Celonis makes compelling case for 'freeing the process' to operationalize AI returns

While most enterprise software vendors are still promoting AI as a silver bullet, Celonis arrived at its annual Celosphere conference this week with a  more grounded message: enterprise AI is failing to deliver, and the reason has nothing to do with model capabilities.

Speaking to a packed audience in Munich, Celonis Co-CEO Alex Rinke cited a statistic from IDC that is an increasing reality for CIOs making decisions: only 11% of companies are getting any measurable benefit from AI projects today. Rather than gloss over this inconvenient truth, Rinke used it as the basis for Celonis' pitch - that process intelligence provides the missing operational context that AI needs to actually work in the enterprise. Rinke told the audience:

I remember being on a Christmas vacation in 2022, and ChatGPT had just come out. I was on it 24/7. I wasn't just terrorizing my family - I was also terrorizing everybody at Celonis. I was texting people like, 'This is going to change everything.' And I still believe it will and is starting to - but we've also all learned that it's harder than we thought.

That honest assessment set the tone for Celonis' announcements, which focused less on flashy ‘AI will solve everything’ features and more on providing the infrastructure and methodology for making AI actually deliver business outcomes. The company's core argument: enterprises need to "free the process" from legacy system constraints and create a living digital twin of operations before AI can be effectively deployed.

A three-part framework: analyze, design, operate

Rinke outlined three fundamental problems holding back enterprise AI today. First, essential business context is trapped inside individual systems, with critical operational data spread across process designs, enterprise architecture diagrams, log files, and even employee actions that aren't captured anywhere. Second, AI solutions are often deployed based on "the loudest voice in the room" rather than strategic analysis of where they'll actually move the needle. Third, siloed AI solutions don't integrate with existing workflows and technology investments.

The solution, according to Celonis, is a structured approach built on what it calls the Process Intelligence Graph - a semantic layer that creates a digital twin of business operations by extracting and enriching data from across the enterprise. On top of this foundation, the company has built capabilities to analyze, design, and operate AI-driven processes.  Rinke explained:

Our analysis capabilities allow you to discover the process and where to strategically deploy AI. Then you have our design capabilities so that you can redesign your target state based on these insights. And lastly, you can now operate these new improved processes, orchestrating AI solutions directly within your existing workflows.

During a pre-conference press event, Divya Krishnan, Celonis' VP of Product Management, demonstrated how this works in practice using a composite example built from actual customer deployments.

Krishnan's demo centered on a fictional company called Keystone Steel facing flat sales and rising costs. The demonstration walked through how Celonis' approach would tackle this challenge, moving from analysis through redesign to agent-enabled operation.

The analysis phase began with building what Celonis calls a "living digital twin" - unifying data across Keystone's ERP, manufacturing execution system, and crucially, task mining data that captures clicks, spreadsheets, and desktop activity. This comprehensive view revealed that products were being scrapped after successful production runs because they fell slightly below the highest quality standard.

The insight came from the Process Intelligence Graph's ability to provide context across systems and regions. The analysis showed that Denmark had a consistently lower scrap rate - not because of better manufacturing, but because sales representatives were manually reaching out to customers via email to sell outclassed units at a discount. This process wasn't documented in any system, but task mining captured it.  Krishnan said during the demo:

That's the power of analysis with Celonis. Denmark very impressively turned scrap into sales manually, but wouldn't it be even better if it could be done with agent-enabled workflows?

The design phase used Celonis' Process Designer to redesign the workflow with AI assistance, recommending two agents: one to match materials with appropriate customers, and a voice AI agent to follow up if there's no response within 48 hours. Finally, the ‘operate’ phase showed how Celonis' Orchestration Engine (now generally available) coordinates AI agents, people, and systems in one flow - with intelligent triggers automatically starting the process when materials are at risk of being scrapped.

The continuous feedback loop here is worth noting. As Krishnan explained: 

We're continuously monitoring the performance of these agents. These are the new process steps we just created. This isn't historical data. We're looking at every interaction between the voice agent and the customers to assess every bottleneck, every opportunity, and how we can refine this engagement.

This is where I see Celonis differing from other vendors in the market. I’ve been to many conferences this year that have spoken about context being critical to AI deployments, but they don’t take it to the next phase, which is making this context operational. Celonis is showing that it understands the dynamic environment of enterprises, where change is continuous. For AI to become effective, Celonis sees a contextual layer that takes knowledge and redeploys change regularly. 

Making the vision operational

Making this vision reality though requires significant technical infrastructure, and Celonis announced several platform enhancements at Celosphere.

The company's Data Core - its high-performance data infrastructure - is now generally available, with Celonis claiming it provides the scale and low latency needed for operational use cases. Chief Product Officer Dan Brown emphasized that some customers are now refreshing data in less than a minute, enabling real-time operational decision-making rather than historical analysis.

