A promise to pay is something a customer says. A payment is something that appears in the bank’s systems, days later, or does not. In our view, treating the first as if it were the second is the most common way collections results are overstated, by internal teams and suppliers alike. We report the difference explicitly, and we think every buyer of customer contact should insist on it.

The distinction applies far beyond collections. Stated interest is not a sale. A confirmed appointment is not attendance. A completed survey is a result in itself, but a promise to renew is not a renewal. In each case there is a declaration made in the conversation and a fact confirmed later in the client’s systems. Good measurement keeps them apart.

Two kinds of success#

We use two precise terms, and we agree them with every client before launch.

Declarative and actual success
Declarative successActual success
DefinitionA result heard and recorded during the conversationA business result confirmed in the client’s system after a defined period
ExamplesPromise to pay, stated interest, acceptance of a next step, confirmation of a visitPayment received, purchase, attendance, activation, retained customer
Where it is recordedContact supplier’s conversation dataClient’s core systems
When it is knownAt the end of the callAfter the agreed measurement window
What it measuresThe quality of the conversationThe effect on the business

Definitions from Vee’s measurement framework.

Both are useful. Declarative success tells you whether the conversation is doing its job and is available immediately, which makes it the right signal for daily optimisation. Actual success tells you whether the business benefited. The mistake is not measuring promises. The mistake is reporting them as if they were money.

Why promise rates mislead#

Promise-to-pay rates are easy to inflate, often without anyone intending to. A conversation that pushes hard for a date will collect more dates. Some will be made by customers who simply want the call to end. A script that accepts ‘sometime next week’ as a promise will report more promises than one that insists on a day. A supplier paid per promise has every reason to drift in both directions.

The horizon matters as well. A promise to pay within a few days is a different thing from a promise to pay ‘by the end of next month’. Banks define the maximum horizon a promise may have, and a good conversation works inside it. A report that counts every date as a promise, however distant, flatters the conversation and tells the risk team very little.

The opposite distortion is just as real and gets less attention. In our early-collections projects, some customers pay without making any promise. They say they know and have no time, or they never answer but see that their bank called. A report built on promises counts these as failures. So does a report built on minutes. In the bank’s ledger they are cured accounts.

There is a useful bridge between the two: the share of promises that are kept. It requires payment data from the client, but it is the most direct test of whether a conversation produced real decisions or polite agreement. A falling kept-promise rate is often the first sign that a dialogue change has made promises easier to give and less likely to be honoured.

Measuring against a comparator#

Actual success on its own still does not tell you what the contact contributed. Many customers in early delinquency would pay anyway. The honest figure is the incremental effect: how much better the accounts handled by one channel performed than comparable accounts handled another way.

That requires a comparator on the same buckets and portfolios, over the same period, measured at the level of contracts rather than calls. In practice that may be the bank’s existing call centre handling a parallel share of the allocation, an earlier period with the same mix, or a control group. Results should be measured the way you would measure a contact centre: in your systems, on your definitions, against a baseline.

This is one reason we always recommend starting on part of the volume rather than the whole base. A controlled start creates a natural comparator: the bank can run its existing channel and ours side by side on comparable accounts, watch the payment data arrive in its own systems, and decide on scale with evidence. Starting on everything at once removes the baseline, and with it the ability to prove anything.

+7–13%On-time payment frequencyFrequency of on-time payments among accounts handled by Vee compared with accounts handled by a human-agent comparator.Early-stage collections, historical; versus a human-agent comparator.

That is the form of our headline collections figure, and deliberately so. It is not a promise rate, and it is not a gross recovery rate that would credit us with payments that would have happened without any call. It is a difference in payment behaviour against a human comparator, measured on the bank’s data. We would rather publish a modest, defensible difference than a large number that falls apart in the first audit.

What the data has to look like#

Comparator reporting depends on data flowing back from the client, and that is usually the hardest part. The outcome lives in the client’s systems, not ours. We work with whatever the client can provide, from the most detailed to the most aggregated.

