Among the many interesting points from my interview with Dan Brown, Chief Product Officer at Celonis, was his forward-looking view of how agentic AI systems will reshape enterprise operations in the coming years. AI agents will drive analysis, design, and execution faster and with greater synchronization than ever before. In sum, this emerging technology will support a major shift in business workflow.
Additionally, Brown—who spoke to me at Celosphere 2025, in Munich—detailed how enterprises are using digital twins, semantic data models, and autonomous agents to uncover bottlenecks, govern complex workflows, and measure real ROI from AI-driven change.
“Agents will soon be analyzing, designing, and operating business systems in real time — accelerating decision cycles and reshaping the digital operating model of every major enterprise.”
Core Takeaways
-
- Process mining: Provides a real “X-ray” of how work actually moves through enterprise systems, exposing variations, bottlenecks, and unnecessary manual touch points.
-
- Digital twins: Creates a unified semantic model across front-, middle-, and back-office systems, enabling truly end-to-end visibility and cross-process optimization.
-
- Agentic AI: Delivers scalable automation and decision-making, though companies need to be aware of issues around governance, transparency, and KPI-based accountability.
-
- Generative AI acceleration: Dramatically speeds up the creation of digital twins, ETL pipelines, and semantic models, reducing setup time and amplifying enterprise ROI.
Key Quotes
The Real Value of Process Mining
“What process mining does is it looks at all of these systems and these digital footprints of documents that are moving through and that work that’s happening, and we take that data and we use inference algorithms to actually show you exactly how the process works. Once you have that X-ray, you can immediately understand the distribution of all variations—where the highest-percentage flows are, where bottlenecks appear, and where unexpected human touches are slowing things down.”
“From there, enterprises can analyze, design, and operate differently. You take the X-ray, see that the real process is not what you expected, modify it through optimization or engineering, and then run and monitor it. That closed-loop cycle of visibility and improvement is the foundation for every transformation initiative that follows.”
Why the Digital Twin Matters
“A digital twin fuels everything: discovery, modeling, and operations. We take data from source systems, format it into objects and events, and infer the business process as each object moves through its lifecycle. Because enterprise processes intersect constantly—front office with middle office, middle office with back office—that unified twin becomes the only reliable way to see how work really moves across an organization.”
“Many enterprise challenges come not from one system’s behavior but from mismatches between systems. A CRM’s idea of an order may not match the ERP’s or the finance system’s. The Process Intelligence Graph standardizes those shapes and meanings so companies can finally get true end-to-end process intelligence, not just fragmented data views.”
Governing Agents and Measuring ROI
“Most companies feel more comfortable when agents provide recommendations that humans validate—but as agents begin taking direct action in systems of record, governance becomes critical. With Celonis, we can see system updates driven by an agent, and we can also use ‘agent mining’ to observe when an agent is working and how it impacts KPIs. You get two independent signals validating that an agent acted and what the result was.”
“When you tie agent behavior directly to process KPIs, you can measure whether a solution is improving speed, cost, quality, or compliance. That’s the essence of real ROI on AI investment—seeing if the KPI is going up, going down, or not moving at all. Enterprises need that visibility before they hand more decision-making authority to autonomous agents.”
The Future of Analyze–Design–Operate
“Agents will soon run other agents. You’ll have agents sensing the environment, reasoning about what they detect, planning the next steps, executing, and learning from outcomes—and orchestrating other agents to perform specific tasks. As precision and trust increase, we’ll see many new agentic patterns: agents invoking code, agents calling RPA, agents orchestrating other agents in dynamic workflows.”
“This will massively accelerate the analyze–design–operate loop. Instead of organizations pausing to assess before making changes, agents will continuously analyze end-to-end processes, reorient the enterprise, execute adjustments, and measure outcomes. That speed and elasticity—especially during peak loads—will profoundly change digital operating models and transform the way businesses run.”
