Citi AI Summit 2026: Enterprise AI - Moving from experimentation to execution

During this panel, I had the pleasure of facilitating a discussion among leaders who are building and operating AI systems inside large organizations. The discussion focused on what is actually slowing AI adoption in production environments, which use cases are generating real returns today and how agentic workflows are beginning to reshape core enterprise functions such as identity, CRM and customer data management.

I was joined by Gary Kotovets, Chief Data, Analytics and AI Officer, Dun & Bradstreet; Rodger Desai, Founder and CEO, Prove Identity; and Ishan Mukherjee, CEO, Rox. By the end of the panel, it was clear that data readiness, organizational design and governance were the keys to fully realizing the potential of AI.

Data quality

Data, not models, is the dominant bottleneck in enterprise AI. Panelists addressed how messy, fragmented or untrusted data can sink even the most well-intentioned AI initiative.

Before agents can reason or act effectively, enterprise data must be cleaned, standardized, unified and made accessible across systems. This work is slow, operationally complex and largely invisible, often taking months before any AI system can deliver value.

Many enterprises struggle with inconsistencies such as multiple identifiers for the same customer across systems. Without resolving these foundational issues, even advanced AI agents produce unreliable outcomes.

Individual productivity vs. organizational impact

Before agents can reason or act effectively, enterprise data must be cleaned, standardized, unified and made accessible across systems. This work is slow, operationally complex and largely invisible, often taking months before any AI system can deliver value.

The gap between those two realities is not a technology problem, it is an organizational one. Unlocking enterprise level impact requires redesigning workflows, redefining roles and managing change across teams. Simply deploying copilots or one to one AI assistants does not translate into scalable performance improvement.

Identity as the backbone of the agentic economy

Questions of identity and trust become foundational when it comes to agentic AI. Panelists stressed that organizations need to set up advanced identity infrastructure that spans humans, agents and systems.

Without strong identity controls, enterprises face growing exposure to fraud, misuse and unauthorized activity. At the same time, robust identity systems create new opportunities to prevent losses in real time. As agents take on more responsibility, identity becomes the connective tissue that enables trust, accountability and compliance across automated workflows.

Reviewing system of record for the customers through an agentic lens

The panel also explored how CRMs are being reinvented. AI native CRMs are designed with agents as the primary users, automatically capturing activity, normalizing data across multiple sources and proactively surfacing insights and actions. The challenge, however, is that enterprises often lack a unified view of the customer. When the same entity exists under multiple names across systems, agents struggle to operate coherently. Panelists agreed this was a key issue to solve in the months ahead.

Measuring outcomes is now table stakes

When it comes to AI deployment, cost-savvy enterprises are now insisting on measurable ROI. The era of AI pilots justified solely by experimentation or token consumption is over.

As a result, evaluation infrastructure is becoming as strategically important as the AI models themselves. Enterprises must be able to measure what improved, by how much and why. Organizations that cannot demonstrate value will struggle to justify continued investment, while those that can are repositioning AI from a cost center to a growth driver.

Building the enterprise AI core

Enterprise AI success depends less on model sophistication and more on foundations: trusted data, customer and employee identity infrastructure, governance and outcome oriented design. Companies that invest in these layers are not just improving efficiency but also rebuilding how work gets done

In sum, enterprises are either preparing for an agent driven future or delaying the inevitable and risking being left behind.

Are you building enterprise AI solutions? Reach out to vibhor.rastogi@citi.com.