Why your collections numbers lie.

Board packs from lender collections functions read the same everywhere. Roll rate stable. Cure rate stable. Right-party contact rate above 60%. Vulnerable customer flag rate roughly 5%. Every number is inside its historical range and the board approves the deck.

None of these numbers mean what the board thinks they mean. Under Consumer Duty they are actively dangerous. This note is why.

The three structural biases

First, roll rate is measured on the accounts you can see, not the accounts you have already lost. Second, right-party contact is measured on calls made, not calls made at times that actually helped the customer. Third, vulnerable customer flag rate is measured against the agent-detected pool, not the true incidence in your book (which is invariably higher by a factor of 2-3x).

Each of these biases makes the number look better than reality. Under a Consumer Duty conversation with the FCA, that gap is what gets a firm into remediation.

The numbers you should actually be looking at

1. Pre-arrears roll rate (rising-risk to 1+, not 1+ to 2+).
2. Outcome-quality index per treatment path (understanding, action, harm).
3. Vulnerable-customer detection recall against a labelled sample (not agent-flag rate).
4. Time-to-first-meaningful-intervention (not time-to-first-call).
5. Post-treatment persistence rate (did the arrangement hold, not was it agreed).

What this looks like in practice

A lender that shifted from the classic KPI set to the five above discovered that their true roll-to-2+ rate was 40% higher than the reported number, that half of their 'right-party contacts' were happening at times the customer had explicitly said were unhelpful, and that their vulnerable-customer detection recall against a labelled sample was 22% — not the 82% the agent-flag rate suggested. The remediation programme that followed was significant, painful and ultimately produced a materially stronger collections function and a materially better regulatory relationship.