A Voice AI Chief Says Borrowers Open Up More to a Machine About Debt
The ElevenLabs chief says people feel ashamed telling the real situation to a person, but are more open with a machine they cut off.
All-In with Chamath, Jason, Sacks & Friedberg · The Trillion-Dollar Industries AI Is Disrupting: Voice, Law & the End of the Billable HourAsked how people should talk to an AI system, Mati, the chief executive of the voice AI company ElevenLabs, turned to an awkward job his technology now does. His company provides voice agents to financial services firms, he said.
People, in his words, are “much more open to share what actually happened” when speaking with AI than with a person.
The example he gave was payment reminders and efforts to collect from people who have stopped answering. Those conversations are sensitive, he said, because describing personal circumstances to another person can feel embarrassing.
He described the emotional block around saying everything in front of another human as “very different” once the listener is a machine.
He also said callers treat the machine with less ceremony. People are usually “more snappy” with an AI voice agent, he said, offering quick responses, and they do not mind cutting it off.
Questions people ask
Why would someone be more honest with a machine than a person about money?
Psychologists point to the lack of social evaluation. People fear looking irresponsible or poor in front of another person, so they dodge, minimize, or stop answering. An AI does not sigh, judge, or remember you later, so admitting trouble can feel safer.
Do companies already use AI voices to collect debts?
Yes. Banks, lenders, and buy-now-pay-later firms use automated calls and text agents for payment reminders, late bills, and early-stage collections. Humans usually still handle complex disputes, legal threats, and final settlements.
Does being rude or threatening to a chatbot actually get better answers?
No, not reliably. Clear, specific prompts with context, examples, and constraints work best. Threats may change tone in some models because of how they were trained, but they do not make the information more accurate.