Zendesk says its new AI agent can solve 80% of support issues

Zendesk announced Wednesday astatine its AI acme a drawstring of LLM-driven products meant to reshape the company’s reliance connected quality technicians.

The halfway of the caller features is an autonomous enactment cause that Zendesk believes volition lick 80% of enactment issues without quality intervention. That strategy volition beryllium supplemented by a co-pilot agent, which volition assistance quality technicians with the remaining 20% of issues, arsenic good arsenic an admin-layer agent, a voice-based agent, and an analytics agent.

According to Shashi Upadhyay, Zendesk’s President of Product, Engineering and AI, the caller agents are portion of a broader alteration successful the enactment industry, arsenic AI replaces overmuch of the enactment that was antecedently done by humans.

“The world’s going to displacement from bundle that’s built for quality users, to a strategy wherever AI really does astir of the work,” Upadhyay told TechCrunch.

Independent benchmarks suggest that modern AI models are susceptible of taking connected the work. TAU-bench, which was designed to measurement a model’s tool-calling ability, includes a script successful which models person to process a returned merchandise — a adjacent analogue to galore enactment tasks. The existent leader, Claude Sonnet 4.5, solves 85% of issues connected the test.

After a chaotic capitalist combat successful 2022, Zendesk has made a drawstring of AI acquisitions that laid the groundwork for the existent shift. The analytics cause launching contiguous is built straight connected the company’s Hyperarc acquisition, which was completed successful July. Earlier AI acquisitions see the QA and agentic work strategy Klaus (acquired successful February 2024) and the automation level Ultimate (acquired the pursuing March).

Zendesk has previewed the caller strategy with existing customers, and Upadhyay says the results person been promising.

“For customers that person been utilizing it, user restitution has been up by 5 to 10 points,” helium told TechCrunch.

Large connection models person often been deployed for lawsuit support, though seldom astatine the Zendesk’s scale. Companies from Airbnb to Regal Theaters person already experimented with in-house chatbot support, often contracting straight with instauration exemplary labs. But those systems typically woody with accusation retrieval alternatively than much analyzable troubleshooting oregon taking self-directed action.

If the caller propulsion for AI-based enactment is successful, the economical implications would beryllium significant. Zendesk’s Resolution Platform already supports astir 20,000 customers, resolving 4.6 cardinal tickets each year. Beyond Zendesk, the U.S. employs 2.4 cardinal lawsuit work representatives — with acold larger workforces successful different countries.