Executive Intelligence Podcast - Celonis President Carsten Thoma on Why Enterprise AI Needs Operational Truth Before It Can Deliver

Why It Matters

Without a grounded, cross‑system view, AI agents cannot deliver measurable ROI, forcing CIOs to reconsider vendor lock‑in strategies and prompting broader industry shifts toward data‑centric, process‑aware automation.

Key Takeaways

• Celonis builds transformer models to translate event logs into process language
• Operational digital twins give AI a system‑agnostic view of enterprise processes
• Vendors that lock data limit AI effectiveness across 25‑30 enterprise systems
• AI adoption requires grounding in operational truth before deploying agents
• Thoma urges a new social contract to address AI‑driven workforce changes

Pulse Analysis

Enterprise AI hype has outpaced the reality of complex, heterogeneous IT landscapes. While large language models excel at language and unstructured data, they stumble when faced with fragmented event logs spread across dozens of legacy systems. Celonis tackles this gap by deploying proprietary transformer models that convert low‑level system codes into a unified process language, creating an operational digital twin. This approach provides a holistic, vendor‑agnostic snapshot of process flows, enabling AI agents to target the right tasks, assess human‑in‑the‑loop requirements, and avoid costly mis‑automation.

The shift toward operational truth also reshapes vendor dynamics. Thoma highlights the danger of data lock‑in, where AI providers restrict access to only a subset of an organization’s ecosystem, limiting the AI’s ability to see the full end‑to‑end process that typically spans 25‑30 applications. Celonis’s "free the process" mantra advocates for open data ownership, arguing that the customer—not the vendor—controls the business‑generated data. This perspective aligns with the emerging "SaaS‑pocalypse" sentiment, as CIOs explore building internal AI‑powered solutions to reduce reliance on opaque third‑party agents.

Beyond technology, Thoma warns that AI‑driven efficiency will reshape the workforce and societal expectations. He calls for an early, honest dialogue about job redesign and a renegotiated social contract that balances productivity gains with purpose and belonging for employees. While the United States leads in model development, Europe and Asia possess deep industrial expertise that could accelerate applied AI adoption, provided regulatory frameworks support open data practices. The convergence of operational digital twins, data sovereignty, and societal considerations will define the next wave of enterprise AI, turning it from a speculative buzzword into a measurable business advantage.