Why Data is the necessary but insufficient condition for AI transformation
Every CEO I talk to tells me the same thing: "We're using AI." And when I ask what that means, I get two answers — almost every time.
The first: "We've rolled out Copilot." Translation: employees are writing emails faster, building decks that look prettier, summarizing documents faster. Individual productivity. That's not transformation. That's a typing upgrade.
The second: "We have a lot of pilots." Translation: teams across the company are experimenting — a chatbot here, a proof of concept there, a hackathon last quarter. This is great for employee engagement. People feel empowered. They share tips on Slack about their favorite prompts. But pilots seldom produce sustainable change.
I've led large-scale transformations at Disney, Albertsons, and DISH — operations spanning thousands of employees, billions in revenue, dozens of systems that don't talk to each other. And what I've learned is that the hard part of transformation was never the technology. It was the operations. The workflows. The governance. The humans.
The Three Levels of Enterprise AI
Here's how I think about AI maturity in the enterprise. It's a progression — and most companies are stuck on the first rung.
Level 1: Sidekick Mode. Copilot rollouts. Chatbot access. Individual employees getting faster at individual tasks. Some companies reach a Level 1.5 — an agent that aggregates data from ERP, POS, loyalty, CRM, and market sources into a single prep report so a merchant doesn't spend half a day pulling data before a vendor meeting. Useful — but still individual productivity. If this is your AI strategy, you don't have an AI strategy. You have a software license with a few curious people.
And letting a thousand flowers bloom doesn't move the needle. A thousand small productivity gains, uncoordinated and unmeasured, feels like progress. It isn't. Focus is what separates motion from progress.
Level 2: Surgeon Mode. This is where it gets real. You take one end-to-end business process — customer onboarding, procurement, claims processing, demand planning, financial close — and fundamentally rethink it as a hybrid human-agent workflow. Not "add AI to a step." Rethink the entire process from trigger to outcome.
Here's a thought experiment I use with boards: imagine if gasoline becomes nearly free - say, a penny to fill up your tank. What would change? Everything. You'd rethink where you build factories. You'd rethink supply chains, commuting patterns, the entire economic geography. That's what's happening with intelligence right now — the cost of generating an insight, analyzing a contract, triaging a customer inquiry is collapsing toward zero. And most companies are responding by letting their employees type faster.
Now apply that to a real process. Mortgage processing today has four distinct roles: opener, processor, underwriter, closer. Each role exists because of a specific combination of capabilities, skills, regulatory requirements, and training needs. The boundaries between these roles were drawn decades ago based on what humans could do and what certain skills cost. Given that intelligence is now nearly free, should those boundaries exist at all? If specialized AI agents can handle document verification, data validation, compliance checks, and status communication — all based on the latest underwriting rules set by the company — do you still need four roles? Or do you need two roles, one customer-facing and one for underwriting, at a different gearing ratio and different productivity expectations?
That's not adding AI to the processor's desk and making one person 10% faster. That's rethinking what a mortgage operation looks like from scratch — with an expectation of 3X throughput improvement.
Level 3: SuperCo Mode. At its highest aspiration, every process, every function, every value pool — can be reengineered as though the company were born AI-native: process and people, governance and decision rights, technology and tools, risk management, customer and ecosystem engagement. This isn't a project. It's a new company wearing the old company's clothes. Most incumbents can't jump here directly without proving out Level 2 as a bridge.
But PE firms are skipping straight here with some of their portfolio companies. The playbook is to buy a company that has the things that are hard to build from scratch — brand, domain expertise, licenses, customer contracts, distribution relationships. Then hollow out the operations and rebuild them as AI-native. Keep the assets. Replace the operating model. Re-launch. With absolute ownership and mandate, they blow past the inertia and nostalgia to execute their thesis on their timeline with skills they are building up. Risky, but ambitious and high reward.
The question for your board isn't whether this playbook works. It's whether your company is the acquirer or the target.
Advancing beyond the Sidekick mode is an Operations, not a Data challenge
In my experience, most companies have spent months discussing data strategy. Very few have spent equivalent time discussing how their operations need to change. The dominant narrative in AI strategy is "get your data house in order." Data quality. Data architecture. Data governance. And yes — data matters. I spent years at Albertsons building a single source of truth for customer data, and it was foundational to everything we did afterward. But data readiness is necessary and wildly insufficient.
To rethink and implement a process as a human-agent workflow, you need a leader who understands the process at a deep level of granularity but can also reimagine it using AI. Most people in an organization who can do the former are unable to do the latter.
