No AI without PI: Celonis bids to be the context layer powering enterprise AI

Enterprises are awash in AI pilots that never graduate to production. The root cause is a lack of context. AI models can predict, generate, and assist. But without understanding how value actually flows through the enterprise (i.e., orders, invoices, claims, shipments, or tickets), it can’t drive outcomes that matter.

At Celosphere 2025 in Munich, Celonis redrew its line in the sand, from a process mining vendor to a process intelligence platform that gives AI the business context it lacks. This includes the data, rules, and cross-process telemetry needed to connect AI to how the business really runs. For COOs, CFOs, CIOs, and GBS leaders, the business case is straightforward: if only a small percentage of firms can measure any AI benefit today, you need a context layer and an operating model that turns pilots into production. Celonis is bidding to be that layer.

Celonis hardens the platform that turns AI context into enterprise control

Celonis’ platform now coheres around three production grade layers:

• Data Core (GA): High-performance ingestion and zero-copy integration with Microsoft Fabric and Databricks, letting data stay where it lives
• Process Intelligence Graph (PI Graph): A living, system-agnostic digital twin enriched with KPIs, rules, and enterprise-architecture context
• Orchestration Engine (GA): Real-time coordination of agents, humans, and automations

Data Core is generally available and pitched as 20x more powerful than alternatives, while zero‑copy, bi‑directional integrations with Microsoft Fabric and now Databricks leave data where it lives and still feed and be fed by the PI Graph. On top of that, Celonis shipped Orchestration Engine for agents, humans, and automation to coordinate in real time, expanded object‑centric models across domains (lead‑to‑order, TMS, customer service, etc.), strengthened task mining and unstructured input, and added application lifecycle management (ALM) capabilities to control the app lifecycle. For enterprises, this is a pragmatic stack for production AI. Essentially, it could be the structural intelligence fabric that lets them operationalize AI safely and at scale. In essence—context in, orchestration out.

The cost of staying blind to your own processes

The stakes are clear: without process intelligence, enterprises burn through AI budgets, lose control of compliance, and erode trust. Celonis realized $8.8 billion in value across customers, showcased a maturity model that accelerates time‑to‑value, built a center of excellence (CoE), and expanded to domain digital twins and operational orchestration. That technical stance aligned with operating guidance (change maker programs, AI labs) to push outcomes beyond dashboards and into run‑state. HFS estimates its process debt at more than $3.3 trillion, twice its technical debt. Crucially, Celonis’s object‑centric process intelligence is a structural advantage that’s tough to replicate and essential when multiple processes and objects intersect.

Open, composable, and designed to avoid lock-in

Celonis’ new architecture is open by design. Its context layer feeds and is fed by other orchestrators and agent frameworks such as Microsoft Copilot Studio, AWS Bedrock, and LangChain. That openness reduces lock-in risk and positions the firm as a context spine rather than a closed ecosystem. Partners can now build, sell, and monetize apps atop the PI Graph—from process-aware copilots in Microsoft Teams to agent governance modules for software development. This composable model shows what enterprise AI actually looks like beyond the slogans. Examples include a process–collaboration agent embedded in Microsoft Teams for credit‑block handling and Bloomfilter’s process‑aware agent governance for SDLC and beyond.

Customer success stories show PI is your MOAT

Customers such as Vinmar, Mercedes-Benz, Uniper, and Barclays illustrated how the PI Graph fits into the entire enterprise puzzle.

Case 1: Vinmar is treating Celonis as an operating system for order execution and consolidating execution across SAP, TMS, and documentation workflows, reporting a ~40% productivity improvement for supply chain teams. That was a CEO‑level, SG&A‑to‑growth story rather than a tool metric. The CEO also talked about their strategic enterprise stack, which is a combination of Microsoft, SAP, and Celonis.

Case 2: Mercedes‑Benz is expanding from a semiconductor crisis response to enterprise‑wide digital twins and data democratization across the supply chain, production, quality, and aftersales. It is explicitly working on cultural change alongside tech and fusing enterprise architecture into the operational reality led by the Celonis PI platform.

Case 3: Uniper + Microsoft showcased Celonis’s process‑aware copilots that tie SAP, ServiceNow, HR/workforce, and plant systems to cut downtime in hydro plants. This was backed by a workforce already at 76% copilot utilization and a vision of process‑aware agents governed by shared context and rules.

Case 4: Barclays exemplifies compliance at speed, using Data Core and PI to transform and surface case data fast enough to meet five‑minute regulatory response windows, underscoring that value is not around P&L but involves risk, resilience, and trust.

What enterprise leaders must do now

• Treat process intelligence as the foundation for AI. Copilots can’t act responsibly or measurably otherwise.
• Start with one domain digital twin (for example, order-to-cash or logistics). Connect via zero-copy into Fabric or Databricks.
• Codify governance. Define 8–10 guardrail rules (segregation of duties, adherence, compliance) and measure outcomes before and after orchestration.
• Blend deterministic and agentic models. Decide what AI must never decide (sanctions, approvals) and where it can propose and learn (pricing, routing).
• Keep orchestration choices explicit. Use Celonis when it accelerates value but ensure bidirectional APIs so the architecture stays portable.

Scaling this model is as much an operating transformation as a technology one. Done right, you’ll finally break free from pilot purgatory.

The Bottom Line: Treat process intelligence as the context fabric for AI. If you can’t trace a dollar (or a risk) to a process change, your AI isn’t intelligent.

Design target processes that blend automation, AI agents, and human oversight. Clarify roles across platforms, services, and solution delivery and demand bi-directional integration with measurable outcomes. Above all, ensure every change is traceable to financial impact or risk reduction.