How does agent matching fit with agent assist, speech analytics and other contact centre AI?

Four tools, four jobs

Contact centre AI is sold as one category, but on a collection floor the tools answer different questions, at different points in the life of a call.

The first three work on the conversation. Agent matching works on who is in it. That is why, in our view, they rarely compete for the same budget line, even when they come from the same vendor pitch.

What speech analytics and compliance monitoring do

Call Centre Helper’s guide to speech analytics in debt collection (Jonty Pearce, 2010, updated 2019) describes software that analyses both sides of a recorded call. It says recordings suggest many collectors do not ask for payment, and that the analysis can find collectors who need more training, spot the ones who put the agency at risk, and show which practices to spread across the floor. The promise-to-pay ratio and script adherence are among the areas it names.

The compliance case is made most strongly by vendors. Jason Napierski of CallMiner, a speech analytics vendor, argued in insideARM in 2013 that software can screen every call for required disclosures, right-party contact language and risky words, and grade each call by risk, where a person sampling recordings sees only a few. Treat the results in that piece as one vendor’s account.

What agent assist does

Agent assist works during the call: prompts, answers and reminders on the collector’s screen. The best-known study of an assistant like this that we found is outside collections. In "Generative AI at Work" (NBER working paper, 2023), Brynjolfsson, Li and Raymond studied a generative AI conversational assistant rolled out in stages to 5,179 customer support agents. Issues resolved per hour rose 14% on average and 34% for novice and low-skilled workers, with minimal effect on the experienced and highly skilled. The authors suggest the tool spread the practices of the stronger workers.

That is customer support, not collections, and one study. On a collection floor, a prompt is only a suggestion: required disclosures, offers and settlement authority stay under the agency’s own scripts and compliance rules.

Where agent matching sits

Agent matching acts before the call. It changes which collector works each account, and leaves the conversation alone. BestPair’s Order Form puts it in one sentence: BestPair changes only which agent works an account. It does not contact consumers, and it does not change whether an account is worked, what a consumer is offered, or any collection terms. BestPair does not listen to calls.

In our view the tools meet in one place. Speech analytics shows how collectors differ in what they say. Outcome results show how they differ in what they collect, and that a collector can be strong on one kind of account and average on another. Why collectors convert at different rates covers how to see that in your own history, and using collector results for coaching covers what a manager can do with it.

Test one change at a time

If assist goes live in the same month as new routing and collections rise, nobody can say which did it. Two ways to keep the answer clean:

1. Stagger them. Settle one change, measure it, then start the next.
2. Hold everything else equal inside the test. Split accounts at random into two groups and give both the same scripts, assist settings, offers and dialer rules, so the only difference is the change being tested. Compare the groups in the same weeks.

In our view this matters more with assist than with most changes. If the NBER pattern holds on your floor, assist lifts newer collectors most, which narrows the gaps between collectors while a routing test is running. How to prove a routing change worked sets out the method, and how to check a vendor’s lift claim covers what to ask any vendor, whichever tool it sells.

Sources

• Call Centre Helper, Speech Analytics in Debt Collection Jonty Pearce, 2010, updated 2019, on what speech analytics finds on collection calls.
• insideARM, Speech Analytics Can Improve Collection Results and Reduce Compliance Risk Jason Napierski, 2013, an opinion piece by a product marketing manager at CallMiner, a speech analytics vendor.
• NBER, Generative AI at Work Brynjolfsson, Li and Raymond, working paper, 2023, on an AI assistant for customer support agents.