Collections Measurement
Check dunning decisions, promises, holdout lift, stage recovery forecasts, experiments, aging, and operator performance.
Collections measurement connects follow-up activity to receivable outcomes. Use it to understand what the system decided. It shows which invoices the system treated or held out. It also shows whether payment behavior changed after a sequence.
Dunning Summary
Open Collections → Dunning to check:
- reminders approved in the last seven days
- actions skipped in the last seven days
- active promises and promises due soon
- broken promises
- common reasons for a skipped action
- the invoice and client attached to each available decision
Filter the activity feed between all, sent, and skipped decisions when you need to check a rollout or investigate unexpected behavior.
Decision Reasons Matter
A send count alone does not show whether the system behaved well. Check skip and defer reasons. They confirm that payment, disputes, promises, compliance, and policy stop the wrong actions.
The decision feed records the mode of operation, action, reason, channel, template when available, days past due, collection stage, and time.
Sequence Lift
Sequence lift compares the paid rate for treated invoices with a holdout cohort over the selected window.
The panel shows:
- invoices enrolled in each cohort
- the paid rate for treated invoices
- the paid rate for holdout invoices
- the difference in percentage points when the sample is large enough
When either cohort is below the configured minimum, the panel reports an insufficient sample. It does not present an unstable lift number as proof.
This view measures paid-rate difference. For now, it does not claim incremental dollars recovered.
Stage Recovery Forecast
The stage forecast uses recent collection-stage history to show:
- open invoices in each stage
- open amount in each stage
- observed recovery probability when the sample is large enough
- probability-weighted expected amount
Use the forecast to compare stages and prioritize review. It is an AR-operational estimate, not a company-wide cash forecast or bank-balance projection.
Experiment Results
When an active or completed sequence experiment exists, the Dunning view can show:
- control and variant enrollment
- paid conversion for each arm
- the percentage-point difference
- the current significance result
- whether the experiment reached the target sample
The page displays results for configured experiments. It does not create or start experiments.
Aging and Performance
Use Collections → Aging for outstanding exposure by age. Use Collections → Performance for collection and operator metrics over a chosen period.
Do not rank operators by message volume alone. Recovery, resolution quality, promises kept, dispute work, and safe suppression are stronger signals than raw activity.
Interpret Results Carefully
- A paid-rate difference is not the same as incremental cash.
- Small samples must keep the insufficient label.
- A holdout is useful only when the system suppresses the treatment.
- Changes in risk mix can distort comparisons between cohorts.
- Forecasts explain likely collection outcomes, not the full financial position of the business.