Stop minting coins and start hosting the room
By Craig Joseph, MD ·
The University of Illinois once offered employees cash and paid time off to join a wellness program, then did what most employers never do and let economists assign people at random to be offered the program or not. Random assignment matters because volunteers differ from everyone else; a lottery produces two groups that differ only in whether they got the offer. If the researchers had simply compared the employees who signed up with the ones who didn’t, the program would have looked like a success, because the joiners were healthier and cheaper to insure. Compared with each other, the two lottery groups barely differed.
Paying for habits people already had
Many vendor slides make that first comparison. Among the more than 4,800 employees who were randomized, the offered group got screened more frequently, and nothing else changed: not health behaviors, not medical costs, not productivity. Before the program even began, the eventual participants were already spending roughly $1,400 a year less on care and were already the regulars at the gym.
A randomized trial in JAMA at BJ’s Wholesale Club came out the same way. Workers at 20 program sites, measured against 140 controls, finished 1.3 of eight modules on average and after 18 months reported more exercise and weight management, but … with no change in cholesterol, blood pressure, BMI, absenteeism, or spending. A seven-year observational analysis at PepsiCo put the lifestyle piece’s return at 48 cents per dollar (#oops), while disease management for employees already living with a chronic condition saved $136 per member per month.
The common design pays for results that biology partly decides, issues the reward from an institution, keeps it private, sets the target centrally, and stops at the end of the benefit year.
A two-dollar coin
AA’s bronze medallion costs a couple of dollars from a recovery supplier, and people carry it for decades. The behavioral scientist Kristen Berman’s analysis of Starbucks Rewards offers two principles that help explain why. Good loyalty programs “design for the small stretch,” asking for a little more of what you already do. And visible status motivates: “One reason airline loyalty programs are so effective is that status is public,” while a coffee tier hides inside an app.
The chip follows both principles, though nobody planned it. Recognition starts at 24 hours and is densest early on, when the risk of drinking again peaks: typically 30, 60, and 90 days, then at intervals until the one-year medallion. Newcomers commit only to the next day. The chip counts days, which the person controls, rather than liver enzymes, which they partly don’t. It is handed over in front of people who have earned their own, and it never lapses. What it confers is an identity: a discount records a purchase, but a chip says its holder is someone with a year of sobriety.
The part a clinician can deliver
The best evidence for AA has nothing to do with chips. It comes from a 2020 Cochrane review by Kelly, Humphreys, and Ferri, covering 27 studies and 10,565 people, and it is about what clinicians do. Its high-certainty finding is that manualized twelve-step facilitation, a structured treatment delivered by professionals to get patients into AA and keep them there, beat established therapies such as cognitive behavioral therapy on continuous abstinence at 12 months. The authors judge it probably saves substantial healthcare costs too.
What won was a clinician deliberately handing a patient to a community the health system doesn’t own. Nobody in healthcare designed that community, and it has worked since the 1940s.
Status that can’t be issued
AA doesn’t manufacture chips or approve any official medallion; the custom spread from local groups, said to have begun in Indianapolis in 1942. A chip’s worth comes from whoever hands it over, so a health system can’t print its own.
Institutional programs also end. In Mitesh Patel’s STEP UP trial, 602 adults with overweight or obesity chose their own step goals above a measured baseline and earned points and levels, with a social design of support, collaboration, or competition layered on. For the 24 weeks the game ran, all three groups walked more than a control group that had trackers but no game, and the competition group did best, adding 920 steps a day over controls. In the 12 weeks after the game ended, activity fell in all three groups, and only competition stayed significantly ahead, one more reminder that nudges have limits. Patel told me on our 2024 podcast episode that Penn’s Nudge Unit grew out of an idea contest with an iPad and gift cards as prizes, and that entrants cared far more about pitching the CEO and executive team.
Alcohol use disorder carries existential stakes; a borderline A1c does not, and I doubt any hospital can make someone want applause for three months of walking. On Strava, kudos go to the slow runner as readily as the fast one, which puts it partway between. Most chronic disease probably lands in that middle too, where the right reward is smaller than a medallion and bigger than a gift card.
What to build instead
Reset each patient’s goal to one step beyond current behavior and recognize progress within days. Targets like 7% weight loss or 150 minutes of weekly exercise are finish lines that discourage the patients who most need to start.
Then build recognition into the groups a health system already hosts: cardiac rehabilitation, Diabetes Prevention Program classes, group prenatal care, and shared medical appointments. Reward attendance and effort, not lab values; let patients opt in; let peers do the recognizing; and never let a milestone expire. Give clinicians the facilitator’s job: the warm handoff and the follow-up call. Then judge the program by how many people still attend a year after it ends, not by enrollment.
A question for the next renewal
Ask your engagement vendor which comparison produced its results. Then ask how many of last year’s patients are still at it.