Discussion about this post

User's avatar
Ben Wilson's avatar

I think your thoughts on what-if analysis are spot on. Innovation in that area is still happening, even if it is on the fringes of the MDS ecosystem.

I have been tinkering with Summit (https://usesummit.com/)?

It is an event-based simulation tool for building out forecasts. Initial assumptions can be tweaked to model out what-if scenarios.

The company positions itself as more "whiteboard" over spreadsheet (https://summit.readme.io/docs/what-it-is-why).

The app even has input hooks for feeding in metrics and output hooks for bringing the forecasts back into a database (https://summit.readme.io/docs/model-output-api), although it could certainly benefit from a tighter integration.

Expand full comment
Mark T's avatar

Really enjoyed this post!

One piece of the imagined "great scenario analysis tool" I thought was missing was functionality to *easily combine events that have probability distributions*.

I think that the statistical wherewithal to do this is surprisingly rare in our collective analyst's toolkit.

For example - let's say two teams each have "new ARR" forecasts that come with Bear/Base/Bull scenarios, and they both tell you their Bull scenarios have just a 10% chance of being met or exceeded.

What's the probability of an "Overall Bull" scenario? (I.e. that new ARR >= the sum of both teams' Bull scenarios.)

Naively we might expect it's 10%^2, if the plans are independent. But this is not the case, since there are many ways we could get to the same overall new ARR. (E.g. a big overhit in one team, and a miss in the other). Plus in reality, the plans are not independent.

I had this question and put it to an actuary friend, for whom the answer was obvious - this type of question is complicated to solve analytically, but easy to solve with a (stochastic) simulation model.

I.e. we input the Bear/Base/Bull probabilities & ARR impacts for each team's plan, define the "correlation factor" between plans if needed, run N simulations, and receive an output in the form of a probability distribution of overall new ARR exceeding certain values.

So would love to throw "easy simulation modelling in an analytics context" into the mix here :)

Expand full comment
6 more comments...

No posts