Dead Reckoning: No Smooth Sailing for Startups
Prior to the mid 1700s (and long before the advent of GPS receivers in every smartphone), mariners at sea calculated their position by Dead Reckoning, a process in which you simply assume that whatever course and speed you are on can be straight-lined ahead with a ruler and pencil, day after day, regardless of wind, current, and human error. As its rather ominous name suggests, dead reckoning often ended in disaster, with ships hundreds of miles away from where they thought they would be. But while dead reckoning as a trans-global navigation tool fell out of use with the discovery of reasonably accurate ways to measure time (and therefore longitude), the concept is alive and well as applied to virtually every attempt to predict the future.
I suppose it’s only human nature; we got pretty far up the evolutionary chain by remembering what happened the last time and learning how to avoid or repeat the situation depending on the outcome. The last time we ate those awesome looking purple berries we all threw up and old Grunthead died. So let’s not eat them this year. The last time we used chunks of antelope fat on our hooks we caught some huge fat fish, so let’s definitely do that again! We survived because we noticed patterns and learned from them, and the easiest pattern of all is dead reckoning—the straight-line interpolation from the past to the future.
Witness how we predict business outcomes. In 2006, home prices rose by 15 percent, so our best guess for 2007 is another 15. Last year our company grew revenue by 22 percent, so this year we’re going to do 25! The problem, of course, is that these predictions can fly in the face of reason and observable outcomes. The expectation of continuous double–digit exponential growth of any quantity is definitely going to end in tears at some point (though predicting exactly when is worth billions).
For example, take the following graph. We can’t help but see the straight line interpolation showing “trending” growth from 1999 to 2012!
But now look at a simple subset of the above graph from 2001 to 2009, and one tends to see a different picture:
In case you’re curious, the data series from the above graphs is purely random (a random walk to be precise) courtesy of random.org. The fact that we “see” patterns in randomly fluctuating data sets is a human trait and fallibility that gets us into all sorts of trouble. When you start looking for it, you’ll find dead reckoning assumptions built into just about everything we touch, from news reports about the stock market to assumptions about job markets and college majors (e.g. last year there was a tremendous shortage of police officers, so next year the shortage will surely continue).
And the startup world is certainly no exception. The adoption of a new consumer behavior, such as Flash Sales or Daily Deals, tend to be self-limiting, diffusion dynamics where at some point saturation occurs, and the number of people engaging in the behavior plateaus or even declines because the behavior wasn’t worth repeating.
The first part of such adoption curves tends to look something like this:
And of course our natural tendency is to dead reckon that graph straight on up towards infinity. I mean, why wouldn’t we? This is clearly not random—this looks positively awesome, right?
But here’s the rest of this classic S-shaped logistic curve:
Let’s say the above data represents consumer adoption of participating in Flash Sale websites. If you decided to jump into the business at year 8 based only on the seemingly promising first graph, you would have soon discovered, much to your dismay, that the market was already 75 percent saturated and the remaining 25 percent would be largely eaten up in only 2 years! To make thing worse, entrepreneurs are born optimists, so we tend to err hugely on the side of upward growth. I would bet good money that most startup types would extrapolate that first curve sharply upwards even faster than the straight-line interpolation.
It’s not that straight-line predictions are always wrong, but in the same way that naturally clumsy people should be extra careful in china shops, we should recognize our tendency to dead reckon and cast a skeptical eye on linear forecasts. There are natural physical limits to everything, including businesses, and it’s often worth ignoring historical data entirely and trying to get at least a gut feel for what the major dynamics are going to be in the future. Doing a little back of the envelope math on market sizing can go a long way to figuring out if you have some unrealistic implicit forecasting assumptions, such as assuming you will acquire more US customers than actually live in North America!
Clearly predicting the future is extremely hard, and no matter how much math, science, and theory you throw at it, there’s always going to be a huge amount of risk and guesswork. But it’s infinitely better to know when you’re guessing and when you’re not so you can constantly be on the lookout for the nasty rocks that show up exactly where it’s supposed to be smooth sailing.