Allocator studio

How many agents do you have?

Conventional collections sorts by probability of default and works down the list. The allocator maximises expected recovered value per agent-minute. The gap between them is not a constant. It depends entirely on how scarce your capacity is. Move the slider.

15%
of the agent-minutes it would take to make a plan offer to every delinquent account. — minutes across — accounts.
2%15%30%45%60%
—recovered value, allocator
—recovered value, risk-ranking
—the allocator's edge at this capacity
—contacts made, allocator vs risk-ranking
allocator risk-ranking your capacity

Both strategies converge as capacity approaches full coverage: given enough agents to call everybody, ordering stops mattering. The edge lives where real operations live, at the left of this chart. Reporting only the flattering end of it would be the easy thing to do.

01

Who each strategy actually calls

Fifty-two accounts from the live queue, at the capacity you set. Highlighted rows are the ones only this strategy funds. The disagreements are the whole argument.

Allocator

ranked by expected value per agent-minute · — of 52 funded

    Risk-ranking

    ranked by probability of worsening · — of 52 funded

      02

      The trade, in two accounts

      These are real rows from the queue, picked because they are the clearest case of the two strategies disagreeing. Neither is a hypothetical.

      03

      Silence is a scored action

      The accounts neither strategy funds are not an oversight. Most delinquents self-cure. Contacting them spends the most expensive resource in the business on someone who was already going to pay, and irritates them while doing it. So “do not contact” has a value in the objective, and wins on its merits. Invariant I9 makes this the only component allowed to produce a queue, so no path exists that quietly ranks by probability instead.

      —of the 52 the allocator never funds, at any capacity in the sweep
      —risk-ranking never funds, even at 60% capacity
      — of delinquent accounts cure without contact

      One caveat, stated plainly. The uplift shape the objective uses is modelled from published practice rather than measured here, because the servicing data carries no treatment history to measure it from. The logic beating risk-ranking is what this demonstrates. The magnitude stays badged simulated until a control arm produces it.