From "80 separate use cases" to "thinking in processes": this is how you approach AI experiments well

AI is a hot topic, and everyone is looking for ways to make the most of it. This is often quite a challenge, because the possibilities are almost endless. That is precisely why one question is now coming up everywhere: 'How do we prioritize our AI use cases?' 

From faster payment payments to detecting bank fraud, the average organization can easily identify dozens to hundreds of use cases that look interesting. Often a quadrant is created: you compare 80 use cases against their value and impact, choose one or two and get to work. Understandable, but at the same time very time-consuming. 

Moreover, this approach entails a risk: if, after much deliberation, you eventually choose one use case and leave the rest, you throw the added value of all the other use cases over the fence. 

Isn't there a smarter approach to think of? 

One process, multiple use cases 

Take a look at your organization's value chain. Why do you exist and what processes underlie it? If that is the starting point, the AI approach will change drastically. As soon as you start thinking in terms of processes, you realize that one process usually contains several use cases. 

Take the process of resolving IT incidents: you receive a notification, for example: "Employees can no longer log in." You listen to the question, you implement a solution and you sign off the incident. This process consists of several steps. And in every step lies a potential use case. If you add all these use cases together, you have one chain in which you can run multiple use cases.

What if you bundled these use cases and tackled them as one project? In that case, AI suddenly facilitates an entire process. This prevents you from achieving a result that is fantastic in itself, but adds little value for the organization. You realize process improvement. 

Process mining = making data-driven decisions

How do you approach this in concrete terms? At least not by randomly tackling a process without first analyzing it. 

Our advice: start with an understanding of the process. With a technique called process mining, you analyze historical data to make patterns visible. This way you can see what is going well, where improvement is possible and where unnecessary delays or waste occur. More than once it turns out that the bottleneck is not in the expected place.

To stay with the above example: suppose your first hunch is to make the front end of the process more efficient, such as the first response, the classification and the follow-up of a report. However, the analysis may show that the biggest delay occurs later in the process. For example, tickets remain open for too long or are regularly reopened after completion.

Then it makes much more sense to deploy AI agents there. Think of an agent who predicts which tickets are likely to fall outside the agreed resolution time, so that you can make adjustments in time. This is how you achieve concrete gains: a shorter lead time, better compliance with SLA agreements and more tickets that are resolved within 48 hours.

What are your process challenges? 

If you are going to work with AI, do not focus on the best AI tool or use case first. Choose a process that you know can deliver more value. Measure where the biggest challenges are and solve them in a targeted way with AI. You usually take several use cases into account. And that's how you really create value.