A Cognitive Handshake™ with Sandeep Maira
Enterprise AI has no shortage of pilots.
The harder question is whether those pilots are actually changing how the enterprise works.
That distinction emerged quickly in my recent Cognitive Handshake™ with Sandeep Maira, an enterprise AI transformation leader and founder of Raven Risk AI. Our conversation moved from operating models and process redesign to ROI, institutional knowledge, and the controls required to scale increasingly capable AI.
The central idea: the next phase of enterprise AI is not primarily about automating more work. It is about reimagining the work itself.
Automating yesterday’s process is not transformation
One of the traps Sandeep sees is that organizations can successfully put AI into production and still miss the larger opportunity.
Companies can take processes that are already less than ideal and use increasingly capable agents to reproduce what people do today. That may improve efficiency, but it can also automate yesterday’s assumptions.
“The missed opportunity here … is actually taking decades-old processes and then reimagining them, leveraging AI, but also seeing what really the ultimate customer wants to achieve from those processes.”
There is a meaningful difference between asking, “How can AI make this existing process faster?” and asking, “If we designed this process today, with AI available from the beginning, what would the work look like?”
The first can create efficiency. The second can create transformation.
Scaling AI is a multidisciplinary problem
Reimagining the process does not mean handing the problem to an AI team.
Sandeep identified four groups that need to work together: business and domain experts, AI technologists, risk and control professionals, and business sponsors.
His observation from initiatives that struggled was straightforward: one or more of those constituencies was often missing.
That resonates with what I have seen in large transformations. AI may be the enabling technology, but enterprise transformation crosses organizational boundaries. The people who understand the work, the people who understand the technology, the people accountable for risk and controls, and the executives able to change the operating model each hold a different piece of the answer.
The failure mode is treating AI as a technology deployment when the actual challenge is an enterprise operating-model change.
Quick wins and foundations are not competing strategies
Executives face another tension: they need measurable results quickly, while scalable AI also requires data, knowledge, controls, architecture, and organizational capabilities that take longer to build.
Sandeep argues that enterprises need both.
Quick wins can demonstrate value and generate economic support for the larger transformation. He described a “self-funding aspect” in which early ROI can help fund the deeper capabilities required for scale.
That offers a useful way out of a common false choice. Enterprises do not need to spend years building a perfect foundation before producing value, nor should they accumulate disconnected pilots indefinitely.
My synthesis of that sequencing is:
Prove value → learn → strengthen the foundation → scale what works.
Sandeep’s operating-model view: centralize first, federate deliberately
The organizational model should evolve too.
Based on his experience across multiple firms, Sandeep favors a more centralized hub-and-spoke model early in the transformation. Scarce skills, technology partnerships, standards, and enterprise priorities can initially be managed more effectively through a strong hub with representation from the businesses.
As maturity increases, more capability can move outward.
The point is not permanent centralization. It is deliberate decentralization.
That distinction becomes more important as AI becomes easier to deploy. Democratizing access without shared capabilities can produce duplication, inconsistent controls, and fragmented architectures rather than enterprise transformation.
The constraint is shifting toward controlled capability
Our conversation became particularly interesting when we turned to agentic AI.
Sandeep’s view is that raw AI capability is becoming less of the central question for many enterprise use cases. The challenge increasingly becomes whether organizations can surround those capabilities with sufficient guardrails, accuracy, consistency, reliability, and intervention mechanisms.
I see a closely related shift: as capability advances, the enterprise constraint increasingly moves toward control, resilience, and accountable decision-making.
That does not mean model capability is solved, or that every AI system can reliably perform every enterprise task. It means that for a growing set of use cases, the question is changing from “Can the AI do this?” to “Can the enterprise trust it to do this repeatedly, safely, and at scale?”
ROI has to show up in the business
None of this matters if AI transformation cannot produce measurable enterprise value.
Sandeep pointed to implementations he has worked on that, according to him, generated approximately 60% productivity gains and millions of dollars in annual savings in complex business workflows.
Those are Sandeep’s reported results, but they illustrate the standard executives should increasingly demand:
Not number of pilots. Not number of users with access to an AI tool. Not number of agents deployed.
Business outcomes.
Productivity. Cost. Revenue. Risk. Cycle time. Customer outcomes. Decision quality.
Activity tells us that AI is being used. Outcomes tell us whether it is creating value.
Models may commoditize. Institutional knowledge can still differentiate.
That brought us to perhaps the most consequential question of the conversation.
If increasingly powerful foundation models are broadly available, where does sustainable competitive advantage come from?
Sandeep sees proprietary data as part of the answer — but not all of it.
He emphasized human domain expertise and institutional knowledge: the accumulated understanding of how a business actually operates, makes decisions, interprets exceptions, and navigates complex domains.
His argument is that organizations can encode more of that knowledge through semantic layers and knowledge graphs, allowing AI systems to work with context that a general-purpose model does not inherently possess.
For years, enterprises have talked about data as an asset. Agentic AI may force them to recognize something broader: institutional knowledge is an asset too.
Much of that knowledge remains trapped in people, processes, documents, and organizational memory rather than represented in a form machines can reliably use.
The real race is not for more AI
The conversation with Sandeep reinforced something I expect we will see much more clearly over the next few years.
The organizations that create durable value from AI may not be those with the most pilots — or even those with access to the most powerful model.
They will be the organizations that can redesign work around outcomes, bring business, technology, and risk together, build foundations while producing measurable wins, encode what they uniquely know, and create enough governance and resilience to let increasingly capable AI operate at scale.
That is a much harder transformation than deploying another AI tool.
It is also a much more defensible one.
And perhaps that is why, when I started our conversation by asking Sandeep whether the AI glass was half full, his answer was simple:
“In my opinion, the glass is actually half full.”
I agree.
The opportunity is substantial. But realizing it will depend less on how much AI enterprises can deploy — and increasingly on how well they can transform around it.
That is a Cognitive Handshake™
About Sandeep Maira
Sandeep Maira is an enterprise AI transformation leader and founder of Raven Risk AI. His experience spans large-scale transformation across major consulting and financial-services organizations, with a focus on enterprise and agentic AI, operating-model change, and measurable business outcomes.
