Predictably Irrational: How Behavioral Science Reshapes Product Design

Kristen Berman co-founded Irrational Labs in 2013 with the behavioral economist Dan Ariely. It now has about 20 behavioral scientists and has worked with hundreds of companies, including Google, TikTok, One Medical, Credit Karma, Intuit, Microsoft and LinkedIn. She spent roughly three years embedded at Google, where her team worked with more than 25 groups, from self-driving cars to YouTube.

Berman defines behavioral economics as a reaction to classical economics, which assumes people decide rationally and without emotion. In practice people are emotional, favor their present selves over their future selves, and follow social norms. The useful part is that these deviations are systematic, so once you understand how and why people behave, you can start to change it.

A popular fintech app found that budgeting was its most requested feature across support forums and interviews. Berman's team convinced it to run an experiment first, with a control group and two budgeting variants across 10,000 people. The result was nothing: no change in average spend and no change in spend variability, which shocked the client but not her team.

The reason is what she calls a behavioral diagnosis: map every step needed to perform the behavior. To budget, you must know your budget, know where you are in the month, make a plan and then follow it repeatedly. That is a lot of work, and Berman says any work you put on the user deserves skepticism until you measure that they actually do it.

Her first line of defense is making things easier, and in personal finance that usually means defaults, the way automatic 401(k) enrollment drives retirement saving. Where defaults are too costly to build, she suggests rules of thumb that remove in-the-moment decisions, such as "I only take a Lyft on weekends". She adds that behavior is driven by what people do rather than what they intend, so a goal alone is not enough and the environment has to change.

The usual rule is that reducing signup friction raises conversion, but Berman describes one case where adding a step helps. Asking a question puts an idea in someone's head, just as she briefly made Lenny think about her mother. If the question makes users think about the benefits of the product, their motivation rises and they push through later friction more willingly.

Apartment List asks what kind of apartment you want, and Trunk Club reported that a similar approach increased conversion by 133%. Berman's team put a quiz on the website of TytoCare, a remote-physician device. Among visitors who finished it, 53% went on to purchase, against 37% of those who did not.

The questions must be easy, though. Berman recommends multiple choice or drop-downs and says to avoid open text fields. Removing one open-ended question that the team thought was easy improved conversion by 40%. A good question can also double as setup, such as a bill-pay app asking how quickly you want bills paid, which both highlights the benefit and configures the app.

Irrational Labs summarizes its approach in a three-B model. The first B is behavior, and Berman says companies are good at outcomes but vague about the action they want. She asks teams to be uncomfortably specific, for example a Peloton user doing two 10-minute workouts with two different instructors within seven days of starting. If your team is not arguing about the definition, she says, you are doing it wrong.

The second B is barriers, which are logistical (a form field, a credit card entry) or cognitive. Cognitive barriers include uncertainty aversion, optimism bias, information aversion and the status quo effect. Uncertainty, such as not knowing whether a ride will arrive on time, pushes people to look elsewhere or decide nothing, and the status quo means that asking people to do something new requires more motivation or fewer barriers.

The third B is benefits, specifically immediate ones. People are present-biased, so long-term reasons are not enough. Asana users log a task because it keeps the project on time, but also because of completion bias, the satisfaction of a checkbox, and social desirability, since teammates are notified when you finish something.

Asked about the dark side of these techniques, Berman says the answer comes down to incentives, and that we are what we measure. She gives a personal example from LendingClub, which offered her a large payment if she achieved a five-point conversion bump and nothing otherwise. Under that structure she found herself suggesting things the legal team rejected, even as she led a Duke lab focused on low-income Americans' financial health.

She offers two tips for setting incentives. First, put the KPI on the behavior, which aligns more closely with customer outcomes than raw active use or retention (which should still be measured). Second, lengthen the horizon, since a quarterly target encourages short-term bumps while an annual or longer one pushes teams to look after both company and customers.

TikTok asked Irrational Labs to reduce the spread of misinformation. The team chose a narrow behavior, decreasing shares, rather than likes, comments or engagement in general. Because the goal was to get people to do less, the lever was barriers, and Berman cites a study in which a slower-closing elevator door pushed people toward the stairs.

TikTok is a fast, hot-state platform, so the team slowed people down in two ways: a label marking the video as unverified, and an "Are you sure?" pop-up after tapping share. Together they added logistical friction and cognitive friction by asking people to reconsider. The combination reduced shares by 24%, and Berman says Facebook and Twitter use similar tactics.

The process began with a literature review, which showed that reminding people of their value of accuracy works only at the point of sharing, not before or after. The team drafted about 30 implementations and tested five pop-up versions with over a thousand users recruited through Prolific. They compared conditions against each other rather than asking whether people liked them, then ran two variants plus a control in the product.

One Medical asked for help with engagement, and Berman first had the team agree on what that word meant, noting that teammates asked separately usually give different answers. The chosen behavior was booking a doctor's appointment right after signup. Without that, a new member who gets sick six months later will probably just call their usual doctor.

The barriers were choosing a provider, finding a time and understanding what One Medical is for, while the benefit was getting health questions answered quickly. During onboarding the team asked a few questions about mental, sleep and physical health, recommended one provider with a reason, and offered a limited set of times, including a virtual appointment the next day. That raised bookings during onboarding by 20%.

Asked for general patterns, Berman cautions that behavior is contextual, which is why she insists on testing. Her rules of thumb are to show the benefit immediately, make steps easier, and move benefits from abstract to concrete. She also suggests reminding people who drop out of long flows and returning them to where they left off, as Wealthfront does.

Berman distinguishes the real benefit of an action from an immediate one, such as following social norms. In one fintech example, framing each step of syncing a bank account as something completed, or as an error message to make disappear, tapped completion bias. She calls this Right for Wrong, doing the right thing for the wrong reason, and warns it should be used sparingly because nobody loves an app full of error messages.

Other examples include prizes for vaccination, lottery tickets in Ohio, and pizza at polling lines. At Peloton, the real reason to exercise is health, but the pull of an instructor shout-out or a streak gets people to class. Clubhouse had a similar status mechanic, with those friends with people on stage in one tier and everyone else in another.

She stresses that the end must be good for the customer. Some nonprofits resist adding deadlines as too harsh, but Berman calls deadlines a gift because they help people prioritize. Adding a deadline to Kiva's borrower signup flow raised conversion.

For product teams without the budget for consultants, Berman suggests starting with process. First, run a workshop to agree on the uncomfortably specific behavior you want to change, since that makes conversations about the underlying psychology easier. Second, do a behavioral diagnosis, which she describes as a journey map on steroids that studies what people actually do rather than what they say.

In practice, her team builds a deck of 200 to 300 slides of product screenshots and lays out every step toward the behavior, then overlays the psychology at each point. In a health app, for instance, information aversion might explain why people avoid viewing test results, which opens a brainstorm on how to ease them past that step. She also mentions Irrational Labs' self-paced eight-week bootcamp for designers, product managers and marketers.