Will AI voice agents replace telesales agents?

Where persuasion is needed

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

Where the buyer has already decided, almost anyone, or anything, can take the order. Where they have not, the sale is won on the conversation. That is our view, and the studies below are the evidence we know of.

What the studies show

Buyers who know it is a bot buy less. Xueming Luo, Siliang Tong, Zheng Fang and Zhe Qu ran a field experiment with more than 6,200 customers randomly assigned to highly structured outbound sales calls from chatbots or human workers (Marketing Science, 2019). When customers did not know, the chatbots sold as well as proficient workers. When they were told before the conversation that the caller was a chatbot, purchase rates fell by more than 79.7%. Customers became curt and saw the disclosed bot as less knowledgeable and less empathetic. Telling them later in the call, and customers who had used AI before, softened the drop.

AI agents are behind on objections and persuasion. Krittanon Kaewtawee and colleagues built a voice agent from telesales call recordings, following a playbook drawn from the best human agents, and had experienced evaluators rate it blind against human agents (arXiv, 2025). It came close to the humans on the routine parts of a call, such as the introduction and explaining the product, and fell behind on objection handling and sales drive. After refinement it improved, but in the hardest scenario it still trailed the human agent. The test covered three scripted scenarios for one company’s telesales, and it is a preprint.

The same pattern on collection calls. In a study of real collection calls at an online consumer finance company in China, Choi, Huang, Yang and Zhang found that on the first answered call, 21 percentage points fewer borrowers promised to repay when the caller was an AI rather than a person, and promises made to the AI were less often kept (NBER working paper w33669, 2025). What that study measured, and what it did not.

AI may work best on the routine part. In a telemarketing field experiment by Luo and colleagues, published in the Academy of Management Journal and summarised by Temple University, a chatbot handled the first screening and passed serious prospects to people for the selling. The AI-assisted teams nearly doubled credit-card conversions, and most of the gain went to the higher-skilled agents.

Why the gap sits in persuasion

Persuasion on a sales call rests on the buyer believing the person asking. A question about their situation lands when they think the agent wants to hear the answer. A commitment binds more when it is made to someone. Empathy counts only when the other person perceives it, and a buyer who knows the caller is a machine has little reason to. What SPIN, Cialdini and empathy say about sales calls.

The evidence fits that reading: bots do the scripted parts of a call well and fall behind where the buyer needs persuading. Voice agents are improving fast, and these studies will date. In our view, persuasion of an undecided buyer will be the last part they match.

Disclosure, consent and the rules

In the United States, the Federal Communications Commission’s declaratory ruling of February 2024 confirmed that the Telephone Consumer Protection Act’s restrictions on calls made with an artificial or prerecorded voice cover current AI technologies that generate human voices. Such calls need the called party’s prior express consent, absent an emergency or an exemption, and the ruling notes that the FCC’s rules require prior express written consent where the call carries telemarketing. The rules also require these messages to identify the business responsible for the call, and telemarketing messages to offer a way to opt out.

Whether and how a voice agent fits your calling is for your compliance team and counsel. BestPair does not provide legal or compliance advice.

If you put a voice agent on part of the floor

Start where the persuasion needed is lowest, and measure it like any change in who takes a call: split leads at random before anyone dials, run both ways in the same weeks, and compare sales per contact. How to prove a lead routing change worked.

The more persuasion a lead needs, the more it matters which person takes it, because agents differ most on the leads that are hardest to sell. That is the part of the floor BestPair works on. People make every call; BestPair decides which agent works each lead. Why some agents close more on the same leads.

Sources

• Luo, Tong, Fang and Qu, "Machines vs. Humans: The Impact of Artificial Intelligence Chatbot Disclosure on Customer Purchases": a field experiment on outbound sales calls by chatbots and people, Marketing Science.
• Kaewtawee, Modecrua, Pachtrachai and Kraisingkorn, "Cloning a Conversational Voice AI Agent from Call Recording Datasets for Telesales": a preprint comparing an AI telesales agent with human agents in blind ratings.
• Choi, Huang, Yang and Zhang, "How Good is AI at Twisting Arms?": an NBER working paper on AI voice bots in debt collection.
• When and How Artificial Intelligence Augments Employee Creativity (Temple University summary): a one-page summary of a telemarketing field experiment published in the Academy of Management Journal.
• FCC declaratory ruling on AI-generated voices: the Federal Communications Commission’s February 2024 ruling under the Telephone Consumer Protection Act.