Celonis, the process intelligence vendor, today announced the launch of the Celonis Context Model (CCM) and a definitive agreement to acquire Ikigai Labs, an MIT-founded AI decision intelligence company. The deal brings Ikigai’s Large Graphical Model technology – patented MIT research focused on tabular and time-series data, planning, simulation, and forecasting – into the Celonis platform. As part of the agreement, MIT will become a shareholder in Celonis, and Ikigai co-founder Devavrat Shah will join as Chief Scientist, Enterprise AI.

The headline news is of course the launch of the Context Model itself, which Celonis is positioning as a new layer in the enterprise technology stack. The vendor is arguing that AI agents and automation cannot deliver value unless they understand how a business actually runs – and that the operational ontology Celonis has built provides the only neutral, system-agnostic context for them to operate against.

However, what’s also interesting – and what Carsten Thoma, President of Celonis, spent most of our briefing yesterday talking through – is what this represents internally at the vendor. In his words, Celonis has decided to turn its own platform around.

A new approach

I’ve been covering Celonis for years and have written extensively about its process intelligence positioning, its ‘free the process’ argument, and most recently its defence and security offer. The argument that operational context is the missing ingredient for enterprise AI is not new from this company – however, Celonis has changed how it thinks about the platform internally and how it plans to deliver on this context for buyers.

Thoma explained that the company has used the past two years to take everything it has built – process mining, object-centric mining, the Process Intelligence Graph, KPIs, the orchestration engine – and recompose it under a full operational ontology. Process mining, which used to be the product, is now one function consuming the ontology. As he noted, there are hundreds of others. He said:

We need to turn our own platform around. We’re not doing anything different in terms of what we talk about, necessarily, but we are doing something very different when you look at the output.

We come from process mining, as you know. We managed to abstract this to object-centric mining, which allows you an end-to-end view across all systems. We added the platform and then partially the KPIs. But now we really have a full operational ontology…we were able to take what we’ve done so far and mould the objects into the flow – not only into the object view, but actually combining the business context and the objects together.

Thoma argued that the trigger for this was a recognition – two years ago – that the foundations enterprises had been relying on to describe how their businesses work were no longer fit for purpose. He added:

That was so important for us, because we always knew – when we saw everything that was coming two years ago – that legacy documentation is obsolete. You need to have a way, in the agentic frameworks and also with the other applications, to sit at the intersection of the applications, the data, the ontology, and the agentic frameworks. You need something that is able to orchestrate it.

Hands down – where we come from and how we built it – we are, for sure, and in particular with this edition, the only solution that can provide that.

The practical implication for buyers is that the ontology can now feed anything downstream. If a customer wants analytics, process mining draws on the ontology. If they want to instruct an agent, the ontology provides what Thoma described as “the how, the why, and how good” – giving the agent a single place to understand what good looks like. If they want a human workflow, the same context applies.

It’s a meaningful architectural shift and the thing that used to be Celonis is now a feature consuming what Celonis has rebuilt as its core asset.

Where Ikigai fits

The Ikigai acquisition extends and Thoma argued that the operational ontology gives customers the hindsight and insight – what is happening, why it’s happening – but the missing piece is the “what if”. Ikigai’s Large Graphical Model technology is built for simulation, scenario planning, and forecasting against structured and time-series data.

Thoma said:

With the Ikigai acquisition, we’ve now added the ‘what if’. So it’s not only the how and why and how good – it’s also the ‘what if’. Because now you actually have an orchestration question: what do you want to do afterwards? Is it analytics and you instruct someone? Is it a workflow? Is it an agent? Is it a different type of automation? Or is it recomposability?

As I’ve noted previously, a lot of vendor messaging around context still operates at the level of slogans – ‘context for AI’, ‘understanding how your business runs’, ‘grounding agents in reality’. The Celonis pitch, with Ikigai bolted on, is doing something more specific – it allows the platform to simulate process redesigns and outcomes before any change is committed. Thoma also referenced the broader concept of a “world model” for enterprise operational data – a phrase that travels well in current AI discourse, though customers will want to see this translate into deployment evidence before taking it at face value.

