Zendesk is shifting from charging companies for each employee using its software toward billing for the work AI completes. As AI handles more customer service tasks, customers may have less reason to buy another employee a seat. CEO Tom Eggemeier’s projections suggest products outside Zendesk’s newer AI offerings will still generate roughly $1.6 billion in annual recurring revenue (ARR) this year, leaving a substantial business to manage through the transition.
Eggemeier is betting Zendesk’s existing customer relationships will help it make that shift faster than larger incumbents.
“We have more than 70,000 happy customers who want Zendesk to succeed. And that is why we’ll finish the year with almost $400 million in AI annual recurring revenue, specifically from our new AI solutions,” he tells me. “We have a competitive advantage because we’re moving more quickly. That puts us in a sweet spot where we can keep pace with some of the AI-native companies while also moving faster than some of the traditional SaaS incumbents.”
Eggemeier expects AI to account for about 20% of recurring revenue this year, implying company-wide ARR of roughly $2 billion, and has set a target of $1 billion in AI ARR by 2028. Moreover, Zendesk expects its platform to handle 1 trillion calls via application programming interfaces and Model Context Protocol servers this year, enabling its software to access other systems and perform tasks.
What counts as an AI resolution worth paying for?
Zendesk introduced outcome-based pricing for AI agents in August 2024. Eggemeier now wants to take that shift further, offering unlimited seats while charging customers for the work its software completes. “I think employee service and customer service are the easiest areas for automated resolutions,” he says. “Customers generally don’t want to pay for an interaction. They only want to pay for results. So even though that might make things a little more complex, we’re moving aggressively in that direction because that’s what our customers want. We also think that’s where the industry is going, and we want to lead that transition.”
That approach requires Zendesk and its customers to agree on what counts as a completed job.
“We sit down with our customers and discuss our definition of an automated resolution. We then have large language models (LLMs) analyze whether those interactions were resolved,” Eggemeier says. “Then, on a periodic basis, we review samples with our customers and say, ‘Here’s a sample we’ve pulled for manual review. Would you assess whether these interactions were resolved? And here’s what our LLM said about those same interactions.’”
He revealed that private ownership gives Zendesk room to experiment with pricing. Hellman & Friedman and Permira took the company private for $10.2 billion in 2022.
“It’s great being private during this kind of transformation because we’re focused primarily on getting customers to adopt the technology and driving automation for them. We’re not as concerned about quarterly financial results,” he says. “That has really allowed us to experiment. We think we’re going to be one of the first companies to push aggressively toward a model with no seat charges or unlimited seats, meaning customers won’t pay according to how many seats they use.”
Better AI Models Won’t Fix Outdated Workflows
Eggemeier argues that Zendesk’s advantage lies in the work required to make a model useful inside a customer’s business, including the software that connects it to daily operations.
“The last kilometer, or last mile, is the most important part. Frontier models and specialized models can do a great deal, but the harness around them, including skills and other capabilities, is really important,” he says. “Even when you have the right harness around those models and train them with data, that alone is not enough. For instance, we have 20 billion customer and employee interactions labeled as positive or negative, and that number is constantly growing.”
Zendesk encountered that limitation in its own lead-nurturing process. Adding AI agents produced a modest improvement, but left the underlying workflow largely intact.
“It’s not enough simply to put AI into your existing workflows. You really need to think about how those workflows should change,” Eggemeier says. “We initially added AI agents to our lead-nurturing process, which made it a little better, but we were still following the same process. Now we’re looking to fundamentally change how we nurture leads, with an AI agent at the center of the process.”
He compares that initial approach to paving a path that cows have already worn into the ground. “You can pave that path and gain some efficiency. But if you really want to serve your customers better, you have to reimagine how things work,” he says. “That kind of reimagining takes a partner with deep expertise who can work alongside you on site and help you make those changes.”
That puts implementation experience at the center of Zendesk’s defense against increasingly capable general-purpose AI. It also makes human expertise essential to delivering the autonomy its agents promise.
Acquisitions are Reshaping Zendesk’s AI Service Business
Zendesk has acquired seven companies since early 2024 to expand its customer support and employee service businesses. Ultimate brought virtual agents that automate support conversations, while Forethought added AI agents designed to improve as they resolve requests. Local Measure strengthened Zendesk’s contact-center technology, including voice AI and call routing.
Unleash extended its employee support capabilities, helping workers find company information and get answers through the collaboration tools they already use. HyperArc brought technology for combining and analyzing data, while Klaus added tools for evaluating customer service quality. The June acquisition of Beams gives Zendesk technology for tracking which software employees use and how much their company spends on it.
Zendesk’s specialized agents can also work within competing platforms, including Salesforce and ServiceNow. The company reported more than 1 million Custom Agent executions in seven weeks. Eggemeier says employee service already accounts for about 20% of Zendesk’s business. He identifies it and voice contact centers as particularly fast-growing areas, arguing that customers increasingly want fewer vendors responsible for more of their service operations.
“Customers don’t want to buy 17 different solutions and then have to piece them all together. Who do you hold accountable for the outcomes when you have 17 different vendors, each pointing at the others?” he says. “That’s why I think we’re going to see more and more platform consolidation. Customers want an AI platform provider and an AI service provider whose interests align with theirs and who is focused on delivering the outcomes they need.”
Market competitors bring different advantages to that contest. AI service startups such as Sierra and Decagon can pursue customers without having to move a large existing subscription business to a new pricing model. Likewise, Salesforce and ServiceNow already supply systems that store customer information or manage internal workflows, giving them an established place in the operations Zendesk wants its agents to handle.
AI Agents Could Increase Customer Service Demand
Eggemeier pointed to his own household bills to explain why AI could generate more customer service requests. After roughly three or four years without contacting his providers, he asked an AI agent to negotiate better rates for his family’s mobile plan and home internet service. The task also included his TV streaming subscriptions. When he asked how many interactions the negotiations had required, the agent reported 37. He then set a monthly reminder for the agent to check for better deals. For those providers, a customer who rarely got in touch could become a regular source of requests without personally picking up the phone.
“In the near future, 80% of interactions are going to involve people using their own personal AI agents to communicate with a company. That is going to cause an explosion in the number of interactions,” he says.
Eggemeier expects AI to handle a similar share of interactions on the company side. Even as agents take on more interactions, he expects people to oversee them on both sides, whether they represent a company or act on an individual’s behalf. “Ultimately, people will supervise the AI agents working on behalf of our company. I’ll be supervising my own AI agents as well, whether I’m using them in my personal life or in my role as an employee,” he says. “We believe people will continue to play a role in overseeing those interactions, whether they involve customer service or employee service.”
THE CONTROL LAYER VERDICT
Zendesk wants customers to pay for the work AI completes, even when fewer employees need access to its software. The company’s $100 million startup commitment, including up to two years of free access, gives it a way to win founders early. The commercial opportunity is to turn those relationships into customers willing to pay for resolutions as their businesses grow. Eggemeier’s bill-negotiating agent shows where additional demand could come from. Software can keep asking for a better deal long after a person would have stopped. Zendesk’s challenge is to turn those extra requests into completed work that customers trust enough to pay for.
