Case study · Completed
Account scoring that routes each account
A national debt buyer and legal-collections servicer.
The problem
The client must choose, for each account, between a law firm, its own in-house team and an outside agency. The agency file was picked by time left on the statute of limitations, not by the chance of payment. A check later found that 79% of the accounts at the agency scored below the median.
What we built
- A scorecard that predicts the chance of payment over 36 months, with expected dollars and a plain-English reason for each account.
- A weekly score run across every active account. The run stops itself if the score mix drifts too far.
- Placement rules on top of the score. Legal-ready accounts go to law firms, strong accounts stay in-house, and the rest go to the agency, best first.
- A person approves every batch before it goes out. 5% of accounts are held back as a control group.
- A pool collection forecast built from older pools with the same profile, tested against the existing forecast on three past dates.
Results
- 97,000 active accounts scored each week.
- Tested on accounts it had never seen, the score ranked accounts well, and its predicted payment rates matched what actually happened.
- In a backtest, the top 5,000 accounts held 84.6% of the 36-month collection dollars.
- In a backtest on past pools, the new pool forecast had about 60% less error than the existing one at 12 months.
- The pool forecast ties to the general ledger within 1% in 99.6% of pool-months.
Timeline
Scoring live since August 2026.
Related work
Law-firm files posted to the servicing system automatically
$1.1M in law-firm remittances posted with no hand keying in the first 3 weeks, plus status reports, cost invoices and skip-trace results.
A reporting database on the current system
487,000 accounts and 6 daily dashboards, with no platform change.
Start with a free consultation
A 30-minute call, then a fixed estimate.