Efficiency gains with AI

Where are the opportunities? 

You can make almost any process more efficient using AI, says Maartje. “The trick is not to be distracted by the possibilities of the technology, but by understanding your greatest challenges and how AI can be part of the solution. The solutions are often in very simple things. Employees on the shop floor often know exactly where the bottlenecks are. So start by spending a few days on the shop floor. Try to understand the work processes and ask employees about their challenges.” 

AI can often help in tracing bottlenecks. One of the instruments is text analytics, in which you use NLP (Natural Language Processing) technologies to analyse text. For example, you can analyse the reasons why people contact a call centre: what are the most common questions, or which process causes the most complaints? Maartje: “There have been several occasions when, by analysing customer phone calls or emails, we found out about problems that could be solved within a few hours. For one client, we were able to prevent 23 per cent of incoming phone calls by providing better information on the website.

You can also analyse operational data with AI. Take a company that manages parking garages and buildings, for example. They can collect real-time data from all the devices that are used such as barriers, pay machines, entrance gates, et cetera. Based on that data, you can detect deviations from business as usual. For example, on Saturday afternoons, 80 per cent fewer payments are suddenly recorded in a downtown parking garage. The Operations Control Centre is then alerted immediately to check what is going on and possibly send a technician, even before customers report a problem,” says Maartje. While IT professionals often only think of their own IT operation when they think of AIOps, Gartner uses the term for AI solutions that analyse any kind of operational data. It works best when combining it with real-time decision making.  

What is your process? 

Using the example of an online pharmacy, Maartje outlines how Mediaan Conclusion goes about identifying and removing bottlenecks with AI. We started by understanding their biggest challenges. That is quality control of packages. After all, you must never send the wrong medication to a patient. It was a very labour-intensive process, which Mediaan Conclusion supports with computer vision.

The contents of each box are photographed from above. This photo is analysed by an image recognition algorithm within seconds. Maartje: “We started with a Proof of Concept lasting two months. Within a short time, we were able to analyse 76 per cent of packages automatically. This gave us and the client enough confidence that it could work. We then drew up a roadmap and invested in a good camera setup. We also developed a scalable cloud platform that can analyse multiple photos simultaneously.

Finally, more data has been added to the machine learning model such as knowledge from supply chain experts combined with information about things like weight and historical checks. We can now analyse 94 per cent of packages automatically and only 6 per cent need to be checked by hand.” 

A second improvement involves checking the prescriptions and processing them correctly. If you want to send medication within a day, this process has to be quick, so any time savings are invaluable. Maartje: “Here, we try to capture pharmacists’ knowledge in a machine learning model. We do this by recording the decisions pharmacists make. We use the data to train a machine learning model. Multiple iterations are then done with experts to understand why the model makes errors and what the logic behind them is. In this way, we can improve the model step by step.

The pharmacists also help us understand the drugs with which there is absolutely no room for error, so we can make sure to be extra careful with them. For example, we do an extra manual check on all new drugs and on certain categories with very high risks.” All prescriptions where there is no doubt now go to order picking automatically. And prescriptions where the model says something is wrong, or where the model has doubts, are reviewed by a pharmacist.  

The pharmacists not only check prescriptions, but also assist customers who call the contact centre with questions about the medication. Mediaan Conclusion used ChatGPT’s language model, questions that pharmacists have answered in the past, and the data in the company’s knowledge base to train an NLP model that prepares the answer for the pharmacist in advance, making their work much more efficient. This is exactly the same way that pension administrator APG operates. 

Copy ideas from other sectors 

Maartje emphasises the importance of taking inspiration from other domains when you want to innovate. “Many companies think that innovation means they have to do substantively new things. No, innovation is doing things that are new to you, but which may have been used somewhere else for much longer. There are many smart people in the world who have already developed solutions. Some of these products are sold under licence, but many are also available for free as open source. We keep an eye on what is coming onto the market for our customers, so they can take advantage of technology developed by others quickly. The way I see it, that really is smart innovation.”