Zendesk, a global software company that develops solutions for customer and employee service, is transitioning from being a software-as-a-service (SaaS) customer relationship management (CRM) provider to focusing on artificial intelligence (AI). As part of this shift, Zendesk has established a Global Capability Centre (GCC) in India that is dedicated to innovation and AI development.
Pune Powerhouse
The company began its journey by partnering with Persistent Systems to help build its initial team and took full control over the operation a year ago. It opened its own GCC in Pune, equipped with a team of 400 and the capacity to accommodate 700. This centre is specifically focused on developing AI products.
Shashi Upadhyay, Zendesk’s President of Product, Engineering, and AI, stated, “We’re going to see the AI business slowly replacing the standard seat-based SaaS model with a much more outcome-based model.” Zendesk has fully committed to this outcome-based pricing model and released its Zendesk Resolution Platform. The platform is designed so that customers only pay for successful resolutions of their problems, and it eliminates charges if issues remain unresolved. According to Upadhyay, Zendesk is the only major global CRM company offering outcome-based pricing.
New Pricing Frontier
In 2024, Zendesk launched its AI products and saw an increase in demand by the end of the year. By 2025, customer adoption surpassed all expectations. Currently, 10,000 customers are utilising Zendesk’s primary AI offerings, which include both an autonomous AI and a co-pilot AI. Additionally, about 20,000 users are engaging with some form of AI. Notable customers include the cosmetics brand Lush, fashion retailer NEXT, music streaming service SoundCloud, and sports brand Decathlon. The company has seen significant traction in e-commerce, retail, financial services, telecommunications, and technology, with e-commerce being the fastest-growing segment, accounting for 20-25% of their business.
“AI is all about outcomes now. It has worked out very well. We’ve gone from zero to $200 million in annual recurring revenue (ARR) from AI in 18 to 24 months,” Upadhyay noted. “If AI is going to do most of the work, then customers will want to pay based on whether the job gets done, rather than paying for seats. This represents a significant shift from a seat-based to an outcome-based model.”
Upadhyay explains that Zendesk offers two types of AI: one that is completely autonomous and another that operates in co-pilot mode. When the autonomous AI agent is unable to resolve an issue, it escalates the case to a human agent, a process known as “human in the loop.” Currently, Zendesk is experiencing growth in both its seat-based and AI businesses. However, the AI segment is growing at a faster pace, having reached $200 million in just two years.
For most companies, customer service constitutes about 7-8% of their total spending. Across the CRM and customer service SaaS landscape, all major players—including Salesforce, ServiceNow, Freshworks, and HubSpot—are transitioning toward AI-driven solutions that automate and enhance customer interactions. Salesforce has launched Agentforce, which promotes autonomous AI agents within its CRM system.
ServiceNow offers generative AI features through Now Assist. Freshworks provides Freddy AI Copilot and Freddy AI Agent to automate customer support, while HubSpot has introduced Breeze AI with multiple specialised AI agents. Microsoft and Oracle also participate in this market. Salesforce reported approximately $900 million in data cloud and AI ARR for FY25, while ServiceNow’s Now Assist was around $600 million in annual contract value (ACV). Freshworks’ Freddy AI Copilot and Agent collectively exceeded $20 million in ARR.
Zendesk employs a metric called the “automation rate” to assess how many issues can be resolved by its AI agents. This metric is measured alongside the customer satisfaction (CSAT) score. They found that for e-commerce companies, the automation rate is nearly 80-90%, meaning 8 to 9 out of 10 cases handled by an AI agent are resolved. The remaining 10-15% require intervention by a human agent.
“We found that 70% of all interactions are what we classify as low-value interactions. Only 30% are high-value interactions, which are those that cannot be automated and necessitate a human touch. Therefore, humans will focus on these high-value interactions,” Upadhyay explained. He acknowledges there are concerns about job losses due to automation, but he suggests there will be more displacement than outright loss, as job roles evolve and people undertake different responsibilities.