Celonis Thinks Enterprise AI Has a Context Problem

A commonly cited report by MIT's NANDA initiative, which studied over 300 deployments and surveyed 350 employees, found that 95% of enterprise AI pilots deliver zero measurable business impact. The same analysis cited S&P Global data showing 42% of companies scrapped most of their AI initiatives in 2025, up from 17% the year before.

Carsten Thoma, Celonis PresidentCelonis, the process intelligence company, says it knows why this is happening. On May 12, the company launched the Celonis Context Model (CCM) and announced a definitive agreement to acquire Ikigai Labs, an AI decision intelligence company built on nearly two decades of MIT research. Together, the two moves mark Celonis' most significant repositioning since it became, by analyst measure, the market leader in process mining.

"AI is only as good as the context it has," says Carsten Thoma, Celonis President. "Every organization needs to give its Enterprise AI a holistic, living model of how a business truly operates."

Why AI Agents Fail Inside Real Companies

The argument Celonis makes is architectural. In its telling, enterprise AI agents fail because they lack an accurate, real-time understanding of how a business actually operates. ERP systems record transactions and large language models generate language, but neither captures how work flows between systems, why processes break, or what the exceptions are.

The CCM is Celonis' proposed fix. The company describes it as a dynamic, real-time digital twin of operations, built from process data across every system, application, device, and interaction inside a business, designed to give AI agents the operational clarity they need to act reliably.

The reliability problem the CCM targets is well-documented. A 2026 paper examining AI agent reliability in enterprise contexts found that even a model with 99% per-step accuracy achieves only 36.6% success across a 100-step workflow. For enterprise applications processing millions of transactions annually, the margin for error is effectively zero.

Jerome Revish, SVP and Chief Technology Officer of Digital and Technology Services at Cardinal Health, puts it plainly. "Precision is paramount in the healthcare industry, and you can't accept AI that's only right most of the time," Revish says. "Ultimately, context is what makes the difference between AI that's impressive in a demo and AI that's trusted and safe to deploy."

From Process Mining to Forecasting and Simulation

The Ikigai acquisition gives the CCM its forward-looking dimension. Celonis describes its existing capabilities as providing hindsight (how operations ran) and insight (why processes are performing as they are). Ikigai adds foresight: the ability to model what will happen, and to simulate what should happen next.

Ikigai Labs was co-founded in 2019 by Devavrat Shah, a Chaired Professor of AI at MIT, drawing on two decades of his research in graphical model learning. Shah previously co-founded Celect, a retail inventory optimization company acquired by Nike in 2019. Ikigai's core technology is the Large Graphical Model (LGM), designed specifically for structured enterprise data: tabular and time-series data found in spreadsheets, ERP databases, and enterprise platforms. Where large language models are built for unstructured text, LGMs are built for the rows, columns, and timestamps that make up most enterprise operational data.

In practice, Ikigai has used LGMs to reduce supply chain planning and forecasting cycles from months to minutes for enterprise customers, with additional applications in financial reconciliation and fraud detection.

As part of the deal, Celonis gains exclusive rights to MIT-owned patents that Ikigai had licensed. MIT itself becomes a shareholder in Celonis, an unusual arrangement that signals the depth of the research relationship. Shah joins as Chief Scientist of Enterprise AI.

"Together, we now provide the fullest operational representation of business reality," Shah says. "With the Context Model, AI agents have the hindsight, insight, and foresight to intelligently and continuously adapt."

Celonis Enters a Fight It Did Not Start

The enterprise AI execution layer is already contested. Salesforce, Microsoft, ServiceNow, SAP, Google, Adobe, and dozens of third-party vendors are all competing to become the runtime control plane for enterprise AI. The vendor that wins, as Futurum Research notes, effectively sets the terms for how enterprise AI operates across the entire stack.

The competition is moving fast. At SAP Sapphire 2026 this week, SAP and Microsoft made their most ambitious joint push yet, with SAP Joule Studio now generally available and a Copilot-Joule integration designed to orchestrate tasks across finance, HR, and procurement. ServiceNow launched its own Context Engine in April 2026 and, at Knowledge 2026 last week, unveiled an Autonomous Workforce suite aimed at completing entire business processes without human intervention.

The word "context" is now standard vocabulary across all of them. That shared language narrows the distance between Celonis' pitch and the field, and means the company will need deployment evidence, not just architecture, to stand apart.

Celonis' core claim is that its process intelligence runs deeper than what rivals can offer from data infrastructure or application workflows. It captures how work flows between systems: the process layer that ERP transaction logs and data lakes do not record. As Diginomica noted, "the thing that used to be Celonis is now a feature consuming what Celonis has rebuilt as its core asset." The CCM also connects to Microsoft Copilot, Oracle OCI Enterprise AI, Amazon Bedrock, and IBM watsonx Orchestrate, positioning Celonis as a neutral layer rather than a walled platform.

Celonis has held the top position in process mining for six consecutive years on the Everest Group PEAK Matrix, with the highest reported market share across manufacturing, banking and financial services, healthcare, and retail. That installed base gives the CCM a foundation that no new entrant can replicate quickly.