Start at the End: How to Build Products That Create Change

by Matt Wallaert

asking “Why would people want to do that in the first place?” and “Why aren’t they doing it already?”

The first maps motivation to the conditions that make behavior more likely, called promoting pressures. The second helps us understand the factors (like “AIDS kills” stamps) that make the behavior less likely, called inhibiting pressures. Identifying and consciously influencing the strength of those pressures is the basis of designing for behavior change, codified here as the Intervention Design Process (IDP)—interventions being the things we build to change the pressures and thus the resultant behaviors.

creation is one of the single best ways to better the world.

Designing for behavior change is about creating the conditions that allow us to act on our original motivations.

It is fundamentally true that any pressure that can be used to make behavior more likely can also be used to make it less likely; in the same way that stronger promoting or weaker inhibiting pressures make a behavior more likely, weaker promoting or stronger inhibiting pressures make it less likely.

Dualities

When we want to change behavior, we start with a potential insight—an observation about the distance between our current world and the counterfactual one in which we want to live. We then validate the insight and flesh it out into a behavioral statement, which we use to map the pressures that are creating the current state of the behavior and thus are the levers we push to change it.

Organisational need vs current state

the lesson plans were framed as “designed specifically to foster curiosity.” Good behavioral design uses science as a process but isn’t about producing pure truths,

How do you measure curiosity?

We got 40 percent more searches out of a simple product and some lesson plans, because we knew what behavior we wanted at the end and, Marketing be damned, worked backward to design the interventions to get us there.

Does more searches equate to more curiosity? Can curiosity happen offline?

Bing in the Classroom, and a happy ending if there ever was one.

Changed behaviour: more students use one specific tool (Bing) to search for answers to questions. Not necessarily curiosity. Nor creativity.

a potential insight really is: the recognition of some potential split and the opportunity to move closer to the utopian end of the spectrum. A potential insight expresses the distance between those two universes and, once validated, allows us to start understanding the gap and how to design interventions that bridge it.

New word: bifurcate (branch into two).

The IDP exists simply to make such alternative realities come into being, to take potential insights and run that potential to ground to see if it can create value.

There are four major types of potential insights: quantitative, qualitative, apocryphal, and external.

confirmation bias: as soon as you start to believe something, it begins to selectively attend to evidence in order to support that belief, because changing your mind costs cognitive resources

Confirmation bias

Good behavioral scientists are T-shaped: they have one area of deep expertise (the legs of the T) and broad interests across other disciplines (the arms of the T).

We can never be absolutely certain about the truth behind any insight, pressure, or intervention, but by triangulating from data and observation and structurally resisting groupthink, we can eliminate risks and increase our chances of a successful, scaled outcome. This is behavioral science, and part of science is a willingness to be wrong.

The key is to focus those insights on behaviors, not on how we create them. You have to fall in love with the problem, not the solution.

When [population] wants to [motivation], and they [limitations], they will [behavior] (as measured by [data]).

Population = the group of people whose behavior you are trying to change Motivation = the core motive for why people engage in a behavior Limitations = the binary preconditions necessary for the behavior to happen that are outside your control Behavior = the measurable activity you want people to always do when they have the motivation and limitations above Data = how you quantify that they are doing the behavior

So why describe the absolute? Because it increases the likelihood you’ll get there.

Your anchor inevitably affects the interventions you design to reach it.

Anchoring and adjustment

autonomy and accountability. The behavioral statement should be as large as possible but small enough that the person could be held individually accountable for its success or failure.

Autonomy and accountability

notice the similarity between a behavioral statement and an objective and key result (OKR). “As measured by data” is really just the KR and the rest is just the O, but phrased in such a way as to be observably descriptive of the world you’re trying to create. If you’re already doing OKRs for your planning process, you’re already ready to simply sub in behavioral statements and reorient entirely toward behavior.

KPIs

when we talk about designing for behavior change, we are actually talking about changing the pressures that determine the behavior, rather than directly changing the behavior itself.

I want you to actually create behavior change, every single day. And sometimes simple frameworks are the best way to get that done.

These arrows represent the balance of competing pressures that create our behavior: promoting pressures—the up arrow—make a behavior more likely and inhibiting pressures—the down arrow—make a behavior less likely. What we actually do is determined by the net product of those forces. If the promoting pressures overcome the inhibiting pressures, we act. If the inhibiting pressures are stronger, we don’t. And both sides are equally responsible for the ultimate behavior:

Promoting pressures vs inhibiting pressres

a successful intervention is about creating a world that doesn’t currently exist, you’ll always need to be working through thought experiments and inhabiting imaginary realms.

complete a pressure map as possible. Focus on just inhibiting or promoting pressures and then switch. Try reversing the polarity of your behavioral statement; if you think about how to make sure no one ever takes an Uber, you’ll likely uncover some pressures you can use to make sure they do. And make sure you have a diverse room. The more gender, ethnic, cultural, cognitive, and other variation you can introduce, the smaller your blind spots and the less likely you’ll fall victim to biases.

