In the first article in this series, I wrote about why we need to experiment with AI carefully. Here, I want to look at AI process improvement: where does AI actually belong? My view is that AI shouldn’t be your system. It should make your system better.

When organisations start with “Where can we use AI?”, they can end up looking for a place to put the technology before they’ve worked out what needs improving. I’d start by looking at the work itself: where people lose time, and what keeps making the same task harder than it needs to be.

Looking at the work this way might indeed lead us to AI. But it might also lead us somewhere simpler.

Where does AI fit into an existing process improvement?

A process can look straightforward on paper, but the person doing the work may have to copy information between systems, chase it from another team, or search for outstanding information. You only see that extra effort when you look at how the process works day-to-day.

Before choosing any tool, we need to understand where the effort is going and why. Is the problem the amount of information someone has to work through? Is there a gap between systems? Each problem needs its own solution.

The outcome is just as important. “Use AI” is difficult to judge as a goal. Reducing the time spent completing a task, making information easier to find, or removing a step that adds no value are things you can measure. They also let the people doing the work see if the change actually improves outcomes.

AI and administrative work in the NHS

A discussion at the Public Sector Data & AI Summit put a live example in front of me.

It was about how many administrative tasks surround patient care, and what happens when that extra work starts taking time away from clinicians.

Keeping an accurate record of care is necessary. The question is how much work sits around doing it, and whether that work could be made easier.

NHS England has recognised that manual administrative work can create delays and interrupt care, and has published guidance on AI-assisted tools for clinical documentation. This is a sensible area to keep improving carefully, as long as any changes help both staff and patients.

To judge whether AI would help, we need to know which steps are really needed, where time is being lost, and what a clinician must check or decide themselves. Only then can we decide whether AI might reduce the workload or just add something else to manage.

When is automation a better choice than AI?

This is a good way to check if we’re really addressing the problem. If a rule is clear and the same action needs to happen every time, conventional automation may do the job well. For staff are copying information between systems, better integration may be the answer. If a step exists only because “that’s how we’ve always done it”, perhaps it can be removed.

AI can help when the work involves information that’s hard to manage with fixed rules. Even then, we need to know where human judgement is needed and how we’ll measure if the tool is making things better. AI is only worth adding if it improves things for the people using it or those getting the service, and guarantees it’s done securely

At Datagraphic, this way of thinking is familiar. We work with processes, data and customer communications, so we ask how information moves and where a workflow could work better. As we explore AI, I believe those questions should continue to come first. Then we can decide what fits best in the process, whether it’s AI, regular automation, or actually changing how the work is done.

That is what I mean when I say AI shouldn’t be your system. It should make your system better. We do not need to build a process around AI just because the technology is available. We need to understand the process well enough to recognise whether AI genuinely earns its place.