Making The Case For Context-aware AI

TL;DR

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

A MIT Technology Review Business Lab report, produced with Celonis, argues that enterprise AI agents need operational context to make useful recommendations within real business processes. Celonis chief scientist Devavrat Shah recommends starting with a specific business decision, testing AI on difficult steps, and measuring results before expanding use.

A MIT Technology Review Business Lab report published October 22 argues that enterprise AI systems need information about how a business operates to make useful recommendations and support decisions. Devavrat Shah, Celonis’s chief scientist of enterprise AI, says agents that handle individual tasks without understanding their place in a larger workflow can miss constraints and relationships that affect the outcome.

The report presents operational context as a requirement for moving AI beyond experiments and pilots into everyday business use. Shah describes the broader aim as “decision intelligence”: using AI to assist with tasks that contribute to decisions, while accounting for the processes and conditions surrounding those tasks. “You can’t just do that in a vacuum,” he says.

Without that context, the report says, agents may overlook important constraints, make costly mistakes, or use excessive computing resources while trying to determine what information they need. It proposes connecting AI to the processes and data enterprises already use, with a contextual layer that can help agents follow workflows and operate within guardrails. Their recommendations could then be reviewed and approved by people.

Shah advises companies to begin with a specific business decision, identify its most difficult steps, and test whether AI can improve them. Organizations can assess whether to expand an application by measuring changes in speed and value, the report says. It does not provide figures from a particular company pilot or an independent evaluation of results.

Why Workflow Context Shapes AI Decisions

The argument matters to organizations deciding whether to expand enterprise AI beyond demonstrations. A system that produces a plausible answer may still be of limited use if it does not know the business rules, dependencies, and handoffs that shape a real decision. Connecting AI to operational information is presented as one way to make its outputs more relevant and keep its work within defined limits.

The report also links context to the economics of deployment. Shah’s suggested test—select a decision, focus on its difficult steps, and measure speed and value—offers a way for companies to evaluate an application before scaling it. That is a proposed approach, not evidence that context-aware systems will deliver a specific return for every organization.

If repetitive work and coordination can be automated, employees may have more time for complex decisions. But that outcome depends on how organizations implement the systems. The report says employees need training to interpret AI outputs and their implications; handing decisions over to agents without that understanding would not meet the report’s stated goal of supporting human judgment.

From AI Pilots to Business Processes

The report situates its argument amid a push by business leaders to move from AI experimentation toward broader operational use. It describes a gap between the capabilities of general-purpose models and the specialized conditions under which organizations make decisions. The focus is not on a new model launch or a quantified deployment, but on how companies might connect AI to their existing work.

Shah distinguishes between frequent operational decisions and less frequent strategic ones. Both involve processes and tasks, he says, and AI agents can assist with parts of those processes. He sees potential applications in finance, workforce management, marketing, and sales, including resource planning. These are areas of possible future use described in the interview, not confirmed deployments or measured results.

The item is an episode of Business Lab, a MIT Technology Review program produced in partnership with Celonis. Shah is identified as Celonis’s chief scientist of enterprise AI. That partnership is relevant when weighing the material: the case for operational context includes his perspective as a company executive, and the report does not present it as an independent study comparing AI deployments.

“You can’t just do that in a vacuum.”

— Devavrat Shah, Celonis chief scientist of enterprise AI

Evidence Behind Context-Aware Claims

The report makes a case for operational context but does not establish how much it improves AI performance across companies or industries. It provides no comparative benchmark, deployment data, or quantified return on investment for the approach described. Shah’s examples of potential gains should be understood as his assessment, not as independently verified outcomes.

It is also unclear what technical design or governance standards would make a contextual layer reliable across different enterprise systems. The report does not specify how companies should resolve outdated or conflicting process data, determine which information agents may access, or assign responsibility when a recommendation is wrong. Those questions may shape whether a system works safely and usefully in practice.

The interview’s producer partnership with Celonis is disclosed, but the report does not include perspectives from other vendors, independent researchers, or customers describing deployments. The scale and maturity of context-aware enterprise AI adoption therefore remain unclear.

Testing AI in Specific Decisions

The next step proposed in the report is for companies to choose a business decision, identify the tasks that make it difficult, and test whether AI can help. They would then measure changes in speed and value before deciding whether to expand the application. The report does not announce a formal testing program or a specific schedule for such trials.

Organizations considering this approach would also need to define workflows and guardrails, decide how people review recommendations, and prepare employees to interpret the output. Shah’s longer-term outlook includes operational tasks and planning across several business functions, but the report gives no timeline for those uses. Evidence from measured deployments would help clarify whether the proposed benefits extend beyond individual pilots.

Key Questions

What is the report’s main argument?

It argues that enterprise AI agents need operational context—including relevant processes and business conditions—to support decisions usefully rather than handle tasks in isolation.

Who is Devavrat Shah?

Shah is identified in the report as Celonis’s chief scientist of enterprise AI. He describes the use of AI to support business decisions as “decision intelligence.”

How does Shah suggest companies start?

He recommends selecting a specific business decision, finding its most difficult steps, testing AI on those steps, and measuring changes in speed and value before considering wider use.

Does the report show that context-aware AI produces a measurable return?

No. It offers an approach and Shah’s views, but provides no company pilot results, comparative benchmark, or quantified return demonstrating the size of any benefit.

What role do employees have in this approach?

The report says employees would need education to understand AI outputs and their implications. It describes AI as supporting work and decisions, not as a reason to hand every decision to an agent.

Source: rss