The Category Everyone Suddenly Wants
In July 2026, Celonis acquired Ikigai to give AI agents what it called “operational intelligence.” Weeks earlier, Bespin Global had put the argument plainly: AI competitiveness hinges on operations, not models. The market has arrived at a truth IGX360 was built on. Enterprise AI does not fail on the model. It fails on the operating reality the model is dropped into.
That is the right instinct pointed at the wrong layer.
Operational intelligence, as the category currently defines it, means process mining: the reconstruction of process reality from event logs. Mining is a genuine capability, and it answers a real question. But it answers only one. Mining tells you what happened. It cannot tell you why. And the why is precisely what an AI agent needs before it can act correctly.
What Process Mining Actually Does, and Where It Stops
Process mining reconstructs how a process actually ran by reading the digital exhaust your systems already produce: timestamps, status changes, transaction records, the sequence of clicks across ERP and case-management platforms. It is statistical inference applied to event logs, and it reveals the process as it executed, not as the procedure claimed it would.
This has real value. BPX reports that 60% of S/4HANA migrations stall without this visibility, because teams cannot migrate a process they cannot see. When the documented process and the executed process have quietly diverged over a decade, mining exposes the gap. That is worth having, and dismissing it would be dishonest. Mining is a legitimate first move for any organisation that has lost sight of how its own work runs. You can see how we treat that visibility as an input in our related insights.
But notice what an event log contains. It contains records of actions. It does not contain the reasons for them. A log shows that an invoice over £50,000 was routed to a second approver. It does not record that the threshold is £50,000, why that threshold exists, or what happens to the invoice that arrives at £49,999 with a known-risk supplier attached. The log shows the path taken. It is silent on the criteria that chose the path.
Event logs record clicks, not judgement. They capture the exceptions that were handled in the system, and miss entirely the exceptions resolved in a phone call, a Slack message, or a practitioner’s head. Mining reconstructs the visible trace of a decision. It cannot reconstruct the decision.
Why AI Needs the Why, Not Just the What
Hand an AI agent a mined process and it will do exactly what the data describes: replicate the observed behaviour. That includes the workarounds, the undocumented shortcuts, and the errors that were baked into the trace. The agent has no way to distinguish the correct path from the habitual one, because the event log does not label them. It automates the pattern, at machine speed, without the judgement that made the pattern tolerable when a human was in the loop.
To act correctly, an agent needs the decision logic: the rules, thresholds, and criteria that actually govern the process. It needs to know not just that invoices over £50,000 go to a second approver, but that the rule exists, what it protects against, and which conditions override it. Mining shows the path that was taken. Articulation shows why that path was chosen, and when a different path applies.
This is the gap behind AI’s stubbornly low production success rate. Pilots automate the observed behaviour and are surprised when they inherit its flaws. The problem was never the model. It was that the process was captured as a trace instead of as logic. You cannot delegate judgement you never made explicit. The way we make it explicit is set out in how we articulate processes.
Process Articulation: Human-Legible, Decision-Ready
Process articulation is the deliberate capture of how work is done and why, in a form both people and AI agents can act on. It is not inference from logs. It is a structured, governed model of the operating reality, built to be read by a practitioner and executed by an agent without either losing the meaning.
IGX360’s canonical model makes this concrete. It uses 20 object types and 16 relationship types to encode not just the steps of a process but the decision logic that governs them: the criteria behind a branch, the conditions that trigger an exception, the authority required at a given point, the constraint that must never be breached. Where mining infers a sequence, articulation states the rule. Where a log leaves a decision implicit, the canonical model captures it as a first-class object, legible to a human reviewer and available to an orchestrating agent.
This is the distinction that matters when the market conflates the two. Mining is a valuable input. It tells you where to look and where reality has drifted from intent. Articulation is the destination: the point at which a process becomes AI-ready, migration-ready, and governable, because the reasoning has been made explicit rather than left buried in event data. This is the work that process excellence teams have always done in their heads and in their documentation. Articulation gives it a structure that AI can finally use.
If you want the mechanism before the argument, see how IGX360’s canonical model captures decision logic.
Do Not Automate What You Cannot Explain
The market is right that operations matter more than models. It has not yet noticed that operational intelligence without decision logic is still opaque. A perfect reconstruction of what happened is not the same as an explanation of why, and only the why is safe to hand to an agent.
The process teams who win with AI will be the ones who articulated the why before they automated the what. If you cannot explain a decision, you are not ready to delegate it.
Request a diagnostic and see what your process mining cannot tell you.