Why most AI pilots stall before they scale

Every service leader has an AI mandate. Far fewer have a plan capable of meeting that mandate in the face of their organizations’ legacy systems, fragmented data streams, complex workflows, and bolted-on approach to AI adoption. 

Freshworks Chief Product Officer Srini Raghavan has a firsthand look at the gap between AI’s promise and its payoff. The Works caught up with him about what separates companies pulling ahead with AI execution from those stuck in pilot purgatory, and why leadership and culture holdups often trump technical aspects. 

What’s changing about how service organizations need to operate in the AI era?

AI has moved from experimentation to expectation. Every service leader is now being asked to show real impact, not just pilots.

But there’s an equally big shift underway related to how service itself is changing. For years, we built service organizations around tickets, queues, and workflows—all systems designed to manage work. AI changes the premise. It lets people resolve, predict, and act across systems, not just route work through them.

But most environments weren't designed for that. They're fragmented across tools, data, platforms, and teams. By bolting AI on top, you get noise, not transformation. The organizations that succeed will rethink service as an operational system, not a collection of tools. 

Read also: The AI ROI roadmap

Where are most companies getting stuck with AI today, and why?

Most teams are applying AI on top of a fragmented foundation. They start with a few use cases, often in isolation. A chatbot here. An automation there. It works in pockets, but it does not scale. 

The reason is that AI depends on context. Without unified data, clear relationships between systems, and governed workflows, AI can’t act reliably. With inconsistent results, trust breaks down quickly. The gap isn't model capability. It's operational readiness.

What does getting AI right actually look like?

Start where the pressure is acute: High-volume requests. Repetitive tasks. The places where speed and consistency actually matter to the business. That’s where teams can prove value quickly, and earn the right to expand. 

From there, build feedback into the system. You need to see what AI is doing, where it's working, where it's falling short. Without that visibility, you're scaling on faith. That’s how trust erodes. 

But the most difficult part, and where most companies underinvest, is the foundation. AI needs context to operate well. That means connected data, clear relationships between systems, workflows it can actually execute end to end. Skip that step, and you end up with AI that answers faster but doesn't get smarter.

When those three pieces come together, organizations stop running isolated automations or isolated AI projects and start running a system that improves itself. 

What principles guide how your team builds products at Freshworks?

We focus on making enterprise-grade software usable, because ‘enterprise-grade’ shouldn’t mean unnecessarily complex. Enterprise software has a long history of adding complexity as it scales—more configuration, more overhead, longer implementations. I've lived on the receiving end of that complexity.

So, our job is to address that complexity within the product, not push it onto the customer. Teams shouldn't have to stitch together multiple systems or rebuild their foundation every time they want to try something new. If we do our job right, customers can move quickly, trust what they deploy, and expand with proven results. 

When you think about the future, what should customers expect Freshworks to make possible that isn’t possible today?

We are building toward a system where signals, context, and action live in one place. That includes signals from tickets, systems, and events. And context about how services, assets, and workflows actually connect. And AI that can act inside that enterprise-governed environment with controls that actually strengthen capabilities. 

When those three come together, service stops reacting and starts anticipating. Over time, it becomes genuinely proactive and eventually autonomous in the places where that makes sense. The goal isn't autonomy for its own sake. It's helping teams get there without adding more complexity along the way, because the organizations that add complexity to chase AI are going to end up further behind than the ones that didn't start.

Don't miss Srini and more experts at Refresh Virtual Summit 2026. Register today.