And even when you get the redesign right, there's a risk most organizations aren't prepared for: the rubber-stamp problem. When you keep a human "in the loop" for oversight, the human starts approving whatever the AI recommends — not because they're lazy, but because the AI is right 95% of the time and the cognitive cost of genuinely evaluating each recommendation is high. The organization believes it has oversight. What it actually has is a false sense of security. If humans are in the loop, they need to be making genuine decisions — not clicking "approve." That means routing only low-confidence cases for human review, running randomized audits, and making humans accountable for outcomes, not approvals.
This is why Levels 2 and 3 require strong operators who are systems thinkers, not only CIOs and Data experts.
The Unicorn Role You Are Missing
To move beyond Level 1 (Sidekick), you need a leader who sits at the intersection of four capabilities that rarely exist in the same person:
Systems thinking. The ability to look at a process, a function, or an entire business and rethink it from first principles. Not incrementally improve what exists — question why it exists. Why are there four roles in this workflow? Why does this approval chain have six layers? What would this look like if we designed it today with no legacy constraints? Systems thinkers see the connections between process, technology, people, and risk that specialists miss. They're the ones who ask "why does this boundary exist?" before anyone else realizes it's a boundary.
Operational leadership and credibility. Deep understanding of end-to-end process mechanics — governance, risk tolerance, checks and balances, and the messy reality of how work actually gets done in a large organization. This isn't something you learn from a vendor demo. It comes from years of running things.
AI fluency. Not the ability to build models — the ability to assess the art of the doable. What can agents reliably handle today? Where do they hallucinate? Where do the guardrails need to be? Where does the risk-reward not justify automation? This is applied AI judgment, and it requires enough technical fluency to ask the right questions without needing to write the code.
Change management. Rethinking a process means rethinking jobs. It means convincing a VP of operations that an agent can handle 70% of their team's intake volume — and that this is a good thing. It means finding quick wins that build credibility before attempting the hard stuff. And it means managing the very human anxiety of people wondering whether they're being automated out of a role.
This is a unicorn role. Most companies don't have this person. Finding or developing this leader — and their team — is, in my view, the single highest-leverage move a CEO can make right now.
The Race You're Already In
Here's where this gets urgent. While incumbents are celebrating their Level 1 productivity gains, a new class of AI-native competitors is being born at Level 3. They're small today. They don't have your distribution, your brand, your customer relationships.
But they're building operating models that assume intelligence is free from day one. No legacy processes. No organizational debt. No governance frameworks designed for a world where every decision required a human in a swivel chair.
And consider the pace of change we're living through. Google took eleven years to reach one billion searches per day. ChatGPT got there in two years. The capability curve is steepening at a rate that has no precedent in enterprise technology. The next generation of AI-native companies will build to scale even faster — because they're not transforming anything. They're starting clean.
The fundamental question every board should be asking is: "Can our AI native competitors achieve distribution faster than we can achieve innovation". Because if they can, your 10% productivity improvement isn't a strategy. It's a rounding error on the way to irrelevance.
What Boards Should Do
My advice to boards is three things:
Audit where you actually are. Most companies think they're further along than they are. If your AI deployment is limited to Copilot licenses and a portfolio of pilots, you're at Level 1. Own it. Overlay that on where your industry is headed. The first step to progress is an honest diagnosis.
Stop letting a thousand flowers bloom. Start picking one garden. Choose a single high-volume, high-cost process — customer service, procurement, mortgage processing, financial close — and commit to rethinking it end-to-end. One process. Full redesign. Measured economics. This is how you build the organizational muscle for Surgeon Mode.
Find your unicorn. Put an operational leader with AI fluency, systems thinking, and change management instincts in charge. Not the CIO. Not a consultant. Someone who has run a P&L or a large ops team, and knows that the hardest part of any transformation isn't the strategy deck — it's the Tuesday morning when three departments push back and the executive sponsor goes quiet.
The companies that move from Sidekick to Surgeon will earn the right to become SuperCos. The ones that stay at Level 1 will eventually discover that the PE firms, the AI-native startups, and their most ambitious competitors didn't need their permission to start the race.
In an AI world where intelligence is nearly free, the cost of inaction can be very expensive.
Narayan Iyengar is an independent board director, former Fortune 500 C-suite executive, and helps boards and C-suites navigate AI strategy and enterprise transformation. If you're a board member or CEO trying to figure out where your company sits on this spectrum, I'd welcome the conversation.