AI or Automation?

How to tell the difference (and why it’s important in process improvement)

Lets imagine you’re looking at a process that isn’t working as well as it should. Maybe it’s slower than it ought to be, inconsistent across teams, or dependent on one or two people who know the unwritten rules well enough to keep things moving. The issues are visible, even if they’re not yet fully defined. And at some point, usually fairly quickly these days, the question of whether AI or Automation might help starts to enter the conversation.

What’s interesting is how often we observe that question, AI or Automation, tripping people up. Not because the answer is technically complicated, but because the two are frequently talked about as if they’re the same thing, or as if choosing between them is just a matter of preference. In practice, they address genuinely different kinds of problems. Therefore the risk is that applying one where the other belongs can add real complexity and slow down progress. Even worse, the wrong solution can lock in problems that would have been much easier to address at an earlier stage.

So let’s try to make the distinction a bit clearer in a way that’s useful when you’re standing in front of an actual process and trying to figure out where and how to start. We’ll first split them up and explain each separately using a relatable example, followed by how you can use AI and Automation at the same time to maximise the impact they have when working in tandem. Finally, we will suggest some next steps you can take to apply AI and Automation in practice.

Where AI fits

The moments when clarity starts to break down

First of all, when we say AI, what we really mean are the current Large Language Models (LLMs) which can all interpret information and draw conclusions. The information (data) can be either textual (subjective) or numeric (objective).

AI tends to become most useful at a specific kind of moment in a process: the moment when the next step can no longer be defined purely through clear pre-prepared instructions, but some level of lesser defined exploration is needed to make sense of the data. Up to that point, work may feel relatively structured. Almost like a formula or programmed set of steps, which follow each other in a standardised and predictable sequence. Information moves through the process, and although there may be inefficiencies, there’s still a clear and shared understanding of what should happen next.

Then something shifts. A handover starts to depend on who is working that day, a decision takes longer than it should, or two experienced people look at the same information and reach slightly different conclusions. These are the moments where interpretation has become the real work, and where maintaining consistency through manual judgement, at any kind of scale, starts to become difficult to sustain. Over time, that difficulty tends to slow things down or introduce exactly the kind of variation that’s hard to get under control through standardisation alone.

AI provides a way of navigating that variation. Rather than relying on a fixed set of rules, it works with patterns in data to support a judgement about what’s likely to be appropriate in a given situation, whether that’s classifying information, flagging risk, summarising content, or suggesting a course of action based on what’s happened in similar cases before.

For improvement practitioners, this feels like familiar territory. PDSA is carried out all the time by experienced people who know the process and its systemic environment inside out, instinctively knowing how best to proceed.  But if you’re new, you clearly have to develop that expertise, and AI can support PDSA in action by exploring theories, ‘what if’ scenarios, and depending on the outcome, help you develop new theories and insights. It can be your friend in accelerating your understanding and taking meaningful next steps.

One caveat is worth adding here. What we see is that there’s a natural pull towards reaching immediately for AI wherever post analysis judgement is involved, but in many cases the real issue sits slightly earlier in the process of interest. Variation has already crept in: inputs may be inconsistent, criteria may not be fully agreed, or different teams may be applying their own unwritten assumptions without realising it. Introducing AI into that situation can produce outputs that seem useful in isolation but don’t hold together consistently when used more broadly. AI works well where there’s still a degree of structure in place. The process doesn’t need to be rigid, but it does need enough shape that inputs are understood and outcomes are broadly aligned.

In summary, AI works best when the process of interest is well understood, defined and standardised so we ensure and are assured that the analysis is meaningful for insights and decision making.

Where Automation fits

When the work is understood but execution is the problem

If AI tends to appear at the points where clarity of next steps is unclear or the complexity of the data is limiting instinctive insights and analysis, automation sits more naturally at the front end of the spectrum, where the work is already well understood, repeatable, and defined, and the real challenge lies in executing it consistently.

In most organisations, there are large parts of processes that follow a stable, predictable pattern, even if they are currently handled manually. The steps are known, the sequence is clear, and the outcome is expected to be the same each time. What those processes need is reliable execution, without unnecessary effort, and without the variation that naturally creeps in when work is done by different people under different conditions.

