Financial Services
Collections propensity from consumer history, not static profiles
On one receivables provider's book (7,926 accounts with 12 months of exposure), a model built only from consumer history available at listing time reached AUC 0.781 against the incumbent score's 0.688. Working the top half of its queue reached 83.7% of all payers, against 71.0% on the incumbent queue.
Business Impact
What changes when raw data stops moving
Rank accounts on behaviour available at listing time
Reach more payers in the top half of the queue
Stop spending contact effort on accounts that will not pay
Keep consumer-level data inside the provider's environment
0.781
AUC against 0.688 incumbent
One provider's book, 7,926 accounts, 12-month exposure
83.7%
Payers reached in the top half of the queue
Against 71.0% on the incumbent queue
3.9%
Payers left in the bottom three deciles
Against 15.7% on the incumbent ranking
2.64x
Top-decile lift over an 18.3% base rate
Statistically level with the incumbent's 2.65x
The challenge
Static profiles strand active payers
Collections scores built on account attributes are strong on the obviously good accounts, flat through the middle of the book, and keep placing payers in the bottom deciles. The signal that separates accounts is the consumer's trajectory: how recently they paid anyone, how contact converts to payment, how their other accounts resolved. That history is sensitive and usually cannot leave the provider.
Approach
How Datasent enables this use case
Features
Strictly as-of listing
Every input is computed from events dated before the account's own listing date: prior journeys, prior payments, response ratios by channel, sibling accounts placed or paid. Nothing that happened after listing leaks in.
Validate
Grouped and exposure-controlled
Accounts of the same consumer never appear in both training and test folds. Only accounts with 12 months since listing are scored, and the result is confirmed on a fixed 12-month outcome window (0.745 against 0.716).
Deploy
Combine or replace
The incumbent and consumer-history scores disagree exactly where each holds information the other lacks, which keeps a combined score on the table. That decision is made with the provider, against its clients' expectations.
Join our newsletter
Get updates on research, product news, and privacy-first data insights.