  1. Contract-level outcome files with the contract number, so each payment can be matched to the cases that preceded it.
  2. Aggregated effects per portfolio and bucket per month, when contract-level data cannot leave the client.
  3. Automatic return after each closed allocation, the target state, where outcomes come back as routinely as allocations go out.

The units have to be clear too. An account is the customer relationship, such as a loan. A case is each assignment of that account for handling; one account can generate several cases in a month. An attempt is a single dial. Payment is a property of the account, contact is a property of the case, and mixing the two is how two honest teams end up with incompatible numbers.

The same gap in sales, healthcare and service#

Collections makes the gap easy to see because money either arrives or it does not. The same structure runs through every process we operate, and it pays to name the declarative and the actual result for each one before launch.

In lead qualification, the declaration is interest and the actual result is a signed contract. Interest is worth measuring, because it decides which leads reach a salesperson, but the commercial question is what each customer acquired actually cost. In one lead pre-selection campaign, interested leads rose by 26% and the cost of acquiring a customer fell by 73% (one campaign, historical). The second number is the one a sales director takes to the board, and it could only be calculated with the client’s own sales data.

In healthcare scheduling, the declaration is a confirmed visit and the actual result is a patient in the chair, or a cancelled slot refilled in time. That changes what the conversation is for. Learning early that a patient will not come is often worth more than a confirmation that turns out to be wrong.

We call it appointment confirmation. Internally we call it cancellation, because that is the business goal: learning early that a visit will not happen. – Dawid Wójcicki, Founder & CEO of Vee

Some processes have no gap at all. A fully completed customer-satisfaction survey is the result; there is nothing further to confirm. Knowing which kind of process you are running is the first step to measuring it honestly.

When results drive the invoice#

The distinction becomes contractual the moment part of the supplier’s fee depends on outcomes. Outcome-linked pricing is an option we offer where the result can be measured reliably in the client’s systems, not a blanket model. It works only when both sides have agreed, in writing, on the full design.

  • Success: declarative, actual, or both, defined exactly.
  • Source of truth: which system confirms the result.
  • Measurement window: how long after contact a payment still counts.
  • Attribution: how the result is linked to the contact, and against which comparator.
  • Exclusions and borderline cases: disputes, hardship routes, accounts withdrawn by the client.
  • Data return: how and how often outcome data comes back.
  • Disputes: how disagreements about attribution are resolved.

We also apply a simple test before agreeing to outcome-linked terms: the supplier must have material influence on the result, the result must be measurable, the feedback data must be available, and there must be enough history to price it credibly. Where those conditions do not hold, a different fee structure, set out in the agreement, is the more honest arrangement.

Where we draw our own line#

We measure contact on four levels: reach, continuation past the introduction, obtaining the key information such as a promise, and the real business effect such as a payment. The first three are ours to improve directly. The fourth we influence indirectly, through the quality and timing of everything that comes before it, and we say so. A customer’s ability to pay is not something a conversation controls.

That honesty is not modesty. It is what makes our numbers usable. A risk committee, an auditor or a procurement team can trace a comparator-based payment figure back to the client’s own records. They cannot do that with a promise rate quoted in isolation.

What to ask before you accept a results claim#

  • Is this a declarative result or an actual one, and where was it measured?
  • What share of promises were kept, according to our payment data?
  • What was the comparator, and was it on the same buckets, portfolios and period?
  • What is the unit: accounts, cases or attempts?
  • Are unanswered and very short contacts included in the denominator?
  • If part of the fee depends on outcomes, where is the written definition of success, window, attribution and disputes?

Suppliers who can answer those questions without hesitation are usually the ones whose results survive contact with your ledger. It is the standard we hold ourselves to, and the one we suggest you hold everyone else to.

Follow The First Second

Occasional issues on the mechanics and economics of AI-powered customer contact. No noise, no vendor news.

Follow via RSS