Automation addresses exactly that. By defining the steps and triggering them in response to specific events, you remove the need for repeated manual handling, work moves more smoothly, delays are reduced, and the process becomes much easier to manage and scale. This in turn frees up human capacity to focus on those issues which automation cannot deal with (special events), or to focus on the further improvement of that process.

Worth flagging here is that automation tends to make the quality of the underlying process design quite visible, and quite quickly. Firstly, where there are gaps, inconsistencies, or steps that exist more out of habit than necessity, those tend to surface once a process is running continuously. Secondly, automating a poor process can create even more rework and at a faster pace, which could also be hidden as it’s now done with minimal human intervention.

Automation doesn’t fix a poorly designed process; it either magnifies or hides the problems. But, when analysing the output, it can give you a much clearer view of where improvement work is still needed. Regular review of process performance data is key here.

The difference in practice

The distinction tends to become much clearer when you step back from definitions and look at how work actually moves through a process over time. In most cases, AI and Automation aren’t competing for the same role, even though they’re often discussed as if they are. They tend to sit at different points along the same flow of work, becoming relevant under different conditions rather than solving the same problem in different ways.

• Where a process is already well understood and the next step follows naturally from the previous one, the challenge usually centres on execution: keeping things moving reliably, without introducing variation simply because different people are involved. That’s where automation starts to add value.
• The points at which AI becomes useful tend to look quite different when you observe them closely. These are the parts of the process where the next step is less obvious, where the outcome depends on context, or where there is just enough ambiguity that progress slows while someone decides what to do. Over time, those moments of hesitation or inconsistency begin to accumulate and affect the overall flow.

Seen in that light, the distinction doesn’t sit in how advanced the technology is, but in the nature of the problem being addressed. One is concerned with keeping work moving consistently; the other is concerned with helping it move forward when clarity is less certain.

Identifying where each applies

When you start, firstly consider what what you’re trying to accomplish. Is it making a well defined process faster with less human intervention, or is it identifying a way to support decision making?

Some activities can be mapped out clearly from beginning to end, even if they’re currently inefficient or manually intensive. The steps are known, the order is predictable, and the outcome is expected to be the same each time. In those situations, improvement tends to come from making that sequence more reliable and less dependent on human intervention at every stage.

Other parts of the same process may behave quite differently. The input varies, the criteria aren’t always explicit, and the outcome depends on how something is interpreted at that particular moment. These areas often sit slightly outside the reach of straightforward standardisation, even when genuine attempts have been made to define them more precisely.

What becomes more useful, therefore, is recognising that both characteristics can exist within the same flow of work, rather than trying to label an entire process as suited to one approach or the other. The task is less about choosing between AI and Automation and more about understanding where each naturally fits within the work as it actually happens.

That recognition doesn’t always come immediately. In our experience, it tends to only become visible once the process has been examined in enough detail to see where it consistently behaves one way, and where it begins to vary. That’s often where the most meaningful design decisions end up being made.

How AI and automation come together

In the last 5 years we’ve seen a revolution in processing data, where we can now harness the best of both worlds.

When AI or Automation are applied in isolation, the benefits tend to be limited to specific parts of a process. When they’re combined in a way that reflects how work really flows, the impact becomes more noticeable. The technology itself hasn’t changed; what has changed is how well it’s aligned with the underlying structure of the work.

In practice, this usually involves allowing AI to operate at the points where interpretation is required, while relying on automation to carry the process forward once a direction has been established. One introduces a level of consistency to decision-making, while the other ensures that the resulting actions are executed in a predictable and repeatable way.

Why this is important in process improvement

As these tools become more widely accessible, the nature of improvement work is starting to extend beyond refining individual tasks and into shaping how entire processes behave over time. The focus is moving towards understanding how work flows, where it begins to break down, and how different capabilities can be applied in a way that reflects that reality.

In that context, the decision to introduce AI or automation is rarely a technical one in isolation. It sits alongside questions about process design, consistency, ownership, and the degree to which a current way of working is stable enough to build on. The role of the improvement practitioner becomes more about deciding where intervention will be most effective, taking into account both how the process is intended to work and how it actually behaves in practice.

That’s a more interesting challenge than it might initially sound, and it’s exactly the kind of challenge that improvement practitioners are already well placed to take on, provided they’ve given themselves the time to properly understand what these tools do, and where each one genuinely earns its place.