Early-stage collections is easy to measure badly. A conversation that ends with a friendly agreement can look like success in the interaction log and produce nothing in the payment stream. The process has to be defined by the result it is trying to reach, not by the conversation it had.
The chain that matters#
- Right-party contact (RPC): verified contact with the customer or an authorised person.
- Promise to Pay (PTP): a commitment captured under an agreed definition.
- Kept PTP: the payment actually received in the agreed outcome window.
- Downstream result: the change in the customer’s payment behaviour.
Each step has its own denominator; a decision rate among reached customers is not the same as a payment rate.
Most reporting collapses these steps. That is where the illusions live: a high decision rate among reached customers, reported next to a low payment result, with no definition connecting the two.
What the evidence supports#
Where Vee has a cited reference point, it is specific: in a soft-collections reference process, on-time payment frequency was 7–13% higher than the cited human-agent comparator. That sentence carries its scope on purpose — process, comparator and reference. It is not a general claim about collections outcomes.
Architecture and governance#
Because the outcome is a payment, not a sentiment, the process needs a deterministic layer around what is promised: amounts, dates, eligibility and what the system may agree to. Generative components can run the conversation; they should not be the thing that invents a settlement term.
Before publishing a collections number
State the process, the segment, the period, the denominator and the comparator. Distinguish a declaration (PTP) from the downstream result (kept payment).
- Which population was contacted, and over what window?
- Was success measured among reached customers or among all cases?
- Is the comparator a human-agent baseline, and is it comparable?
Practical implications#
If a programme cannot answer those questions, it cannot yet claim an outcome. The right next step is usually a diagnostic: define the steps, pick the denominators, and only then compare configurations.
Follow The First Second
Occasional issues on the mechanics and economics of AI-powered customer contact. No noise, no vendor news.
Follow via RSS