Celonis teams with Microsoft on AI agent oversight

Celonis has formed an alliance with Microsoft Agent 365 to help companies move AI agents from trial use into wider deployment.

The partnership combines Microsoft's agent infrastructure with Celonis's process intelligence tools, with a focus on governance, oversight and business measurement for AI systems.

Over the past two years, businesses have tested generative AI and agent-based software across customer service, finance, procurement and internal operations. Many of those efforts have remained limited in scope as companies grapple with fragmented systems, inconsistent data and concerns about control when software agents begin making or recommending decisions across multiple functions.

The joint approach is intended to address that bottleneck by linking agent activity to the business processes in which those agents operate. Celonis argues that process data can show whether an agent is improving outcomes such as cycle times or cost per claim, rather than simply completing assigned tasks.

The alliance highlights tools to monitor the effect of agents on key performance indicators, identify conflicting automated actions between agents, and feed lessons from agent performance back into governance rules. One example is the risk of different agents working at cross purposes, such as one consolidating purchase orders while another separates them to comply with tax rules.

That reflects a broader problem for large organisations as they expand AI programmes. A growing number of agents may operate in parallel across departments and software environments, increasing the risk of duplication, weak oversight and unintended effects unless companies can see how those systems interact across an end-to-end process.

Dan Brown, Chief Product Officer and EVP of Engineering at Celonis, described the challenge.

"This collaboration brings world-class process intelligence to the Microsoft ecosystem, complementing Agent 365's native capabilities to ensure that AI agents don't just perform tasks but deliver measurable business outcomes," Brown said.

He said technology leaders now face a scale problem as companies move beyond isolated experiments.

"The challenge for today's CIO is no longer building an agent; it is operating hundreds and soon thousands of them with trust and safety. IDC projects that AI agent use will increase 'by an order of magnitude over the next few years, yet many organisations remain 'process blind'. They lack the deep operational context required for AI to be effective," said Brown.

"Without a shared understanding of how the business actually runs, AI initiatives often stall in 'pilot purgatory'. Even when agents are performing as they were designed, they often lack visibility into the larger value chain: systems remain fragmented, departments speak different languages, and teams operate in silos. This is 'the great disconnect,' and it is the primary hurdle to achieving a true return on AI (RoAI)," said Brown.

Operational oversight

One part of the integration centres on what Celonis calls Agent Mining, which examines the reasoning and logic behind agent decisions. Software agents produce reasoning traces that can be reviewed to understand why a decision was taken, creating an audit trail that could support both governance and operational refinement.

Under the alliance, users of Microsoft Agent 365 will be able to access those insights to track business impact and return on investment in real time, according to Celonis. That should help organisations assess whether an agent is producing a useful result within a wider workflow rather than in isolation.

Another focus is lifecycle governance. Celonis said its process analysis can identify which agent paths lead to stronger business outcomes, allowing companies to adjust rules and human oversight as agent use expands. The partnership therefore extends beyond deployment to ongoing monitoring and revision.

The alliance also gives Celonis a route deeper into Microsoft's business software base, where demand for AI assistants and agents has grown quickly. For Microsoft, adding process analysis may help answer customer questions about whether AI tools are improving operations in measurable ways, particularly in large companies where automation efforts often cut across enterprise systems.

Early users are beginning to test that proposition. Cosentino, described as an early adopter of the technology, is using Celonis to monitor digital operations and plans to add Microsoft Agent 365 reasoning traces to that work.

"We have already seen incredible value using Celonis to monitor our digital operations. By enriching our existing Agent Mining capabilities with Microsoft Agent 365 reasoning traces, we will move from simply seeing what our agents do to understanding the 'why' behind every autonomous decision," said Rafael Domene, Global CIO at Cosentino.