And where do you go from there then? If you know that churn is (customers that leave/all customers). It might be that more customers leave or that the overall customer base is smaller (which is the same effectively). But how does that help you go from there?

(edited)

From there you disaggregate the elements of the definition! By logic trees. So you work out the factors that cause customers to leave (e.g., availability of competitor products/substitutes, price sensitivity, is the underlying reason/need for the product/service still there, etc. Once you have done this, you can develop ideas whether/how you can influence each element into the desired direction.

but isn't this dis-aggregation just sort of guessing, as it's not a numerical break down. As you say there could be multiple reasons why customers leave or pay late. In the example case I did it was just because the client didn't charge late repayment fees but competitors did. My issue here is how do you 'sensibly' dis-aggregate the criterion holistically, because the answer may just be hidden in another element that you just don't think of, and hence don't ever get to to the root cause etc..?

No, it is quite the opposite of guessing! It is a 2-step-process: Step 1 is a rigorous disaggregation into numerically quantifiable conceptual drivers, Step 2 is a mapping of qualitative influencing factors/reasons to these numerical drivers. So if you feel concerned that you miss out on such qualitative reasons, then this means that you have not drilled deep enough in Step 1 and your numerical categories are still too broad! In my above example, the "number of customers that leave" is of course too broad and needs further disaggregation. For example you could think of "number of people who actively cancel" + "number of people who just don't renew" (if we are talking about subscriptiom models). You can further drill down if needed, in order to reach the level of concreteness where you don't need to "guess" anymore. I hope this is understandable, but tbh I believe it is very hard to grasp it via exchanging explanatory messages. This is a methodology that needs to be taught.

This makes sense and I understand your point RE numerical issue tree, then qualitative mapping. My main concern which links back to the original question, is for metrics that are not easy to break down numerically at first sight, how do you just come up with the dis-aggregation on the spot without just asking the interviewer to define the metric for you!! This is my issue here, not the qualitative mapping, but how to create the rigorous logic tree on the spot for numerical problems that are not clear cut like revenues or profit!

Quote: " how do you just come up with the dis-aggregation on the spot without just asking the interviewer to define the metric for you!!". But this is what you have to do! You have to align with the interviewer on the definition of what you are solving for! Because this is exactly what you would do with a client. You can not just assume something - this is the recipe for failure... And rightfully so. This is true for all metrics that are not universally defined (like e.g., profit). For example, even a widespread metric like "market share" has to be defined upfront! Market share is mostly measured as "Revenue/Market Size", but in some industries it is different (e.g., market size might be measured by means of customers, volumes, or something totally crazy like "total assets" in Banking). So your duty as consultant/candidate is to ask "What is market size for our client? Am I right in suspecting that it is measured via Revene/Market Size? Or something different?". Only after having aligned on the definition with the interviewer, robust structuring is possible.

Ok clear - very helpful thank you!! :)