Will AI voice bots replace human collectors?

1. Where persuasion matters

Whether a bot or a person should take a call comes down to how much persuasion the call needs.

Where the consumer has already decided, conversion is high whoever answers. Where they have not, the call is won or lost on persuasion. That is our view, and the rest of this page is the argument for it.

2. What makes a person persuasive

The persuasion research describes the methods. Robert Cialdini’s Influence (1984) sets out why people say yes, including commitment (we keep to what we have said to someone) and liking (we agree more readily with people we feel understand us). Neil Rackham’s SPIN Selling (1988) shows that the questions that work let the other person describe their own situation and what solving it is worth. What the persuasion models say about collection calls.

Each of those methods works only if the consumer believes the person using it. A question about their situation lands when they believe the collector wants to hear the answer. A promise binds when it is made to someone who listened. Underneath the methods is one ingredient: empathy.

3. What empathy is made of

Empathy is walking in someone else’s shoes, and letting them know you have. The research breaks it into parts. Theresa Wiseman’s concept analysis for nursing (1996) is commonly summarised as four:

1. Seeing the world as they see it. The hours that were cut, the bill that came first.
2. Staying out of judgement. The debt is a problem to solve, not a verdict on the person.
3. Recognising the emotion. Worry, embarrassment or anger, heard before the numbers.
4. Communicating it. Saying back what you heard, so they know it registered.

Carl Rogers (1957), writing about therapy, put the weight on that last step: understanding counts only when the other person perceives it. He also drew the limit: the listener enters the other person’s world as though it were their own, while keeping their own position. On a collection call that means a collector can understand a hardship completely and still ask for a payment. Why empathy matters on a collection call.

4. Why a bot cannot walk in the consumer’s shoes

A bot can say the words of empathy. In our view it cannot supply the parts that make them work:

• It has no shoes to walk in. It has never had a bill it could not pay, so its understanding is a script, and consumers can hear the difference.
• The last step fails. Empathy counts only when the consumer believes it. When they know they are talking to a bot, the understanding is not received, however well it is worded.
• The commitment binds less. A promise to pay is a commitment made to someone. Made to a machine, it carries less weight.

The evidence points the same way. In "How Good is AI at Twisting Arms?" (NBER working paper w33669, 2025), Choi, Huang, Yang and Zhang studied real collection calls at an online consumer finance company in China. When the caller was an AI voice bot, 21 percentage points fewer borrowers promised to repay on the first call they answered, and the AI calls collected less. The AI and the human callers counted a promise on slightly different rules (the AI a same-day promise, humans one up to the end of the next day), which the authors flag and test for, and repayment tells the same story. The authors suggest a promise made to a bot feels less binding. The cases are from 2021 and 2022, before today’s large language model voice agents, and it is one study. Bots will keep improving; we expect empathy to be the last gap they close.

5. Why this is where BestPair works

Empathy and persuasion are not spread evenly across a floor. Collectors differ in how well they persuade, and in our view the difference is largest on the accounts that need the most persuading. So the more persuasion an account needs, the more it matters which collector takes it. Our published case study states this reading as a hypothesis rather than a finding: in a live test at a North American collection agency on one telecom book, the gain from pairing was largest on the biggest instalments, and on the smallest the two halves were level, within chance. The case study.

That is the job predictive pairing does. It puts each account in front of the collector your own outcomes show does best with that kind of account. It is also why a floor that already converts most of its right-party contacts has little to gain: there, little persuasion is left to do. Why collectors convert at different rates.

People make every call. BestPair changes only which collector 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.

6. If you put a voice bot on part of the floor

Start where the persuasion needed is lowest, and measure it like any change in who takes a call: split accounts at random before anyone dials, run both ways in the same weeks, score both on dollars collected per account contacted, and keep every account in its group. Whether a bot fits your contact rules is for your compliance team and counsel; BestPair does not provide legal or compliance advice. How to prove a change in routing worked.