Identifying the Right AI Use Cases
A data-driven approach from practice
AI agents can hold context, carry out actions, and increasingly behave like digital team members. But they do not know on their own which problems are worth solving.
In our experience at myos, AI projects rarely fail because of weak models. They fail because of the choice of use cases. Teams start with the processes that intuitively seem most important. In doing so, the actual operational bottlenecks often go undiscovered. In the Map phase of our framework, we regularly uncover potential that the team had not had on its radar before.
Start with evidence, not gut feeling
Every team carries an idea of where the biggest pain points lie. Which tasks cost the most time, which processes constantly stall. When we compare these assessments with the actual data, a different picture almost always emerges.
A few months ago, a client from the automotive industry came to us. His support organization wanted to cut costs, but the first prototypes brought no noticeable improvements. We are trying to automate everything, he told us. But nothing really moves the needle. When we looked more closely, it became clear: the problem was not the AI, but where it was being deployed.
Understanding the actual workload
When we analyzed the conversation data, a different picture emerged than expected. The workload was not broadly distributed, but concentrated and pattern-based.
Two areas, appointment booking and claims intake, accounted for almost three quarters of all support interactions. Everything else ran at the margins. And almost 80 percent of all calls that even touched on the topic of booking ended up becoming a booking journey.
Every organization has such a center of gravity. At myos, this is precisely the core of our Map phase: making the actual workload visible before we propose a single automation.
Finding the levers with the greatest impact
Knowing where requests cluster is not enough on its own. What is decisive is where the human employees actually spend their time. When we overlaid the handling time on the conversation volumes, the leverage points immediately became visible.
This combination of volume and effort produces the clearest signal. At myos, we map this in the impact-effort matrix: quick wins with high impact and low complexity first, then the larger projects.
Implementability determines the starting point
Some processes appeared hundreds of times with nearly the same structure: predictable questions, stable decision logic, consistent transitions. These were not only important processes, but also ones that can actually be automated reliably.
Other topics turned out to be structurally more demanding: inconsistent phrasing, context-dependent reasoning, unclear boundaries. Even when they were relevant to the business, they were not the right starting point.
The best results come from starting with the processes that combine high impact with high implementability. AI automation works best when you tackle the right things first.
