Can AI Agents Resolve Customer Inquiries Easily?

Sierra co-founder Clay Bavor discussed the state of enterprise artificial intelligence adoption in an interview on CNBC International’s The Tech Download, on July 14, 2026. 

AI agents capable of fully resolving 50% to 80% of incoming customer inquiries without human involvement can deliver a significant return for enterprises, Bavor said. Sierra builds customer-facing AI agents for functions including customer service, sales, and mortgage origination.

Enterprises are increasingly deploying fully autonomous agents rather than tools that simply assist human staff, Bavor noted. The difference in return on investment between an agent that fully resolves a case and one that makes a human employee "10 or 15 or 20% more productive" is significant, he added, both financially and for the customer experience.

Guardrails, Value and Execution

Sierra layers supervisory agents on top of its customer-facing agents to monitor output in real time, Bavor told CNBC. These supervisory agents can flag responses that fall outside approved boundaries, such as dispensing financial or medical advice, and either route the case back to the primary agent or escalate it to a human employee.

Rather than charging a seat-based license fee, Sierra bills customers only when an agent fully resolves a conversation, retains a customer who intended to cancel, or completes an upsell, Bavor said. That model, he added, has pushed the conversion rate on the company's proofs of concept above 90%. "We only win when our customers win. Our success is aligned with our customers' success," he said.

Bavor also told CNBC that Sierra routes tasks across a mix of AI models rather than relying on a single provider, pairing frontier models for complex reasoning with smaller, fine-tuned models for narrower jobs such as detecting voice interruptions. More broadly, he said applied AI companies do not need to use the most expensive frontier models for every task, likening it to not needing "a 10 trillion parameter-model" to help determine "what sweater would go best with that pair of trousers."

Separately, Bavor said some individual engineers are spending more than $100,000 annually on tokens amid a broader push across companies to increase AI usage. He said he expects enterprises to begin budgeting and allocating token usage to employees in a manner similar to a travel or departmental budget, as scrutiny over AI spending grows.

Bavor’s statements point to a broader pattern in enterprise AI: as adoption matures, companies appear to be shifting focus from usage volume toward measurable business outcomes and cost discipline.

>