typically finds the spread between the top and bottom rows is very large.

What happens when you measure it

The most important section, because this is where the metric does active damage.

Suppose collectors are compensated on promises obtained. The predictable result: more promises, of lower quality.

The mechanism is straightforward. A collector facing a person who says they can't pay has two options — accept it and log the outcome, or keep pushing until the person agrees to something. If the metric rewards agreement, the second is what happens. The additional promises obtained this way come disproportionately from people who resisted, which is to say from people whose resistance reflected accurate knowledge of their own finances.

The measured output rises. The cash doesn't. This is a textbook proxy failure: a metric chosen because it correlates with the objective stops correlating once it becomes the target, because the behavior that moves the metric isn't the behavior that moved the objective.

Three secondary harms, each real:

• Pressure to obtain agreement creates compliance exposure. The line between persistence and conduct that violates the standards in our compliance guide is one a collector under quota pressure is more likely to approach.
• It damages the relationship — significant for a first-party creditor whose customer may buy again.
• It destroys information. "This person cannot pay $400 this month" is useful and would route the account toward a plan or a workout. Converting that into a promise that breaks discards the finding and wastes the subsequent cycle.

The fix is not subtle: measure collectors on dollars collected and on payments taken during contact. Both are countable, both are the actual objective, and neither can be gamed by obtaining agreement.

What predicts payment better

Ranked, with what each actually reflects:

• A payment taken during the conversation. Not a predictor — an outcome. The single most valuable thing that can happen on a contact.
• Prior payment behavior on this account. Someone who has made two payments during delinquency will likely make a third. This is the strongest genuine predictor and it's already in your data.
• Inbound contact. A person who called you has demonstrated something a person who answered your call has not. Inbound contacts should be routed and staffed as the highest-value interactions in the operation, and frequently aren't.
• An authorized future payment, which required an action and an account.
• Enrollment in a structured plan with a first payment made — the strongest completion predictor in our plan design guide.
• Contactability itself, which is the binding constraint our contact data analysis describes.
• Cash flow evidence, where a consented view is available — the only input that observes capacity directly rather than inferring it.
• Account and delinquency age, which dominates most behavioral signals.

Notice the pattern: almost everything above a promise on this list involves an action rather than a statement. That's the underlying principle. Actions require capacity; statements don't.

Converting a promise into an instrument

The practical upgrade. Rather than abandoning the promise, change what it is.

1. Take payment now if any amount is possible. Even a partial payment is worth more than a full promise, because it converts intention into cash and establishes a pattern.
2. Anchor the date to an income event. Ask when they next get paid and set the date there.
3. Size the amount to what they say they can manage, not to what clears the balance. An affordable amount that arrives beats an aspirational one that doesn't.
4. Capture authorization with clear disclosure of what will be charged and when, so the payment executes rather than requiring another decision.
5. Send confirmation immediately summarizing amount, date, and method — which serves as a record and as a reminder.
6. Remind before the date, since a meaningful share of breaks are forgetting rather than inability.
7. Provide a self-service path so the person can pay when they have the money rather than when you call — the friction point our collections framework identifies as consistently underestimated.

Steps four and seven do most of the work. A promise requires the person to act again later. An authorization does not. That difference is the whole gap between a promise and a payment, and closing it is largely a systems question rather than a conversation question.

Reading a broken promise

A broken promise is information, and most operations discard it by escalating uniformly.

What the pattern distinguishes:

• Broken, then answers the phone. Almost always inability rather than avoidance. The person is still engaging, which is a good sign. Response: a smaller structured arrangement.
• Broken, then unreachable. Could be avoidance, could be a contact data failure, could be circumstances deteriorating. Response: verify contact data before assuming intent.
• Partial payment made. Strong signal — the person paid what they had. Response: reduce the amount, keep the arrangement.
• Broken by a few days, then paid. Timing rather than capacity. Response: move the date, not the amount.
• Repeated promises, none kept. The promises are ending calls rather than reflecting plans. Response: stop taking promises from this account and move to a structured arrangement or a settlement conversation, per our settlement guide.

The general rule: the amount is wrong more often than the intent is. An operation that responds to a broken promise by re-promising the same amount is repeating an experiment that already failed.

A better metric set

What to report instead, and why each earns its place:

• Dollars collected per contact — the objective, per unit of the scarce input.
• Payments taken during contact, as a share of right-party contacts. The metric that most directly rewards the behavior you want.
• Promise-to-cash conversion, segmented by promise type. The only useful version of the promise metric.
• Plan enrollment and first-payment rate, since enrollment without a first payment is a promise wearing a plan's clothing.
• Self-service payment share, which measures whether you've removed friction.
• Right-party contact rate, the binding constraint.
• Recovery by segment at matched account age, which is the only fair comparison across teams.
• Complaint rate, which is the counterweight that keeps pressure-based tactics visible.

And the deletion that matters most: remove promise volume from any dashboard where it could influence behavior. Keeping it as a diagnostic is fine. Reporting it as performance is what produces the failure mode above.

What this implies about the conversation

If promises are weak because they don't require capacity, the conversation should be redesigned around establishing capacity instead of obtaining agreement.

What changes:

• Ask what they can pay, not whether they'll pay. The first question surfaces capacity; the second invites a promise.
• Ask when they next get paid. Anchors the date and reveals income structure — including the volatility our income analysis describes.
• Accept "no" as a real answer. A person who says they cannot pay $300 this month has given you information. Routing them to an affordable plan recovers more than pressing for $300 that won't arrive.
• Establish whether the difficulty is temporary or structural, which determines the right instrument — the distinction our forbearance analysis identifies as decisive.
• Close with an action, not an agreement. Payment now, authorization, or plan enrollment.

The reframe worth ending on: the goal of a collections call is not to obtain a commitment. It's to find out what this person can actually do and to make doing it as easy as possible. Those produce different conversations, and the second produces more money.

Frequently asked questions

---

This guide is educational and does not constitute legal or compliance advice. Payment authorization requirements, disclosure obligations, and communication standards are governed by federal and state law; consult qualified counsel about your practices.