Context-Aware Enterprise AI: Celonis Makes the Case for Decision Intelligence
Celonis argues enterprise AI needs process context, guardrails and human review to cut costly errors, wasted compute and speed business decisions at scale.
Summary
On October 22, 2026, Celonis chief scientist of enterprise AI Devavrat Shah, an MIT AI professor for more than two decades, argued that enterprise AI needs operational context to deliver “decision intelligence.” Agents can support, augment or automate tasks within operational and strategic decisions, but without each role’s workflows, constraints and guardrails, they may waste massive computing resources, miss critical conditions or make errors ranging from poor recommendations to ruinous financial forecasts.
Celonis positions its process intelligence and Context Model as a bridge between systems of record, workflow tools and event logs below, and business users or agents above. It abstracts processes, assigns certified role agents within defined guardrails and lets other agentic infrastructure consume outputs that experts can review, approve or sometimes automate. Records need not be complete or perfectly clean, although vendor access requires classical information security checks. Shah advises starting with one decision, mapping its process, applying AI to the hardest step, then measuring speed and value before scaling. Automating repetitive labor and coordination could shift employees toward strategic work, but requires rapid AI and tool education. Potential uses include financial, workforce, marketing, sales and product planning.
Positives
- Celonis’s Context Model can draw operational context from existing workflow tools and event logs, even when records are incomplete or imperfectly cleaned.
- Certified role agents operate within defined responsibilities and guardrails, giving organizations clearer control over what each agent may do.
- Human experts can review and approve AI outputs before automation, preserving oversight for consequential decisions.
- Testing AI on one difficult process step lets organizations measure speed and value before committing to wider deployment.
- Repetitive labor and coordination could decline, freeing employees to focus on strategic decisions and enterprise planning.
Risks & concerns
- Context-free agents may overlook business constraints and guardrails, producing mistakes that range from poor recommendations to damaging financial forecasts.
- Agents may consume massive computing resources while trying to infer operational information that a contextual layer could provide directly.
- Connecting vendor software to systems of record and event logs requires classical information security checks and organizational trust.
- Automation could displace repetitive responsibilities faster than employees can move into strategic work, making rapid AI and tool education necessary.
- Scaling before measuring speed and value could leave organizations investing in applications that do not justify deployment.