A challenge that all organizations face is setting goals and creating plans that are internally consistent. For example, if you were to estimate next quarter's revenues by product would it match you estimate for revenues by geography?

What I often find is that there are 6-8 relevant dimensions to estimate a number. When you 'think in ranges', each dimension gives a different range for that particular forecast. By overlaying these forecasts on top of each other, you get a visual depiction of your internal consistency.

If you were perfectly consistent, the forecast would all exactly overlay one another. On the chart, I've overlayed cumulative forecasts. A cumulative forecast shows the likelihood that the true outcome is at or below a given value. I've also added the Median estimate for each forecast.

What we can do is analyze why one method give a different range than another. For example, if our estimate by product is much higher than our estimate by geography, we can drill in and reconcile the two. Through this iterative process, you can harmonize your estimates. As the forecasts get closer and closer, this is what I call the rope. The goal is to have a tight, strong rope.

Finally, I like to look at the dispersion of the means. This give me a 'center of gravity' feeling for where the true number will likely end up. For the true value to be outside of this range, you will have to have been significantly wrong on several dimensions at the same time. If you have to give someone 'the number', the mean of the means is one way to resolve these ranges back to a single number.

Subscribe to:
Post Comments (Atom)

Jim -

ReplyDeleteI would be interesting in seeing you extend these comments to explain the financial market meltdown, where it appears that the quants were wrong on several dimensions at once. It looks like they may have optimized for compensation.

Charles, that's sort of a whole different subject, isn't it? I think it was covered to some degree in a previous post by Jim (the one where he talked about only examining the 3-4 "most likely" scenarios).

ReplyDelete