Perhaps more significantly, Celonis announced a partnership with Databricks using Delta Sharing for zero-copy, bi-directional integration. This allows customers to access process intelligence without moving data from their existing lakehouse environment. Brown noted that Snowflake integration is also on the roadmap. Brown explained during the press Q&A:

If people have invested in a data store repository already that we can attach to, obviously, that makes it easier from an enterprise IT perspective. Reusing the asset allows us to move more quickly.

The company also announced enhanced task mining capabilities with AI-driven Task Discovery, allowing Celonis to capture and organize desktop activity data that fills critical gaps in the digital twin. Combined with the ability to integrate enterprise architecture blueprints, unstructured data like PDFs, and data from beyond organizational boundaries through Celonis Networks, the platform is positioning itself to ingest operational context from wherever it exists in the enterprise.

The context question: Celonis' place in the AI stack

During the Q&A session, I asked Celonis executives to address what's becoming an increasingly crowded market position - the "context engine" for AI. Nearly every enterprise software vendor is now claiming to provide essential context for AI systems.

President Carsten Thoma offered a surprisingly measured response, acknowledging that Celonis doesn't claim to provide all context for all types of data. He said:

We will never claim we have all the context. We pull a lot of context also from large data pools. This is why zero-copy integration with Fabric, Databricks, and Snowflake in the future is incredibly important for us - so we don't have to own all of that data.

However, he argued that the specific type of context Celonis provides - operational process intelligence - is uniquely critical for enterprise AI. He added:

If you look at the enterprise AI level, you can live without many other things, but you cannot live without this operational digital twin that we provide. 

If you don't understand what your company is doing, I don't know how you could put AI to work and feel confident about it.

An open ecosystem approach

Critically, Celonis is positioning itself as an open platform rather than trying to own the entire AI agent stack. The company announced support for Model Context Protocol (MCP) servers, allowing the Process Intelligence Graph to be incorporated into third-party agentic AI platforms like Amazon Bedrock and Microsoft Copilot Studio.

Brown acknowledged that the agentic AI landscape will be distributed rather than winner-take-all: 

I don't really see a world where one or a handful of companies 'owns' this. I think Celonis will fit in by being a part of agents that are external to us. I think we will have our own agents. I think we will have agents that are completely inside of our own platform and domain.

The company showcased several partner applications built on the Process Intelligence Graph, including Rollio (process collaboration agents for resolving exceptions in Microsoft Teams), Trullion (AI agents for lease accounting), and Bloomfilter's Agent Miner (which captures not just human activity in software development but also AI agent activity and reasoning).

My take

The contrast between Celonis' messaging at Celosphere and what most enterprise software vendors are saying about AI is very apparent. While competitors are either leading with features or promising that AI will magically fix everything, Celonis is offering something more valuable but less flashy: a blueprint for how AI will actually change enterprise operations.

The analyze-design-operate framework acknowledges that AI deployment requires understanding where to apply it, redesigning processes to accommodate it, and continuously monitoring and improving those AI-enhanced workflows. This isn't the "sprinkle AI on everything" approach that's producing those dismal 11% success rates.

The emphasis on operational context also addresses a real gap in the enterprise AI stack. As companies race to deploy agents, many are discovering that generic AI without business context produces impressive-sounding but ultimately useless results. The Process Intelligence Graph's ability to capture not just system data but also task mining information, process designs, and enterprise architecture creates a semantic layer that helps AI understand how the business actually operates.

The zero-copy integrations with Databricks (and future Snowflake support) are smart moves, allowing Celonis to meet customers where their data already lives rather than forcing yet another data movement and governance headache. Combined with the generally available Orchestration Engine, Celonis is building the infrastructure for the continuous analyze-design-operate cycle it's promoting.

However, building these digital twins is complex work, requiring data engineering and business process expertise. The partner ecosystem will be critical to making this accessible to a broader market. And while the open platform approach is strategically sound - no single vendor will own the agentic AI space - Celonis will need to demonstrate consistent customer ROI as the market matures beyond early adopters. 

Equally, buyers are often making decisions based on what’s already in their enterprise technology stack - hoping that their historic, strategic partners will figure this out for them down the line. However, I’ve been impressed by Celonis’ realistic talking points - being honest with buyers about the changes required is a key part of winning trust. 

In a market full of AI hype and disappointed enterprise buyers, Celonis' message that successful enterprise AI requires operational context, strategic deployment, and continuous improvement feels refreshingly grounded. Whether "freeing the process" becomes as essential to enterprise AI as Celonis claims remains to be seen, but the company is at least asking the right questions about why AI is failing and what's required to fix it.