Ask most enterprises why their AI program hasn't moved past pilots, and you'll get an answer about the model. It's not accurate enough, not fast enough, not cheap enough yet.
Srikanth Akkiraju, who has run transformation at Philips and now at ServiceNow, doesn't buy it. In a recent fireside conversation with iOPEX, he made the case that the model was never the problem. The problem is that most enterprises haven't decided what they actually want AI to change.
Five questions came out of that conversation. None of them involves picking a better model.
Do You Know the Purpose for Which You're Doing AI?
Pilots don't usually die from bad technology. They die from never having a reason to exist beyond "we should be doing something with AI." Srikanth saw this firsthand: pilots that worked, then sat unused because nobody had defined which customer, employee, or financial outcome they were supposed to move.
The second failure is a habit borrowed from the wrong era. Nine-month hardening cycles made sense when the technology barely moved between releases. AI moves every quarter now. Waiting for a perfect release doesn't produce a better product; it produces a shelved one.
"A lot of people started things because they just needed to start things. Or competition is doing it. But not really thought about the purpose." - Srikanth Akkiraju, CTO, ServiceNow
The Expert Take
Your transformation office is not a pilot factory. If every AI initiative on your portfolio doesn't have a named outcome (a specific process changed, a revenue number moved, a cost removed), it doesn't belong on the list.
Stop funding experiments dressed up as strategy. And if your teams are still running 9-month hardening cycles, they're already being left behind. Because, the enterprises pulling ahead are the ones testing in production, failing cheap, and compounding fast.
Are You Really Rethinking Your Processes, or Just Layering AI on Top of Existing Work?
Srikanth splits enterprise AI into two categories with very different economics. Everyday AI that is used for better email, faster search, drafting help, etc., is popular and roughly a wash. Employees save five minutes and spend four of them reviewing the output. It builds goodwill, not P&L.
Breakthrough AI is different, and it stays small on purpose. Three to five bets an organization can genuinely redesign around, not fifty. As Srikanth candidly said:
"Who cares? I can be working on 5,000 use cases. What is delivering value?"
That's the discipline most transformation dashboards miss. They count use cases instead of tracking which ones actually changed how work gets done.
The Expert Take
Pick three to five bets where AI can fundamentally change how value is created or delivered. Measure them the way your CFO measures a capital investment — not by activity, but by return. Everything else? Let it run on its own. It doesn't deserve a steering committee. And be ready to pull the plug as and when needed.
What Is Your Moat, and Can You Describe It Without Naming a Single Platform?
For two decades, enterprise functions have been synonymous with the platforms that run them — HR is the HR system, sales is the CRM. AI is the first technology that can separate how work moves from where transactions get recorded. That's a real unlock, and it's exactly where Srikanth pushed back on the obvious anxiety about trading one platform lock-in for another:
"Your moat is not your platform. Your moat is how you want to operate as a company."
Two banks can run identical software and still compete on completely different experiences. Because the platform enables the operating model, it doesn't define it. Get that sequence backward. Pick the platform first, design the operating model around it later, and every new system just hardens the fragmentation you already have.
The Expert Take
If your answer to "what's our AI advantage?" starts with a vendor name, you need to reassess the difference between advantage and mere subscription.
Your competitors can buy the same platform by Monday. What they cannot buy is your institutional context: how your teams make decisions, how you serve your customers, the judgment baked into your workflows over decades. That is your moat. The platform's job is to operationalize it, not define it. Design your Enterprise Intelligence Model first. Then choose the technology that fits.
If You Switched On Agent Discovery Tomorrow, What Would You Find, and Who Would Be Surprised?
Enterprises have run this experiment before, first with SaaS sprawl, then with dashboard sprawl, both cleaned up years later at real cost. Agentic AI is repeating the pattern with sharper teeth: an unused dashboard is waste, but an ungoverned agent with live system access is a live risk. Srikanth read on what most transformation dashboards still miss entirely:
"The most underrated is the workflows. You have AI to do your probabilistic thing, and your workflows to do your deterministic thing."
Agents and recommendations don't create value by existing. They create it when a governed workflow turns them into completed, accountable work. Which is also why activity metrics (agents deployed, licenses issued, use cases opened) tell leadership so little.
The Expert Take
You rationalized your SaaS stack. You cleaned up your BI dashboards. Both cost you years and millions, because you waited until the sprawl was unmanageable. You cannot afford to make that mistake a third time. Not with agents.
An ungoverned agent isn't an idle tool. It has access, it acts, and it creates liability you don't know about yet. Build your AI control tower now: every agent cataloged, every access logged, every dollar justified against a return. Governance built in 2026 compounds. Governance built in 2029 is a crisis response.
If You Redesigned Your Process Today, Would Your Investment Still Pay Off When the Technology Changes Tomorrow?
No one can lock in today's model, tool, or platform choice. The technology moves too fast for that bet to hold. But there is a way to future-proof: redesign the process itself.
Srikanth's personal experience makes this concrete. His team built a medallion-architecture data lake and an API-first integration layer years earlier for unrelated reasons. When MCP and agent-to-agent patterns arrived, those foundations were already in place.
"By redesigning a process to get it fundamentally ready for AI-related work, you will future-proof yourself. If you just bring AI into your existing process, that money is completely lost. That's just money down the drain."
A process redesigned to flow, even with human checkpoints today, can have those checkpoints removed the moment confidence allows. The foundation, not the model, is what compounds.
The Expert Take
Every model release for the next three years will be more capable than the last. The question now is whether your processes are built to absorb what’s incoming.
If your workflows still have hard-coded handoffs, point-to-point integrations, and approval chains designed for a world without AI, every upgrade will require a rebuild. Redesign those two or three critical processes now with clean data flow, API-ready architecture, and humans in the loop where confidence demands it. When the next model drops, you won't need a transformation program. You'll just flip a switch.
The One Real Question on Your Table
It's not about which AI platform we should buy or how many use cases should we run? Every enterprise in your competitive set is asking those questions. Most of them will still be asking them two years from now.
The one question that separates the enterprises moving to production from those stuck in perpetual pilot mode is this: Are we using AI to do old things faster, or to fundamentally change how we create value?
The technology is not your differentiator. Your competitors have access to the same models, the same platforms, the same vendors. What they cannot replicate is the Enterprise Intelligence Model, and you build around it — the institutional context, the redesigned processes, the governed intelligence that compounds with every cycle of work.
Every week you spend layering AI onto broken processes, funding pilots without purpose, and ignoring agent sprawl is a week your moat gets shallower.
The five questions - put them in front of your CFO, CRO, and COO. Where the answers diverge, that's not a misalignment problem. That's where your transformation actually needs to start.
Watch the full fireside conversation between Srikanth Akkiraju and Kedar Kulkarni, or talk to iOPEX about which of these five questions your organization can't yet answer.