Not all context is the same

And of course, Celonis isn’t the only vendor in the market talking about enterprise context for AI. Plenty of vendors are throwing their hat in the ring – SAP, ServiceNow, Google, Databricks, Microsoft, Oracle. So, what makes this different? Thoma’s answer is that most context offerings are about storing and exposing data that already exists, often in unstructured form. Celonis, he argues, is doing the harder thing – generating context from operational data that doesn’t exist in a consumable form in the first place.

He said:

There are a lot of context models out there. But it’s a different type of data. We are not claiming we want to deal with large volumes of historic, unstructured data – that’s exactly what makes the difference. We are domain agnostic, system agnostic, very much focused on operational data – process and productivity – that you have to generate, because the data doesn’t exist in a form you can consume…

It’s not like storing your data, because that’s self-explanatory. The difference is: domain agnostic, system agnostic, full control, in and out – it’s not a one-sided context model.

Thoma pointed to a recent customer benchmark in the HR domain – notably an area Celonis doesn’t consider core – as evidence. He said:

The customer literally wrote us an email saying that in the benchmark, the data context and understanding on domain data for HR that Celonis provided was, by a magnitude, better than the other vendor’s own understanding. And why? Because we know the dependencies. You can qualify the data better if you know the dependencies – when it leaves your system – and the others don’t.

The ‘voice of reason’

The CIOs we at diginomica are hearing from in our network are increasingly cautious about vendor claims as it relates to AI and AI ROI specifically. And of course there is lots of chatter in the market about the so-called ‘SaaSpocalypse’ and the role of SaaS platforms in the future. Commenting on what buyers actually need to move forward in this moment, Thoma said:

It’s not all about the frontier models. I had a CIO summit two weeks ago in Tuscany with twelve leaders from very large global companies. Not one of them is planning a broad frontier model AI rollout in their company in the next three years. Not one.

So what do they need? They need someone who is reasonable. Here’s your context. Here’s how your business runs. Here’s why it’s running badly. Here’s what you can do. And here are your options. One option might be a model. One option might be an automation framework. One option might be agents from ServiceNow. One option might be recomposability. That’s exactly what you need.

I suspect this sort of framing will land well with buyers – not a single sweeping rollout, but a sequence of judgement calls about what to redesign, what to recompose, what to automate, and what to leave alone. Not a sale that’s claiming ‘we will fix everything for you’ – but a platform that helps you to make a variety of different decisions, across a variety of different scenarios, where different tooling and people will be needed. Thoma extended the point to the broader market context:

Where do you go? You need a trusted platform that can show you, with deep intelligence, your operational reality – the simulation. So you can see the hindsight, the current state, the improvements, and the foresight. Absolutely agnostic. That doesn’t charge crazy amounts. In our platform replacements, we probably charge 20 per cent of what the customer was paying before…

That is what the world needs right now: a neutral, transparent, trusted context model. And that’s what we can provide.

And this is where Thoma describes Celonis as “the voice of reason here” – a middle layer that is agnostic, intelligent, and trusted. 

My take

What’s interesting about this announcement is how grounded it feels relative to a lot of enterprise AI vendor messaging. There is no claim that AI fixes everything and the pitch is closer to the opposite: that most AI deployments fail because organizations don’t have a clear, neutral view of their own operational reality, and that without one, the agents and automations on top will not deliver. The Ikigai acquisition adds the ability to simulate alternatives – to model what could happen before committing to change.

Whether the architectural pivot delivers in practice will come down to execution. Celonis is making a substantial bet that the enterprise stack is consolidating around three layers – data, context, and agentic execution – and that it can own the middle. The integrations announced today with Databricks, Microsoft Fabric, AWS, Snowflake (coming), Amazon Bedrock, Anthropic’s Claude Cowork, IBM watsonx Orchestrate, Microsoft Copilot and Agent365, and Oracle OCI Enterprise AI suggest the company is taking that seriously, by being consumable from both sides.

There is much to test as customers adopt this. But after watching this vendor for some time, the argument is likely a compelling proposition for many buyers right now. In a market full of vague context claims and aggressive lock-in plays, “operational truth, neutrally provided, with simulation on top” is at least a tangible thing to evaluate.