Outside View

Thinking Tools3.4 Facts About the FutureAlso known as Inside View and Outside View

Developed by Daniel Kahneman and Amos Tversky (Intuitive Prediction: Biases and Corrective Procedures, 1979) and named by Daniel Kahneman and Dan Lovallo (Timid Choices and Bold Forecasts, 1993).

Sets the details of the case aside and asks how similar cases have turned out, anchoring a forecast on that record before adjusting.

A new oven line's first-year sales: the team's forecast, the record of eighteen launches, and the blend

The team forecast $4.0M. Eighteen past launches had a median of $1.8M and a middle half of $1.0M to $2.6M. The last slider is how well launch forecasts have predicted actual sales: at 0 the forecast stays at the class median, at 1 it takes the team's number, and in between it moves that share of the gap. Change the team's forecast or the class and watch where the inside view sits and how the blend follows. Illustrative.

$4.0M
$1.8M
$1.6M
0.3
Gap between the two views
$2.2M
Where the inside view sits in the class
Below the class's middle halfInside the class's middle halfAbove the class's middle halfBeyond almost every past case
Move from the median toward the team
$0.7M
Blended forecast
$2.5M
Middle half of the classBetween the two views$0M$2M$4M$6MClass medianTeam's forecastBlendMiddle half of the classBetween the two views$0M$2M$4M$6MClass medianTeam's forecastBlend

An example

A maker of commercial kitchen equipment was launching a line of compact ovens for cafes. The product team forecast $4.0 million in first-year sales, built from distributor commitments, a trade-show pipeline, and a price under the nearest rival's. Operations needed a number for production.

The finance director set a second forecast beside it. The company had launched eighteen new lines in twelve years; their median first-year sales were $1.8 million, the middle half $1.0 to $2.6 million, and only one had passed $4.0 million. The gap of $2.2 million had two sources: the team knew things about this launch that the record did not, and the record held the slow distributor uptake and rival price cuts that no plan had foreseen.

To weigh them, she checked the same eighteen files for how well launch forecasts had predicted actual sales: a correlation of about 0.3. So she moved three tenths of the way from the median toward the team's number, to about $2.5 million. Production was planned on that, with a trigger to scale up if orders tracked the team's number.

Why it works

The two views draw on different information. The inside view uses what is known about this case: its plan, its people, its early signals. The outside view uses how cases like it turned out, which includes the trouble no plan lists. Each misses what the other sees, so a sound forecast holds both and asks why they differ.

How far to move between them turns on one question: how well has evidence like this case's predicted outcomes before? If such forecasts have tracked results closely, the case deserves most of the weight. If they have barely tracked at all, the forecast belongs near the class median however persuasive the case feels. Moving partway, in proportion to that track record, regresses the inside estimate toward the class and keeps what the case evidence has earned.

The thinking behind it

Daniel Kahneman Daniel Kahneman, a psychologist, showed with Amos Tversky that intuitive predictions follow how well a case resembles a category and so neglect base rates and regression to the mean. In Intuitive Prediction: Biases and Corrective Procedures (1979), in TIMS Studies in the Management Sciences, they proposed correcting such forecasts by starting from the record of similar cases and only then looking at the case itself. Thinking, Fast and Slow (2011) is his popular account of that work.

Amos Tversky Amos Tversky, a cognitive psychologist, wrote On the Psychology of Prediction (1973) and Intuitive Prediction: Biases and Corrective Procedures (1979) with Daniel Kahneman. The 1979 paper drew the distinction at the center of this tool, between singular information about the case at hand and distributional information about similar cases, and recommended building forecasts from the distribution first. That idea was later given its name, the outside view.

Dan Lovallo Dan Lovallo, a professor of business strategy at the University of Sydney, named the inside view and the outside view with Daniel Kahneman in Timid Choices and Bold Forecasts: A Cognitive Perspective on Risk Taking (1993), in Management Science. They argued that planners should anchor estimates on the outcomes of a reference class of similar past projects. Their Delusions of Success (2003), in the Harvard Business Review, carried the argument to executives.

How to run it

  1. Write both numbers down. Record the team's own forecast and its reasons before looking at any record of similar cases.
  2. Get the class's distribution. Gather the outcomes of comparable past cases and note the median and the middle half.
  3. Place the inside forecast in the class. See whether it sits inside the middle half, above or below it, or beyond nearly every past case.
  4. Explain the gap. List what this case has that the class lacked, and what the class went through that the plan leaves out.
  5. Score the evidence's track record. From past forecasts and outcomes, judge from 0 to 1 how well estimates like this one have predicted results.
  6. Blend the two. Start at the class median and move toward the inside forecast by that share of the gap.
  7. Commit, and set a trigger. Plan on the blend, and name the early results that would move it toward either view.

Where it goes wrong

The difference is split by habit. Teams average the two views half and half whatever the record shows; spot it when the weight never changes from one forecast to the next. Set the weight from how such forecasts have fared.

The case is thrown away. The class median becomes the answer even when the case has evidence with a strong record, such as signed orders. Move further toward the case when its evidence has earned it.

The gap is explained away. Every reason the case differs is counted as proof, and the explanation turns out to be the plan restated. Accept a reason only when past cases show it has mattered before.

When to use something else

  • Reference Class Forecasting. When the case offers no estimate with a track record worth weighing, or the job is a project forecast with a reserve: anchor on the class median, adjust only for evidenced differences, and size the contingency from the spread.
  • Premortem. When the plan's specific risks need surfacing, beyond correcting its overall optimism.
  • Superforecasting. When the forecast is the probability of an event, and the class record is only the first of several habits, with frequent small updates.

Where it sits in SOLVE eX

Go to the source

Back to the catalog