“Let’s use Machine Learning for this.”
I was doing a cross-functional client workshop where we were trying to see if we could stack rank of their field force. The field force was involved in multiple programs, each in different geographies. Each program had its own metrics, & data reporting was bad.
As we discussed, we realized that with available data, it would be impossible to get a fair ranking. We started questioning our fundamental assumptions of requiring a ranking scale & realized a single scale would mask individual skills. Three separate scales, each with bands, was better, though it required some basic human intervention.
Many leaders throw problems over the fence at ML experts. There is a tendency to think of ML as magic – give it tough problems & it will spit out business relevant answers.
In practice, we need to start by asking if problems are ‘ML Relevant’? What is the outcome we expect? What is the cost of getting it wrong? Does it fit into our workflow or do we need to modify workflows & roles?
Post this, we can get into data availability & metrics. But without the first step, there are a lot of false hopes & wasted efforts.