Nia Bowers
As AI-powered coaching apps proliferate in 2026, promising to transform habits and change lives, one veteran of the space is warning founders about a fundamental flaw in their approach.
The hardest problem in consumer technology is rarely building the product. It is getting anyone to change what they do – especially when the behavior happens off-screen. Exercise. Sleep. Meditation. Eating well. According to the World Health Organization, about thirty-one percent of adults worldwide did not meet recommended physical activity levels in 2022, a share projected to reach thirty-five percent by 2030. Apps promising to help launch by the thousands every year. Most are abandoned within weeks. Palash Kala has been tinkering with this problem on and off for years – and came away with conclusions that challenge how the industry thinks about motivation.
“Products that require users to be disciplined don’t scale impact,” Mr. Kala says. “If your product needs users to work harder than they already are, you’re not really solving the problem. You’re keeping the ownership with the user.”
The Experiments
Over the years, Mr. Kala tried to change human behavior with technology. He built enterprise nudge systems deployed across eleven countries to Fortune 100 companies. He shipped an AI habit coach to thousands of users. He ran gamified fitness programs inside one of India’s largest fintechs. He conducted habit-building workshops, guided strangers through accountability calls, and ran personal experiments on himself.
The results were mixed – and the patterns that emerged from those mixed results became the insight.
What Worked for Him (But Didn’t Scale)
In 2020, Mr. Kala made a deal with his brother-in-law: he would pay fifty rupees, about sixty cents, for every Netflix episode he watched beyond one per day. Over two months, he successfully limited himself to one episode while working through seven seasons of Suits. The discipline persisted for a while after the agreement ended.
“It worked for me,” Mr. Kala says. “But I was highly motivated. I had ego on the line. I put real effort into it. When I tried the same technique with friends, some succeeded and some didn’t. It never scaled.”
The same pattern appeared in his workshops. Some participants changed their behavior. Others didn’t. The ones who succeeded were the ones who came in motivated. The technique didn’t create motivation – it channeled motivation that already existed.
This is the problem with commitment devices. Stickk, the commitment platform developed at Yale, lets users put money on the line for their goals. It works for people who use it – but it hasn’t scaled to mass adoption. Focusmate, a social accountability platform where users work alongside strangers on video, is a genuinely good product – users who try it tend to love it – but it has reached only around one million dollars in annual revenue with roughly twelve employees. It works for the people who use it, but it hasn’t been adopted by the masses. The constraint is the same: these products require users to opt in and stay opted in. The moment motivation dips, users opt out.
“Opt-in is still your decision,” Mr. Kala notes. “Guilt or fear might drive you for a while, but that can’t sustain. You have to get to the habit loop where it becomes a positive association – and most people never get there.”
What Worked Short-Term (But Didn’t Sustain)
At CRED, Mr. Kala organized internal workshops with Nir Eyal, author of Hooked, and Yu-Kai Chou, the gamification researcher behind the Octalysis framework – bringing these behavioral science practitioners directly to the product team. He also created Platform.Fit, a gamified fitness initiative for employees, and tested four versions of the program.
The third version, built around weekly challenges with betting mechanics, penalties, rewards, and social accountability, worked: sixty-five percent of participants completed at least three workouts per week throughout the six-week program. The program scaled from fifty people to adoption across an organization of a thousand.
“In that sense it succeeded,” Mr. Kala says. “Teams opted in. People showed up. But when the program ended, the habits didn’t stick for most people.”
The social accountability helped short-term, but over time it started to feel forceful. People were looking at each other, feeling the pressure – and that pressure only works while the structure is there. Remove the external structure and most people revert.
“Gaming mechanics don’t transfer to positive behavior change the way they work in games themselves,” Mr. Kala observes. “In games, the activity is inherently engaging. In behavior change, you’re trying to make something that isn’t engaging feel like it is. That’s a fundamentally harder problem.”
What Looked Like Success (But Had a Catch)
Before his consumer experiments, Mr. Kala worked at worxogo, an enterprise SaaS company that uses behavioral science to build high-performance sales teams. Its AI engine, Mia, delivered personalized nudges to sales representatives and managers across financial services, pharmaceuticals, and other sectors. The product deployed across eleven countries to Fortune 100 organizations. Managers were happy. The product made revenue.
But Mr. Kala is candid about what was actually driving the results.
“The real behavior change happened when incentives were aligned,” he says. “Managers are authorities. When a manager sends a nudge, there’s an element of monitoring. Employees respond because there are professional consequences for not responding.”
When the financial incentives aligned – when the nudge matched what the employee already wanted or needed to do – behavior changed. When the incentives didn’t align, the nudges were just noise.
“We didn’t create accountability,” Mr. Kala says. “We instrumented accountability that already existed. That’s a different thing than changing behavior.”
