Meet the Authors

Key Takeaways What you need to know
  1. Agents need shared business context, not just a database connection, to act reliably across processes.

  2. Reltio's context intelligence approach moves from unification to understanding to action.

  3. Radisson Hotel Group documented its use cases and put business stakeholders at the center of its AI operating model.

For years, the enterprise data playbook meant collecting information from different systems, analyzing it, extracting insights, and providing those insights to the right people so they can make better decisions.

Agentic AI is rewriting the rules. As AI agents become capable of making decisions and taking action across business processes, they need to understand what that information means, how different pieces of data relate to one another, and the business rules governing what they should do with it.

Kash Mehdi, VP and Field CTO at Reltio, an SAP company, describes this as a shift from the “analytics era” to the “agentic era.”

Explore related questions

“What used to be the end line for companies’ data leaders, it is now the starting line,” Mehdi said. “Now with agentic AI, it’s not just humans, it’s AI agents, and they need to work off of the same context.”

Connecting the Dots

In a typical enterprise application environment, a customer service representative might look at a CRM system to understand a customer, switch to an ERP system to check a product or supplier, and then open another application to review support tickets.

“We were the ones stitching together this shared context in their head,” Mehdi said.

AI agents can’t do that unless organizations make those relationships digitally available. It becomes more difficult when information remains scattered across different systems and architectural environments, with inconsistent definitions and little explanation of how the data should be interpreted.

Adding to the complexity are the constantly changing relationships. In life sciences, for example, ‘entity resolution’ may address much of the identity challenge, but the harder data accuracy problem to solve can be ‘affiliation’ and determining which integrated delivery network owns a hospital, which group purchasing organization governs an account, and how those hierarchies change over time.

Definitions can be just as challenging. Ambiguity over who has decision-making authority can produce competing definitions of something as fundamental as a key opinion leader or net sales. Multiple hierarchies can reflect legitimate business realities rather than data defects that need to be eliminated.

“Fragmented data produces fragmented results,” Mehdi said.

Connecting an AI agent to a database won’t solve the problem. Enterprises instead need to connect information about customers, products, organizations and other critical entities while providing the business meaning that allows AI to understand the relationships among them.

“It’s also about teaching the agents how your business works,” Mehdi said.

From Data to Context

Reltio describes its approach to these challenges as “context intelligence,” built around three steps: unification, understanding and action.

The first involves unifying and harmonizing information residing across SaaS applications, on-premises systems, data warehouses and other environments.

“It’s about enriching this data with semantics, being able to connect your customer domain with a product domain, an organization domain and other attributes to make sense of your data ecosystem,” Mehdi said.

That information is delivered to downstream applications or AI agents that use the context to make decisions or initiate actions.

Reltio’s platform supports this model through multi-domain master data management, Intelligent 360 profiles and an intelligent graph that connects data and relationships. Organizations can augment unified profiles with transaction and interaction data, while agentic capabilities can work with unstructured information, data quality, work assignments, matching and enrichment.

For example, the financial services industry might believe unstructured data is largely a “pre-LLM challenge” and that extracting reliable answers from structured, tabular information can now be more difficult. Since the extraction itself can create noise, it’s important to combine extraction with resolution so information is connected to the entities and relationships that give it business meaning.

Radisson Hotel Group is a prime example of how organizations can build a foundation for the transition from analytics toward agentic AI. The company documented its use cases, forcing the organization to answer an important question before deploying AI: What business problem are we actually trying to solve?

By asking the tough questions up front, Radisson has gained measurable benefits by improving B2B data quality from 30% to 35% and reduced duplication in customer records from 20% to 25%.

“Every use case, there is a common pattern that they need a unified foundation to unlock their business goals,” Mehdi said.

Put People at the Center of AI

Radisson also developed an AI operating model that placed people at the center of its transformation.

With approximately 150 use cases to manage, the company assembled a diverse group of business and IT stakeholders and evaluated opportunities against corporate strategy and business goals such as customer experience and revenue. Use cases could then be embedded into an existing process, purchased from a vendor or developed through the company’s AI lab.

Mehdi warned that organizations should treat a lack of engagement from business stakeholders early in an AI initiative as a “massive red flag.” Too often, technology teams build something first and worry about adoption later.

“When it comes to operationalizing things, one of the biggest challenges that I hear is, ‘Hey, we built it, nobody came,'” Mehdi said.

And therein lies one of the most important lessons learned in the agentic era. Giving AI agents better data and more context can help them understand how a business works. But turning that intelligence into business value still requires people who understand which problems are worth solving, which decisions matter and where AI can make a meaningful difference.

Events

07Dec
SAPinsider Technology Executive ForumPhoenix, Arizona, United States
View All