Does agent matching raise payment size or how often people pay?

Two numbers inside one

Dollars collected per account contacted is the product of two simpler numbers: the share of contacted accounts that pay, and the average amount collected from each account that pays. A change on the floor can move either one. If more consumers pay but each pays less, the dollars can stay flat. If the same share pays but larger amounts, the dollars rise with no change in the payment rate.

Collection software can keep the two apart. Oracle’s banking collections documentation lists a promise kept rate, which counts promises, next to a collection liquidation rate, which sets the amount collected against the amount assigned. They can also move apart over time. In 2011 Mike Ginsberg of Kaulkin Ginsberg wrote in insideARM that liquidation on larger credit card balances was falling faster than on smaller ones, and linked it to consumers’ reduced ability to make large up-front payments or payoffs.

That is why the score for any routing change should be dollars, with both parts read underneath it. What counts as success in collections.

What the published case study shows

A North American collection agency split a telecom book of 170,000 accounts 50/50 at random before the first call. One half went to the agents BestPair matched; the other ran on the usual process, with the same dialer, the same agents and the same weeks. Cash was counted for 30 days after each account’s first real conversation, with the largest 1% of payments trimmed.

• The May 2026 result. $98.76 collected per account reached in the matched half, against $79.17 on the usual process: 24.7% more. The case study says the difference came from more cash from the accounts that paid, not from more people paying.
• The longer read. A longer read of the same test, through 7 July 2026, on 2,661 reached accounts whose 30-day payment window had closed, not the May result. It split those accounts by instalment size. Across the whole book, the share of reached accounts that paid was about the same in both halves: 1.8 percentage points higher in the matched half, a gap chance alone could produce. On the smallest instalments the halves were level. On instalments of $1,000 or more, about one in three reached accounts paid in the matched half against about one in five on the usual process, a clear difference.
• The case study’s reading. The pattern is reported: the larger the account, the bigger the gap. The explanation, that the more persuasion an account needs the more it matters who is on the call, is labelled a hypothesis rather than a finding.

The full case study shows how each figure was counted.

What that does and does not tell you

• One book at one agency. It shows what happened on that telecom book in that period. It is not a forecast for another book or another floor.
• The match reached about half the calls. Adherence was about 50%: when the matched agent was busy, the dialer sent the call to whoever was free. The test did not measure what happens when the matched agent takes most of the calls.
• Two different windows. The longer read’s figures cover the 2,661 reached accounts whose 30-day payment window had closed by 7 July 2026, so they are not the May result restated.
• A book’s mix matters. On that book, the gap sat with the larger instalments. A book of mostly small instalments could look different, and only your own data can say.

How to measure both on your own floor

1. Split at random and run both groups in the same weeks. How to prove a routing change worked.
2. Count three things per group. Accounts contacted, accounts that paid within the window, and dollars collected. From them, work out the share of contacted accounts that paid, the dollars per paying account and the dollars per account contacted.
3. Fix the window before the start. For example, payments within a set number of days of the first conversation, the same for both groups.
4. Take reversals back out, and decide in advance how to handle the very largest payments, so one settlement cannot decide the result.
5. Choose account-size bands before the test, by balance or instalment, and read each band. Bands picked after the results are in will always find a winner somewhere.
6. Treat arrangements with care. An instalment plan pays over months, so a short window sees only its first payments. RPC, promise-to-pay and kept rates covers promises and kept rate.
7. Expect payment-rate differences to need many accounts. A gap of a few points in how often people pay can sit inside chance on a small campaign. Read the range, not only the point.

Sources

• Oracle Banking Collections, key performance indicators Product documentation defining promise kept rate and collection liquidation rate.
• insideARM, The ARM Industry is in for a Long Recovery, Part II Mike Ginsberg, Kaulkin Ginsberg, 30 August 2011, on liquidation declines by balance size.