Wednesday, June 10, 2009

McKinsey: Obfuscation through Excessive Analysis

One common characterization of McKinsey work is that they will do all sorts of ridiculous levels of analysis, even if the outcome isn't cogent to the actual case. A harsher view would say that McKinsey hangs audacious ideas onto that analysis to push the client into pursuing their recommendations, even if they won't work in reality (i.e. "Ted Airlines").


Consider the article that fell into my reader today:


Consultant Ninja Analyis: Power curves have interesting properties that I remember from undergraduate circuits class. McKinsey's going to use an interesting mathematical formula to tie economics and nature together. Great, let's see what they do!



Consultant Ninja Raction: All this lady has done is sort a few statistics according to size. Once you sort a population by size, it almost inevitably follows a pareto distribution. And a pareto distribution can almost always be fitted by a power curve function.

If this consultant wrote "bank crises by size show us that you should focus on 'Build flexible business models'" she would be challenged by the client. If she instead writes "a power curve analysis with an R-squard of 0.96 shows us that bank crises are dynamically unstable. Therefore you should "Make the system the unit of analysis'", the client, who can't possibly understand her 2 weeks of work in a 60-minute meeting, would blandly nod his head.

Bottom Line: A common (and fair) criticism of consultants it that we whip up terrificly complex analysis for the purposes of intellectually bullying clients into just blindly going along with the larger message we construct. McKinsey is guilty of it here.

PS. What the hell is the Y-axis on the banking chart, anyways?

5 comments:

Anonymous said...

let me help you out my friend. The y axis is most likely the number of crises within the ranges in the x axis (i.e., nearly 70 crises had less than 15% GDP loss).

Also, it's not true that "Once you sort a population by size, it almost inevitably follows a pareto distribution." Try doing it with height of adults and you'll see that the fit will be terrible.

The type of frequency/magnitude analysis is the chart is used often by physicists and geologists to understand complex system dynamics, so it's pretty established. I think it's good to apply that type of thinking to business since there are some similarities

Consultant Insider said...

I am in agreement with Ninja here - All this graph tells me is that, like earthquakes, there are more smaller banking crises than larger ones. I don't really see a useful "so what" coming out of this.

astro said...

I'm not sure how this has a practical application yet. The power law feature is pretty common to many natural systems. Put another way, you can make most any data fit if you can vary the power law index arbitrarily...

One interesting possibility might be related to the idea of fractal dimension though. See the NOVA special on this. In that case, one prime example was that a population of trees showed a similar fractal dimension to the number of branches on a single tree. And using that empirical relation, by measuring a single tree (or a few), the biomass or oxygen production of a whole forest could be accurately estimated, thus saving the work of exhaustive measurement.

But I'm not seeing how this particular idea would be used in a business context... It would be interesting though!

Consultant Ninja said...

Anonymous- of course heights don't fit; they follow a Gaussian distribution. What I should have said is that populations in business applications subejct to analysis are almost inevitably pareto distributions, not gaussian (like the CDO modelers discovered, too late unfortunately).

The reason is that in a Gaussian curve you can take the average and be ok. In a pareto distribution, if you just take the average you missing huge incremental opportunities on the margins.

Pareto have huge value in business - that goes without saying. My criticism is that this article uses fancy terms to pretend to invent an insight that Pareto coined 100 years ago.

Liz said...

I think application and understanding of power laws in a business context is still rare though.

For example the mean in a power curve situation is almost useless as it is completely skewed by the largest (ie least common) value.

But look how common it is in business to use means as inputs for forecasting etc

Better understanding of power curve distributions and where they occur could avoid some catastrophic decisions I suspect.

Wednesday, June 10, 2009