The next AI shift in IT isn't speed. It's prevention.
Freshworks VP of AI products Venkat Venkataraman on why "shifting left" means catching problems before they become tickets, and what it takes to get there
For decades, service management has run on the same cycle: something breaks, a ticket gets filed, a human resolves it. And AI is making that loop run exponentially faster.
But, what if we could do things differently? That’s what Venkat Venkataraman, Freshworks VP of AI products, believes is possible. His perspective is that real transformation isn't a faster version of the old model, but a different one, where problems are caught and neutralized before an employee ever notices, and where IT's value gets measured by what never broke rather than by how fast the break got fixed.
Recently, we sat down with Venkataraman to hear more.
AI-powered automation is already producing real gains for service teams: higher resolution rates, faster response times. Yet a lot of IT leaders say it doesn't feel transformative. Why?
Venkat Venkataraman: I don't dispute the numbers. Across our customer base, AI agents and Copilot are driving 65%-plus employee query deflection, roughly a 40% lift in agent productivity, and a 77% reduction in average resolution time. Those are real, measurable gains. But most of that improvement is still happening inside the same old loop: problem, ticket, resolution.
AI is making an existing operating model run faster and more efficiently than ever. But because the underlying model is essentially unchanged, that makes it feel less transformational. Even if the results showcase impressive levels of efficiency.
Shifting left means moving the point of intervention earlier than the ticket itself. Early enough that, ideally, the ticket is never the first point of contact at all.
Venkat Venkataraman
Vice President, Product - AI Platform & Strategy, Freshworks
You’ve talked about how AI can help service operations “shift left.” What does that look like and how is that changing the old model?
Shifting left means moving the point of intervention earlier than the ticket itself. Early enough that, ideally, the ticket is never the first point of contact at all. Instead of assisting with a problem after it's reported, AI watches for the signal that precedes it: an anomaly, a change with downstream risk, a pattern that has historically preceded an outage. It predicts what's about to go wrong and triggers remediation automatically, with the ticket pushed to the background. It becomes a record of what happened, not something IT has to manage.
If today many IT teams are deploying AI as an automation layer on top of an unchanged loop, then shifting left moves us to a new operating model that’s proactive by design, built on signals, predictions, and interventions.
Instead of reacting to tickets, we get to prevent them from even happening.
If IT shifts to prevent issues before they happen, how will that change how leaders quantify IT’s value to the business?
This is where the conversation has to move past the metrics IT has always reported. Average handling time and deflection rate measure how efficiently you run the old loop, but they say nothing about the incidents that never occurred.
As prevention becomes the point, the value shifts toward what didn't happen: outages avoided, risky changes caught before rollout, capacity freed up because fewer people are stuck triaging routine breakage. Time-to-insight for IT leaders is also moving from hours to minutes, which changes the kind of conversation you can walk into with the business, from "here's what we closed" to "here's what we kept from ever becoming a problem."
It’s a shift that focuses IT’s value on outcomes, not activity. And as AI identifies recurring issues, closes knowledge gaps, improves workflows, or surfaces risky patterns earlier, all of these become part of the business case. IT is not just improving efficiency. It’s improving resilience, continuity, and capacity to keep the business moving forward.
What steps can IT leaders take to start shifting left today? What do you tell customers who are just getting started with AI agents?
I tell customers to start simple, to first automate routine, well-documented, high-volume work. Things like password resets, leave requests, or policy questions, areas where knowledge and workflows already exist, the risk is manageable, and teams can demonstrate value quickly.
Once you have that starting point, build and test your approach in a limited channel, so you can learn, adjust, and gain confidence before you deploy to production.
And, through all these steps, always monitor and learn from the patterns you see, whether it’s resolution rates, escalations, or any other signal, so you can continually tune and improve outcomes.
How McLaren, New Balance & more turn service performance into a competitive edge
Managing a fleet of AI agents is very different from managing a team of people. What should IT leaders be thinking about to keep their people and data safe?
This is something that needs to be built into your approach for building and deploying agents. We think about this in three big ways. The first is managing the agent lifecycle: creation, testing, deploying, and fine-tuning their performance and behavior. Then, we look at governance for agents, so that at any point you know who owns the agent, what data and systems it can access, what actions it’s authorized to take, and where human approval is required. You also need clear guardrails, audit trails, and the ability to pause or shut down an agent quickly if it behaves unexpectedly.
Finally, it's important to have continuous improvement loops, because AI agents aren’t a “set and forget” tool you can walk away from. They require ongoing monitoring, not just to ensure their performance meets expectations, but to also continually tune and improve them over time.
As we continue to add agentic capabilities to Freshworks products, this is an area we continue to invest in to give our customers the transparency and the granular control they need over agentic behaviors.
