Why you can’t maximize both sales and forecast accuracy

When I say that you can’t maximize sales and forecast accuracy, people often are confused or take exception. Here’s why you can’t have both.

Defining forecast accuracy

You make a prediction of total sales sometime early in a forecast period. Then you measure total sales in the period. Accuracy is how close your prediction is to the actual figure. Measured as a decimal, this would be: 

(Prediction - Sales)/Sales. (Multiply this by 100 to get a percent error.)

Maximizing accuracy then means minimizing the (prediction minus sales) "error." When the prediction = sales, the percent error is zero. Note that going over the forecast is still an error. Certainly more desirable than the alternative. But when people talk about forecast accuracy, I hear them bragging about repeatedly nailing the number, not simply exceeding it.

Feedback—the awkward fact

Let’s also recognize an awkward fact. If sales creates (or sees) the bookings forecast, there is feedback between the forecast and bookings. [1] This is your sales superpower. And it's why you have a sales team—because they can change the outcome.

How to maximize accuracy

One way to maximize accuracy (minimize error) is to pick a sales number and stop selling if/when you get there. Pick that number low enough and you are nearly certain to be able to get there. But this leaves sales on the table. You are not maximizing sales. 

You say sales would never do that. Ok.

How to maximize sales

So if you want to maximize sales, you would have to forecast a number that represents the highest achievable sales. Meet that goal and you have maximized sales. Only problem is that you would rarely (if ever) meet this figure. [2] Accuracy would not be repeatable. You have minimized accuracy.

Forecast distributions

A forecast distribution tells the story. It shows the relative likelihoods of different outcomes. If you are closing more than say 20 comparable-sized deals per forecast period, your distribution will look something like this (the amounts of course will be different). [3]

An estimated forecasted sales distribution. The x-axis is the forecasted amount. Higher on the vertical axis means the corresponding amount is more likely. The shaded area is an 80% prediction interval, meaning that 8 out of 10 times, sales will fall in that range.

The chart is annotated with key figures representing forecasts that the business has 90%, 50%, and 10% chances of beating. The chance of landing within a narrow band of any one figure is tiny—unless you stop selling when you get there. The chance of beating the maximum possible figure is by definition zero. The chance of nailing it (with more than a handful of deals in your pipeline) is effectively zero. [4, 5]

You can maximize sales. You can maximize accuracy. But you can’t do both.

Which would your board prefer?

---

[1] If your forecast is completely independent of the sales team, then accuracy is limited by statistical limitations and hyper-accuracy is a myth. 

[2] Of course there are gradations to this approach. You could forecast a more realistic figure—one with for instance only a 10% change of beating. But then if you were fortunate enough to sell that much, you would have to stop selling when you hit that figure (thereby not maximizing sales), and you would only be in a position to do that 1 out of 10 times. You may be close to maximizing sales, but you would not be maximizing accuracy.

[3] Deal-size concentration can make distributions lumpy and skewed. Given enough deals though, the principles discussed here still hold. If you have very few deals then nailing the forecast is a different game. In the extreme, if you have one deal you are working, there are two possible outcomes. You know what the max is and may be able to nail that forecast.

[4] Put that in perspective. If you had 50 deals each with a (very high, unrealistic) 90% probability of closing, the chance of getting them all (the max possible sales) is 0.9^50 or about 0.5% (1 out of 200 times).

[5] What people want, is a realistic estimate of forecasted sales that has a very narrow distribution. For almost all businesses this is not possible, nor (because of the ability of your sales team to change the outcome) is it desirable.