From Six Sigma to Process Intelligence

Over the years, I have had the opportunity to work with several large banking, healthcare and other global organizations on process transformation. One thing that has always fascinated me is how the way we understand and improve processes has evolved.

And when I look at where we are today, I see a very natural progression:

Six Sigma → Business Process Management → Process Mining → Task Mining → Process Intelligence → Agentic AI

I believe the next step in this evolution is going to be particularly interesting. Process Intelligence, in my view, will become the foundational layer for enterprise AI agents. Let me explain why.

From Six Sigma to Business Process Management

The origins of modern process improvement can be traced back to manufacturing. Six Sigma, developed at Motorola in the 1980s, brought a structured, data-driven approach to reducing defects and variation. The methodology later expanded significantly across industries, with General Electric making Six Sigma a major enterprise-wide initiative in the 1990s.

The objective was simple: understand the process, identify where things were going wrong, reduce waste and variation, and continuously improve.

As organizations became more complex, however, processes moved far beyond the shop floor. A process could now span finance, procurement, supply chain, HR, customer service and several different systems. This is where Business Process Management became increasingly important.

Organizations started documenting their processes using process models and swimlanes. Teams could visually understand who was performing which activity, how a process moved from one team to another, and what the as-is process looked like. The next step was to design the future state.

I have seen this firsthand while working with clients. Teams would come together in workshops, map the existing process, identify activities that appeared unnecessary, and design a better future-state process. It made perfect sense on paper. But there was one question that remained: once we designed the future state, how would we know whether the organization was actually following it?

That gap eventually opened the door for process mining.

Process Mining: Seeing What Actually Happens

Process mining changed the conversation. Instead of relying entirely on interviews, workshops and process documentation, organizations could now use the data already being generated by their digital systems.

At its simplest, process mining connects events using information such as a case ID, activity and timestamp, along with other business attributes available in the underlying systems. This allows the actual paths taken by individual cases to be reconstructed.

Take a simple Purchase-to-Pay process: Purchase Requisition → Purchase Order → Goods Receipt → Invoice → Payment. That may be the process we designed. But when we look at the actual data, we may discover that cases are taking many different routes. Some skip activities. Some repeat them. Some follow completely different paths.

Suddenly, we are no longer looking at the process we think we have. We are looking at the process we actually have.

This is one of the reasons process mining has become so powerful. It provides a data-driven view of process execution and helps organizations identify deviations, bottlenecks and opportunities for improvement.

I often think of it as an MRI of the process. The process model shows us what the process is supposed to look like. Process mining lets us look inside and see how it is actually behaving.

But even this picture is not complete.

Task Mining: Looking at the Work Behind the Process

A large part of the work performed by employees never appears as a clean event in an enterprise system. An employee may receive an email, open an attachment, move information into Excel, check another application, speak to a colleague, wait for a response and then finally update the system.

The process-mining data may capture only the beginning and the end of that activity, so we know what happened in the system, but not necessarily what happened between those system events. This is where task mining adds another dimension.

If process mining provides the macro view, task mining provides the micro view. It looks at activity at the desktop level and helps organizations understand how people actually perform their work, including the applications they use and the manual steps that may sit between process events.

For example, a bank may have a ten-step account-opening process that is expected to take a certain amount of time. Process mining may show that one particular step is consistently taking longer than expected. Task mining can then help reveal what is happening inside that step: perhaps employees are switching between multiple applications, manually entering information, searching for documents or waiting for another person before they can continue.

Now we have a much clearer picture. Process mining tells us where the problem is. Task mining helps us understand what is happening inside it. And this is where the concept of Process Intelligence starts becoming particularly interesting.

Process Intelligence: Bringing the Pieces Together

Today, we increasingly see the market moving toward what is broadly being called Process Intelligence. The exact definition can vary across vendors, but the underlying idea is straightforward: bring together process management, process mining and task mining.

Process management tells us how the process is designed and how it is expected to work. Process mining tells us how the process is actually executing. Task mining gives us visibility into the work happening at the desktop level. Together, these create a much richer understanding of the process.

This is also where much of the market is heading, with platforms such as Celonis, SAP Signavio and ARIS increasingly bringing process intelligence, process mining, process management and transformation capabilities closer together.

And I believe this evolution has arrived at exactly the right time, because the next major shift is already underway: AI agents.

Why Process Intelligence Matters for AI Agents

We are moving from AI that primarily responds to us toward AI that can increasingly act on our behalf. An enterprise agent may eventually be asked to process an invoice, resolve a customer issue, investigate an exception or execute a series of activities across multiple systems.

But for an agent to do this effectively, intelligence alone is not enough. It needs context. It needs to understand the process: what the intended process looks like, how the process actually operates, what exceptions occur, what rules apply, what activities people perform and what should happen when something goes outside the normal path.

This is where I believe Process Intelligence will become extremely important. Think of Process Intelligence as the contextual layer between the enterprise and its AI agents.

The organization already has years of process knowledge sitting across process repositories, transaction systems, employee activity and operational data. Process Intelligence brings these pieces together, giving the agent a much richer understanding of the environment in which it is expected to operate.

And this is not just a theoretical thought for me. I have seen organizations I have worked with bring together process-mining and task-mining data and build their own contextual or semantic layer on top of it, with the intention of using that information to support internally developed AI agents. Seeing this transition firsthand has strengthened my belief that this is where the market is heading.

My View: Process Intelligence Will Become a Foundation for Agentic AI

This is where I want to make a distinction between what is established today and what I personally believe is coming. I don't believe every AI agent today requires Process Intelligence. But I do believe that enterprise-scale agents operating across complex business processes will increasingly need it.

In the next few years, I expect organizations to realize that simply deploying more agents is not enough. You can deploy one agent, ten agents, or an entire swarm of them, but if those agents do not understand the processes they are operating within, the systems they are interacting with, the exceptions they will encounter and the rules they need to follow, their ability to operate efficiently and safely will be limited.

That is why I believe organizations will increasingly need to take a step back before scaling their agentic ambitions and get the foundations right first:

Build the process repositories.
Understand the process models.
Mine the actual process execution.
Understand the work happening at the task level.
Bring these together into a contextual layer.
Then let the agents operate on top of that foundation.

To me, this feels less like a completely new concept and more like the natural next step in the evolution of process transformation. We started by asking how we could improve a process, then how to model it, then how to see how it actually runs, and then how to see the work happening inside it. Now we are beginning to give machines the ability to act within those processes.

The better the machine understands the process, the better it can act within it. That is why I believe Process Intelligence will become one of the foundational layers of the agentic enterprise. And with companies such as Celonis, SAP and ARIS continuing to invest heavily in this space, I believe the next few years will make this an extremely interesting category to watch.