Algorithms can give your company the edge it needs in today’s shift to digitalised services. Already, forward-thinking companies are doing the right thing. They’re introducing algorithms to digitise and speed up their own processes, because the benefits are plentiful.
If we take the example of OEMs, algorithms can help minimise unplanned downtime. They predict when an asset is most likely to break down, and can improve the performance of an asset by adjusting the way it operates. Algorithms can also decide when a certain component should be serviced.
The benefits of algorithms, and in general of building valuable digital services using data, are well-known and well-explained already. And as the CEO of Widget Brain, I’m stating the obvious when I say I’m a firm believer in the power of algorithms.
But many of the articles that write about the benefits of artificial intelligence overlook an important fact. And in order for me to properly explain it, I need to give you some context.
As the authors of the excellent book Prediction Machines rightly say: “prediction is the essential output of AI.” That’s ultimately what we use algorithms for, too. Any form of action an algorithm takes is based on a prediction. The action is a derivation of its own prediction. That’s my first point.
My second point is that algorithms in industries such as OEM and logistics cannot make mistakes. An anomaly prediction algorithm that’s accurate only 80% of the time might do more damage than it does good. It can lead to unexpected downtime of a crucial piece of infrastructure, and unnecessary maintenance is expensive and wasteful.
So the conclusion is that an algorithm needs to be exceptionally accurate in its predictions. And for that to happen, it needs context. It needs to know where it’s located, how external variables (such as the weather, or down- and upstream activities) interact with the asset it’s trained on, and how different data streams interact and correlate with the asset.
All of this means that every machine will need its own set of algorithms.
Big deal, you might think. And indeed, big deal it is. Imagine you have industrial pumps and you’re looking for a way to optimise their performance. There’s much more value in optimising the performance of the pumps than it is to simply monitor their health.
By acquiring insight into the pump’s operation, the business case can move from asset health to better energy management, better equipment usage, and lower asset depreciation, which in turn will become the fundamental drivers for rolling out these algorithms.
In general, I’ve seen that performance optimisation adds more value than just monitoring asset health, because it allows you to improve the underlying business process, instead of focusing solely on the cost of the possible breakdown of assets (most of which rarely break down in the first place).
But performance optimisation requires operational context. This is what most people overlook: each individual pump will require its own algorithm with the operational context of that specific pump. And if you have a thousand pumps, you need a thousand algorithms. They might be similar in nature, but each one will ultimately be implemented in an environment with a unique context.
So don’t be fooled. The amount of algorithms your company has can creep up on you. Suddenly, you could have hundreds or thousands of algorithms that need to be trained and managed individually for accurate predictions. It’s entirely unmanageable, and a systemic problem for the whole company.
The solution is a tool to manage your tools. It might seem like a simple insight, and possibly a bit meta, but you need intelligent software to manage your algorithms efficiently. And if you want to know what that software could look like for your company, don’t hesitate to send me a message.
– Joachim Arts, CEO of Widget Brain