Find The Swing Variables

14 May

Most strategic conversations begin too late.

A group gathers around a set of options and quickly starts debating which one is right. People bring evidence for the option they favour. Others point out risks. Someone asks for more analysis. The discussion becomes a contest between recommendations.

That feels like decision-making. How strategic is this? What is strategy anyway?

In Playing to Win, A.G. Lafley and Roger Martin describe a different discipline. Rather than begin by arguing over what is true, they suggest first asking what would have to be true for each option to be a good choice. That is not merely a better meeting technique. It is a different way of thinking.

It is also very close to Bayesian reasoning.

Two very different thinking modes

Conventional thinking vs Strategic/Bayesian thinking

Which option is best? vs What would have to be true for each option to be a great choice?

What evidence supports the option we prefer? vs What do we currently believe about the conditions that must hold?

What are the risks in this option? vs What are the swing variables?

What analysis should we do? vs What evidence would most efficiently change our mind?

How do we justify a decision? vs What must we learn before commitment is rational?

The first mode is recommendation-led. It moves quickly from option to argument. The second is uncertainty-led. It moves from option to conditions, from conditions to swing variables, and from swing variables to evidence. That shift matters because most decisions do not turn on everything we do not know. They turn on a small number of uncertain things that would actually change the choice.

Those are the swing variables.

What are swing variables?

A swing variable is not just an important assumption. It is an uncertainty that can swing the decision.

Suppose a product team is deciding whether to:

build a new digital self-service channel,
improve the existing assisted-service model, or
partner with another provider.

Many things are uncertain: build cost, adoption, service volumes, customer satisfaction, change-management effort, regulatory complexity, supplier reliability. But not all of them matter equally to the choice.

Perhaps the decision changes almost entirely on three variables:

whether enough customers will adopt self-service,
whether the new channel will materially reduce cost-to-serve,
whether the organisation can deliver it before a market window closes.

Those are not merely risks. They are the swing variables. If belief about any one of them moves far enough, the preferred option may change.

This is the part that ordinary business cases often miss. They gather more facts, but not necessarily more decision-relevant facts. They measure what is available, visible, or politically comfortable, rather than what would change the decision.

The Playing to Win move

Lafley and Martin’s reverse-engineering process is powerful because it refuses to let teams collapse too early onto a favourite answer.

Their sequence is:

frame the choice,
generate strategic possibilities,
specify the conditions that must hold true for each possibility to be sound,
identify the barriers to choice,
design valid tests,
conduct the tests,
choose.

In previous posts https://www.recursiveloop.com.au/insights/better-product-decisions-need-bayesian-thinking and https://www.recursiveloop.com.au/insights/on-changing-your-mind I discuss how starting with the information you have and updating this position based on evidence is at the heart of the bayesian decision making process. In a sea of possibilities, step 3 in my view is the where people struggle most.

Instead of asking colleagues to abandon their preferred option, you ask them to help describe the world in which each option would be right. The argument shifts from “my answer versus yours” to “what would need to be true?”

The next move is equally important. The book says to identify the conditions people feel least confident are true — the barriers to choice — and focus testing on those first. Later it describes the need to focus deeply on the binding constraints, rather than analysing everything in parallel.

In decision-analysis language, these are the swing variables: the uncertainties worth resolving because they are capable of moving the decision.

The Bayesian layer

A Bayesian thinker would make the same process more explicit.

For every strategic option, ask:

What would have to be true for this option to be right?
What do we currently believe about those conditions?
Which of those conditions are swing variables?
What evidence would most efficiently change our belief about them?

That second question is the Bayesian addition. Playing to Win talks about confidence. Bayesian thinking asks us to make that confidence more explicit: not merely “we are unsure about adoption”, but perhaps “given comparable launches and our current evidence, we think there is a 30–50% chance adoption clears the threshold that makes this option worthwhile”.

The third question prevents analysis from becoming indiscriminate. You do not need to reduce every uncertainty. You need to reduce the uncertainties that determine the choice.

The fourth question introduces information value. A good experiment is not one that produces data. It is one that is likely to move belief about a swing variable enough to improve the decision.

That is why asking customers whether they “like the idea” is usually weak evidence. You might hear that whether the strategy is good or bad. But observing whether customers change behaviour, pay, return, refer, or switch may be much more diagnostic. The question is not simply “what evidence can we get?” It is “what evidence would we expect to see if the condition were true, and not expect to see if it were false?”

The real shift

The shift is from this:

Find the best answer, then defend it.

to this:

Expose the few things that would make each answer good or bad, then learn about those first.

Playing to Win gives us the strategic discipline for making that shift. Bayesian thinking sharpens it by making beliefs explicit, focusing attention on swing variables, and asking which evidence is actually worth purchasing before we commit.

The organisations that learn fastest are rarely the ones collecting the most information. They are the ones that know which uncertainty matters next.