Don’t throw your human expertise out with the transformational bathwater

Why digital transformations of operational processes require balancing technology with human expertise, knowledge and intuition to be most successful.

Last week I was meeting with a client at their manufacturing facility. It’s a while since I’ve been in a factory so I was excited that after our meeting I was able to do a tour of the shop floor. Doing so reminded me why I studied engineering and started my career in manufacturing.

As I was being shown around, the operations director and I chatted away, though one topic in particular stuck in my mind. Like many manufacturing businesses in the UK, theirs was an aging workforce; a team of people who had built up years of knowledge, experience and expertise in their products, processes and equipment. A team that for the most part carried this critical knowledge in their heads.

Listening to expert intuition

We stopped momentarily whilst my host spoke to one of the team about the latest batch being produced, and was given an explanation of what the team had planned to bring it in line. Their course of action had been determined without AI, ML or even linear regression - it came from human experience and intuition. This didn’t surprise me in the slightest, and brought back a lot of memories from my time as a operations manager and improvement leader in manufacturing. Not least, the importance of listening to the experts in the room.

As with the factory I was visiting, when I started out in my career I was frequently amazed by the depth of knowledge and intuition shown by the shop-floor teams I worked with. Whether it’s the machinist who told me that our new drill bits had the wrong point angle because of how quickly they were wearing out; or the sheet metal worker who identified a mislabelled batch of alloy because of how it reacted on a press brake, the knowledge held within the minds of the teams never ceased to amaze me.

Both of the these examples could have been identified in a more scientific way, and in each case the initial observation was subsequently confirmed through appropriate measurements and tests. But the initial hypotheses were correct, and trusting the team allowed us to react to the issue faster, mitigating its impact whilst we confirmed its cause.

Of course, operational expertise like this isn’t limited to manufacturing. You’ll find examples in every sector wherever people interact with processes, systems and tools. But in our desire to digitise and evolve, do we risk losing this expert knowledge, or at least losing our ability to listen to it?

But AI could spot those things too, right?

I’m a huge advocate for the use of digital technology to improve operations, whether in manufacturing or any other sector. It is quite literally part of my job. The transformational opportunities that can be created through the use of data, automation and AI are incredible, and manufacturing is no exception to this.

For this reason, I also don’t doubt that given the right training data and fed the necessary inputs, today’s algorithms could spot and resolve all three of the operational issues I’ve mentioned so far. But within this there also lies a challenge.

Taking the drill bit wear as an example: spotting that the wear is occurring at an unusually fast rate is one thing, but identifying the cause is another. The abnormal wear rate can be determined through a trend analysis and flagging anything too far out of the ordinary (in 6S you might employ something like an SPC chart to monitor this statistically). But understanding why this is happening requires monitoring and isolating a variety of parameters: the speeds and feeds of the drill, any vibration or undesired motion in the machine, the material properties of the parts, and the properties of the drill bits.

Applying sensors to monitor machine properties is becoming fairly commonplace as part of an asset management or digital twin approach to maintenance, but in this instance it would flag nothing untoward. Unless you had conducted a process FMEA and hypothesised (or retrospectively considered, given past events) that your drill bits may be out of specification, its unlikely that you’d have implemented visual monitoring to confirm their point angle.

Technology as augmentation, not replacement

The result of this, is that even with the latest monitoring and AI in place, we’d likely only be able to spot the occurrence of the problem and not its cause. The operations team would still be required to investigate and resolve it.

And herein lies the crux of the matter. Like any machine, AI will only be capable and assistive in the areas that we allow it to be. It is reasonable to assume that we could develop a model to support with maintaining batch tolerances, to monitor our machines for predictive maintenance regimens or to spot abnormal events. But it is also reasonable to suggest that even when these models are in place, you’ll still need a team of experts to work alongside them. Digital transformation and the adoption of AI doesn’t mean replacing your workforce, it augments it.