Why Context and Decision Intelligence Are Critical to Enterprise AI

Devavrat: That’s a great question. As you rightly pointed out, we’re on the cusp of AI moving from the experimental stage into mainstream business adoption. I have been teaching AI as a professor at MIT for more than 20 years, so it’s no surprise that we’re very bullish about the future of AI in the enterprise.

How AI Can Transform Business Decision-Making

Where I believe AI has the potential to create the biggest change for businesses is in decision-making and decision support—or, put another way, in bringing decision intelligence to the enterprise.

When we think about how organizations make decisions today, whether at MIT, MIT Technology Review, Celonis or other large organizations, we generally make several different kinds of decisions. Some are strategic and made occasionally, while others are operational and need to be made frequently.

In both cases, processes are usually already in place: do this first, then that, followed by another step. Many people contribute to these adjustments and perform what Jensen’s fans call “tasks.” By completing these tasks step by step, an organization eventually arrives at a decision.

These tasks, or roles, are where AI can introduce agent power at an atomic level. AI can support some or all of the decisions involved and potentially automate tedious work. But to do that effectively, AI needs to understand the context and determine which roles and tasks should be performed. It cannot operate in a vacuum.

When AI is combined with context in the right way, it can bring what I call decision intelligence to the enterprise. I believe that is where AI will create the biggest change.

The Commercial Risks of AI Without Business Context

Megan: Digging a little deeper into this, many organizations are now adopting AI with the potential to make autonomous decisions. What are the commercial risks if context is not introduced at the same time?

Devavrat: Let’s return to the settings I described. Organizations are trying to automate, support or enhance roles depending on the task and the type of decision involved. At the same time, frontier labs have built impressive models and AI infrastructure that contain an enormous amount of knowledge about the world.

Now imagine that an organization wants to make a small decision in a very specific business context. Even without direct access to the relevant context, an extremely intelligent system might eventually determine the right answer by asking many questions and reasoning through the situation. The problem is the cost of getting there.

If the system performs an enormous amount of computation to complete a relatively simple task, the organization may spend far too much computing power. That is one risk.

In reality, the system may also miss important context, constraints or guardrails and make a mistake. Sometimes that mistake will not matter. It may be like telling someone to go to Restaurant A and discovering that dinner is unavailable. But the consequences can be much more serious if the system is making financial projections or decisions that could damage a company.

How Organizations Can Add Context to Enterprise AI

Megan: In a corporate environment, these are business-critical decisions. If they are wrong, the consequences can be significant. On a practical level, how can organizations add the context layer we are discussing? For example, how does Celonis provide operational context through a real-time digital twin of the organization?

Devavrat: It may be helpful to look at the history of Celonis. Celonis was created with a mission to help organizations adhere to ideal processes, build better processes and help people work across multiple processes.

That is the company’s core DNA: what I would call process intelligence. Organizations can bring their processes together, digitize them and improve their understanding of where problems occur. They can then redesign processes to make the business operate more effectively.

Connecting Process Intelligence With AI Agents

If we want to bring decision intelligence into this environment, each atomic step in a process may previously have been performed by an expert or another employee. Now, AI agents can help perform or support those steps.

Understanding the context of processes within an enterprise and introducing AI agents—what we like to call role agents—is the right way to bring decision intelligence to businesses. In this way, Celonis provides the context organizations need to make better decisions.

Megan: And in terms of how this works, how does the context layer relate to the system-of-record layer below it and the layer above it?