1
$\begingroup$

Given a clean data set, is there a best-practice way of highlighting likely causes of changes in a KPI -- for instance, examining 10-50 dimensions to find out which contributed most (or was most off trend) to sales declining?

Typically my solution is to pull together all the likely attributes; then chart them based upon % change and magnitude, to see what sticks out; often it's "the West region is down" or "this plan is reducing" ; but the process for finding is surprisingly manual.

As an example, imagine a data set

(salesdate),(store_id),(store_region),(store_type),(planSKU),...

If we note that the total number of sales is down 10% WoW; is there a common way to suggest the likely trends to investigate further?

I don't even know what to call this sort of analysis.

p.s. It gets more complex when you expect sales to be down somewhat; so you want to see which dimensions/attributes are the outliers from the trend

$\endgroup$
1
$\begingroup$

The closest I can get to an answer is that when studying historical data it is often possible to detect "statistically significant" change points . These often suggest the possibility of one time effects or level shift/trend changes. They can also point to changes in model coefficients over time. I would suggest you search for help on Intervention Detection and either get some statistical consulting or be prepared to do a lot of reading.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.