The deeper lesson was about scope: nudges work well for optimization. A/B tests on messaging, timing, framing – these can move metrics. But when you try to build a zero-to-one product based purely on nudges, it struggles. The nudge is a layer on top of existing motivation, not a replacement for it.
What Didn’t Work (And What It Taught)
In 2023, Mr. Kala joined Nintee, a startup founded by Paras Chopra, who had bootstrapped Wingify to fifty million dollars in annual revenue. Nintee’s mission was to build an AI coach for self-growth – one of the first attempts to use generative AI for habit coaching.
Thousands of people signed up. Some loved it. Some returned regularly and built real habits. But overall retention stayed lower than the team needed, no matter what they tried.
“We thought the UX was the problem,” Mr. Kala recalls. “We started on Discord, then built our own app. We tried streaks, reminders, personalized plans. Each optimization moved the needle slightly, but not enough.”
After eight months, the team pivoted to interactive learning for self-help, inspired by Duolingo. They validated content cheaply on PowerPoint slides and Instagram before building. They launched an alpha, ran ads, measured retention. Day-one retention was fifteen percent. Industry benchmarks show health and fitness apps retain only about three percent of users by day thirty.
“We checked other products in the category,” Mr. Kala says. “Retention was low everywhere – not just in self-improvement, but across education technology broadly. Duolingo is the exception, not the norm – and they’ve spent years and hundreds of millions of dollars getting there. Even the Duolingo founders would tell you this is an incredibly hard category.”
Almost a year of experimentation later, Nintee shut down. Paras Chopra, Nintee’s founder, reflects on the experience: “We tried everything – AI coaching, gamification, social features. Some users genuinely loved it and built real habits. But for the majority, we couldn’t solve the fundamental problem: the hard work happens outside the app, and we couldn’t do that work for them.”
For Mr. Kala, the learning was clear: AI didn’t change the fundamental bottleneck. The problem wasn’t that human coaches were unavailable. The problem was that even human accountability doesn’t work long-term at scale – because it’s opt-in.
“The user still has to do the hard thing outside the app,” he says. “Close Instagram and meditate. Put down the fork. Go for the run. Technology can remind. Technology can play with words, use different mechanics, try to engage you – but that work is really hard. And in the end, technology cannot want things for people.”
What Actually Scales
Mr. Kala’s conclusion is simple – and obvious in hindsight: technology scales impact when it radically simplifies existing behavior. It struggles when it tries to create new motivation.
“Strava doesn’t make non-runners into runners,” he says. “It makes tracking effortless for people who already run. Fitbit requires zero effort from the user – the tracking comes for free. That’s why it sticks. Instagram didn’t create the behavior of sharing photos with friends – it made it radically easier.”
The pattern extends beyond behavior change. When Mr. Kala joined Composio, an AI agent infrastructure company, he recognized the same dynamic. People were integrating MCP servers one-on-one with each agent, sometimes building custom integrations themselves. Mr. Kala led the launch of Composio For You, a product that simplified that existing behavior – letting users connect thousands of apps through natural language, without having to worry about connecting them repeatedly across agents or building integrations. The product grew ten-fold in its first two months. It didn’t try to convince anyone to want something they didn’t already want. It made something they were already doing radically easier.
“The products that scale are the ones that find an existing behavior and ask: how do I radically simplify this?” Mr. Kala says. “Not: how do I motivate people to do something new?”
The Honest Conclusion
Mr. Kala is direct about what his experiments taught him – and what they didn’t.
“I tried a lot of things. Some worked for me personally because I was motivated. Some worked short-term because external structures held. Some looked like success but were really driven by incentive alignment. Very little scaled sustainable impact in behavior change – but Composio did, because it simplified what people were already trying to do.”
The founders who will succeed in positive behavior change, he argues, are the ones who accept these constraints rather than fighting them.
“If your product requires users to be disciplined, it won’t scale,” he says. “If it requires opt-in commitment that users can abandon when motivation dips, it won’t scale. If it relies on guilt or fear or social pressure, it might work for a while, but it won’t sustain.”
What works is different: find an existing behavior with existing motivation, and make it radically easier. That’s what Fitbit did. That’s what Strava did. That’s what Composio is doing. That’s what the products that actually scale do.
“It sounds obvious in hindsight,” Mr. Kala admits. “BJ Fogg’s research says this. The Jobs-to-be-Done framework says this. I organized workshops with Nir Eyal at CRED – I knew the theory. But there’s a difference between intellectually knowing something and viscerally believing it. It took a few experiments to really internalize it. Simplify existing behavior. Don’t try to motivate.”
—
Palash Kala is Head of Product at Composio. He previously built behavioral nudge systems at worxogo and consumer products at CRED, and co-founded Hovi AI.