I recently spoke with ServiceNow CCO and AI Enterprise Advisor Chris Bedi about what’s holding AI adoption back, how to scale it enterprise-wide, and what it means to build toward a future powered by digital employees.
This is the fifth installment in a series of interviews with leading CIOs across global enterprises.
Asheem Chandna: Where is AI having real impact inside ServiceNow?
Chris Bedi: Broadly speaking, across machine learning, generative, and agentic, the impact has reached every part of the company.
With machine learning, our aim is for every human decision to be supported by an ML-based recommendation. We want it to feel unthinkable to make a decision without that input. In some cases, we’re already deferring decisions to the model when accuracy is high, but that’s still the exception.
One of the biggest challenges isn’t the tech. It’s adoption. Making AI ambient in the user experience is key. For aspiring entrepreneurs: UX can be the antidote to adoption friction.
We’ve leaned into adoption and now have hundreds of models in production, generating millions of predictions daily. I can offer two examples:
In field sales, reps often have nuanced compensation questions. We used to have a 10–15-person finance team handling those, with responses taking up to four days. Gen AI cut that to eight seconds. Those team members now focus on more strategic work, like redesigning comp models.
In customer support, agentic AI is an obvious fit. We categorize cases from P1 to P4—P4s are high-volume, low-complexity. P2s are lower-volume, more complex. After training AI on our SOPs and knowledge base, we saw an 18% improvement in time-to-resolution for P4s, and 54% for P2s. That surprised me. As a computer engineer, I would’ve expected the opposite. But AI had more room to add value in the complex cases: running diagnostics, pulling logs, executing scripts.
How are you measuring the value of AI?
We measure AI’s impact in four ways:
- Speed: How fast the organization moves, especially in go-to-market and engineering.
- Productivity: Straightforward—more output with the same resources.
- Effectiveness: This is newer. Improving how well people do their jobs, not just how fast.
- Experience and Sentiment: Employee and customer satisfaction.
At the enterprise level, AI’s impact ladders up to growth, margin, and revenue per employee.
We’re also looking at how AI gives people time back through summarization, meeting notes, and automation. It doesn’t replace people, but it increases capacity. Agentic AI already handles 20% of support cases, and that number is growing.
What’s the long-term vision for AI at ServiceNow?
Our long-term goal? The “AI employee.” And I’m serious. We’re building the first group of AI employee use cases now.
We define an AI employee as software that can do 80% or more of a given job. ServiceNow has about 2,100 unique roles. If we build an AI employee for even one of them, we may never need to hire for that role again. Not because we’ll hire fewer people, but because we’ll be hiring for different roles. Ones that don’t exist yet.
The path there is a collection of agents that together can handle 80% of a job’s tasks. You’ll see digital employees first in support, HR, and procurement. It’ll take longer in highly specialized areas, but it’s coming.
One idea that’s really resonating with customers is the “zero-headcount department,” like a support function with no full-time staff. They know they won’t hit zero, but a 70% reduction in headcount? That’s compelling.
Will AI employees replace humans or work alongside them? And how close are we to that 80% benchmark?
AI employees will absolutely work alongside humans. Imagine an AI recruiter supporting a human one, but with infinite scale.
Another key shift is proactivity. Most Gen AI today is reactive, waiting for prompts. But AI employees—especially those powered by agentic AI—will act independently. That’s a big leap.
As for today? It depends on the role and the maturity of the company. Technologically, we could reach 80% for some roles right now. Today, AI agents are driving 20% productivity increases across customer, HR, and IT support.
Most companies are still using AI to boost capacity: summarizing, automating simple tasks. Full-scale AI operations and digital employees are just starting to show momentum.
To be clear, we won’t be hiring less. We’ll just be hiring differently. Think back to 1999. If someone said, “You need an SEO team,” most people wouldn’t even know what that meant. We’re at a similar AI inflection point. The net-new jobs are coming. We just don’t know what they are yet.
