Inversion Thinking

The problem it solves

Asked how to make a plan work, most groups produce a short list of general intentions. Communicate more, start earlier, involve the right people. Asked what would make the same plan fail, the same people in the same room produce a longer list of specific things, and most of those have an owner, a system or a date attached to them.

The asymmetry is reliable and it has very little to do with optimism. Success has roughly one shape, whereas there are a great many ways of missing it, so the failure question searches a much larger space. It also searches it with the part of memory that stores what went wrong last time, which is the part carrying the detail.

What it costs to ask the forward question by itself is a plan whose risks are all phrased as adjectives. Ambitious timeline, tight dependencies, stretched team. None of those can be removed, assigned or watched, because none of them names an event.

The idea

Inversion means stating the outcome you want, asking what would reliably produce its opposite, and then treating the answers as a work list rather than as a mood. Charlie Munger borrowed the instruction from the mathematician Carl Jacobi and spent forty years repeating it.

Invert, always invert.

The version with a procedure attached is Gary Klein's premortem, which adds one move that turns out to matter more than the inversion itself. It puts the question in the past tense. The project has failed, it is a year from now, and the room is explaining why. Explaining something that has happened is an easier mental job than predicting it, and people who would hedge a forecast will describe a fact in detail.

Enlarge: Same plan, same room, two questions, two very different lists.

Two columns side by side. The left is headed ask it forward, with the question how do we succeed, and beneath it two long bars standing for the answers that come back, noted as few and general. The right is headed ask it inverted, with the question what would sink it, and beneath it five shorter bars of differing lengths, noted as more, and each one specific.

Same plan, same room, two questions, two very different lists.Close

Same plan, same room, two questions, two very different lists.

Safety-critical industries have run on this for decades, where the planning question is which sequence of events ends with somebody hurt, and every control is written against a sequence rather than against a general goal of being safe. It transfers less cleanly than it looks, because those industries search half a century of recorded, investigated failures and a project team searches the memories of whoever showed up.

How to use it

Run it once there's a real plan and before there's a public commitment, which is the short window where the detail exists and changing things is still cheap.

Say the outcome in the past tense and be specific about when. The launch happened, it's six months later, and it went badly enough that people are talking about it.

Have everyone write independently for five minutes before anybody speaks. This is the step that gets dropped and it's the one doing the work, because the first cause said out loud sets the direction for the causes after it.

Then convert, item by item, in the room. Each cause becomes a thing somebody removes this week, a thing somebody watches with a named trigger and a named watcher, or a thing the group accepts out loud and writes down as accepted. A cause that becomes none of the three was worry rather than analysis, and saying so is kinder than leaving it on a list.

Where it breaks down

Every concept here has one. It is the section most summaries leave out.

It searches experience, and it stops silently at the edge of it. The list comes out of what the people present have lived through, so a failure mode nobody in the room has met is invisible, and the exercise produces exactly the same feeling of thoroughness either way. The only repair is a different person in the room, rather than a better-run session.

Assigned pessimism is weak pessimism. A person told to find the problems produces a performance, and everyone present discounts it accordingly, including the person performing. The research comparing real disagreement with the role-played sort is unkind to the second, which is awkward for a technique frequently run by appointing somebody to be the doubter.

It converts easily into a document nobody owns. The generative half is cheap and satisfying, so the meeting ends with eleven causes and a warm sense of having been rigorous, and the conversion step is what gets cut when the hour runs out.

The evidence is thinner than the popularity suggests. One study from 1989 on explaining events in the past tense is most of the foundation, the effect it found was a moderate increase in the reasons people generated, and nobody has shown that teams running premortems deliver better projects. The mechanism is sound and the claim usually made for it is larger than the mechanism supports.

Aimed at a person it curdles. What would make this hire fail is a reasonable question about a role and an ugly one about a candidate, because the answers come back as attributes rather than as causes anybody can remove.

In one line

Ask what would make this fail, in detail and in the past tense, then go and remove the answers.