Completing a pressure map

goal isn’t just having the most ideas; it is having different ideas. Better ones.

Generatng novel ideas

When we focus on novelty, we lose sight of behavioral outcomes.

It is entirely possible for a single pressure to give rise to many interventions and equally possible for one intervention to satisfy many pressures.

Intervention design is really just the translation of pressures into something we can actually create; if pressures are the levers, interventions are how we pull them, hopefully in the right order and with the right strength.

although you can’t pilot everything, you also shouldn’t pilot only one thing. Intervention selection isn’t about driving to a single solution but rather about setting yourself up for a range of pilots that maximize the chances of creating behavior change.

behavioral science, like all good science, is based on the assumption that your interventions won’t work.

Pre-mortem

we’re fighting against: confirmation bias and our natural tendency to believe something is true simply because we want to justify the effort.

We want to recoup our sunk costs

optimum distinctiveness: a range of options that together cover as much of the spectrum as possible, but with relatively little overlap.

Optimum distinctveness

The key is to stay focused on the behavior you are trying to change. So in intervention selection, assume nothing. What worked elsewhere may not work here; what has failed in another context may be viable in this one. Stay fixated on optimal distinctiveness so that you can maximize the chances that you’ll find at least some behavior change and move forward from there.

the golden rule of designing sign-up flows is that you don’t add anything unnecessary, because every additional step introduces greater inhibiting pressure and thus lowers the behavior that is registration.

Everything we create, we create to change behavior. And some of our most lauded roles in life are explicitly about behavior change: teacher, doctor, parent, etc. Yet we never talk about them as behavior-changing, and to accuse someone of trying to consciously change the behavior of others is to slander them;

self-serving bias that goes roughly like this: when I do good things, it is because I’m a good person, but when I do bad things, it is because I was affected by my environment. The reverse is true of others: when they do good, it is because of their environment, while their bad deeds are because they suck and are terrible people.

Self-serving bias. Judgement.

behavior change is a war between opposing sides who are invested in specific outcome behaviors, and you should choose consciously which side you’re on. Your interventions will always appear ethical when filtered through the lens of your own values, but by understanding some of the potential issues, you can avoid waking up and finding that you aren’t proud of the work you’ve done.

Ethics and values

ethical check based on two factors: what behavior we are changing and how we are changing behavior. And these ethical problems are mapped to the two fundamental behavioral gaps: the intention-action gap and the intention-goal gap.

The motivation component of a behavioral statement can’t help us in this case, because the person has already rejected the behavior as a method for honoring that motivation. Thus we have to deal both with what and then with how, since solving an intention-goal gap still leaves the potential for an intention-action gap. So let’s start with what and the intention-goal conundrum.

the behavior change is unethical if it does not honor any of the population’s motivations, not simply if it doesn’t honor the most obvious one.

By being transparent about both our behavioral statement and our interventions, we allow for the ethics to be evaluated by people who do not have the same motivations that we do.

Transparency

If: your outcome behavior is not the result of any of the population’s motivations or the benefit of your outcome behavior or an intervention to produce it does not outweigh the cost to an alternative motivation or you are unwilling to publicly describe and take responsibility for the outcome behavior or intervention it is unethical.

if people don’t agree on terms, everything stalls out.

Pilots are tightly scoped interventions that we expect not to work (remember, we have to explicitly prove efficacy as a defense against confirmation bias), so we use small populations, focus on speed to market, and do them in an operationally dirty way.

Pilots

Speed and resource efficiency are also important here. Because we chose multiple interventions during intervention selection, we’ll likely be running three to five concurrent pilots at any given time. If those pilots are too operationally heavy, we’ll stall out, so we have to be constantly focused on finding the lowest-fidelity version of an intervention that will still result in behavior change.

Scale: less than 2 weeks to implement

Pilot validation is just like insight validation: qualitative and quantitative confirmation that you’re headed in the right direction.

Pilots let us make light bets that don’t demand a lot of commitment and nothing kills the confirmation bias faster. By putting a step before tests and running so many pilots outside our normal operations, we avoid advocating for appealing but bad ideas, simply because we weren’t that invested in the first place and have plenty of other things to try.

If only Wellington's cycleways were piloted before stuffing up our roads.