Where are most enterprises, including ServiceNow, in their AI adoption?
Like any major transformation, it’s not linear. Many companies recently hit what Gartner calls the “trough of disillusionment.” They’ve deployed AI and then asked: “Where’s the ROI?”
But no one’s backing away. AI is a priority in almost every C-suite conversation I have.
What we’re seeing now is a pivot. 2024 was full of pilots and proof-of-concepts. 2025 is shaping up to be the year of scaled, production-level use cases. So yes, it’s still early, but we’re entering the first real innings.
At ServiceNow, we’re intentional about how we scale. We’ve already deployed agentic AI across several areas and expect to have 10–15 digital employees in place by year-end. We’ve already seen over $350 million in enterprise value from AI.
Industry-wide, there’s still a lot of noise—point solutions everywhere. But more companies are realizing they need three things:
- All forms of AI: ML, Gen AI, agentic.
- A solid data foundation.
- Strong workflow execution.
Because if AI can generate a plan but can’t execute it, what’s the point? That’s why we focus on AI, data, and workflows as the core building blocks.
Are you focused on a few big AI projects? Or are you taking a more decentralized approach, letting teams experiment on their own?
Both, and that’s true for the top-performing companies I talk to as well.
From the top down, we’re focused on value. We define the metrics that matter at the company and department level, then prioritize the projects that move those metrics.
At the same time, we create space for experimentation. That might mean providing teams with sandbox environments to try out new tools or top-down-sponsored AI hackathons. AI is now accessible enough that anyone with a good idea can build something.
So yes, innovation is blooming, but in a structured way. It’s not the Wild West. It’s what I’d call a “curated bloom”—that balance between focus and exploration.
You mentioned startups, what have you learned from working with them at ServiceNow?
It starts with learning. There’s so much innovation in startups, especially those backed by firms like Greylock, and enterprises need to stay intellectually curious. Even if you don’t adopt every idea, it’s worth seeing where smart people are placing bets.
We also try to create clear on-ramps for startups. We’re about 26,000 people—not massive, but big enough that startups can struggle to get in front of the right stakeholders. It’s on us to lower that barrier.
The outcomes are mixed—some hits, some misses. One success story is Synthesia. Instead of scheduling a full video shoot, I can type a few lines and generate a video of an AI avatar version of “me” delivering the message. It’s a smart, simple solution to a real pain point.
I always tell the recipient, “This is AI Chris Bedi,” just to be transparent. But it’s a great example of usable tech that solves a real problem.
What advice would you give to startups?
First, a word of caution: there’s agent fatigue out there. If you’re pitching an AI agent, you need to be clear on what makes it different and how it integrates. Siloed tools, no matter how clever, won’t get adopted. Enterprises need interoperability.
The biggest pain point I hear? Adoption. Everyone’s asking: “How do we move the frozen middle?” How do we go beyond innovation teams and drive real usage?
So, here’s my advice for startups:
- Make it easy to access. Ecosystem integration is key.
- Prioritize UX. A great model is meaningless if it’s hard to use.
- Be honest about what you’ve built. If it’s an app, call it an app—not a platform.
- Focus on real, discrete problems.
- Show value in 90 days. Buyers are tired.
Looking at your own AI journey, are there areas where you still think, “I wish someone would solve this?”
Definitely. And it’s often less about the what and more about the how.
Sales effectiveness is a great example. Everyone wants it. There are plenty of tools: curated content, chat interfaces, sales data platforms. But the real challenge is the last mile—making tools ambient, seamless, and part of the daily workflow. And that’s why I’m so excited that ServiceNow is reinventing CRM with an AI-first approach—so that someone is us!
Some startups help you build an account plan, but the moment a rep wants to edit or use it, the UX falls apart. So, they give up and start over.
The opportunity isn’t just in solving problems. It’s executing well. Poor UX kills adoption. If the better option is too much work, people will default to “good enough.”
