12
people interviewed about why they fell off
3 of 12
would give the job to the person they’d most hate to disappoint
$0
ever at stake. Money makes people cheat.
How it works
Build a plan with Snitch
Snitch is the AI coach. You set a weight goal and pick your workout days and your gym. The app works out the numbers; Snitch explains them.
Weigh in three mornings a week
Photograph the scale and the app reads the number. The photo isn’t kept.
Show up at the gym
Stay 30 minutes inside the gym you pinned and the workout counts. There’s no button to tap and nothing to fudge.
Go quiet, and Snitch tells
Name one to three witnesses. Miss your weigh-ins for the week, or two workouts in a row, and each gets a message. Keep your word and they never hear a thing.
From idea to TestFlight
I’m building this alone, so I didn’t run a full team process. I kept the part that matters: each stage had one question to answer before the next one started.
1. Idea
What exactly am I betting on?
2. User research
Who would people trust with the job?
3. Market
Why do the tools that exist stop working?
4. MVP scope
What’s the least I can build that tests the bet?
5. Build
Does it run on its own, with the app closed?
6. Test
Does telling a witness bring people back?
1. The idea
I’m overweight, and not for lack of knowing what to do. I’ve had the apps, the reminders and the plans. What worked was my mom and my brother texting to ask if I’d been to the gym. Nobody else had noticed I’d stopped.
So I built a watchdog. Snitch keeps an eye on your weigh-ins and workouts, stays quiet while you keep your word, and snitches on you when you don’t. A watchdog that snitches: SnitchDog.
The bet: a message to someone you respect, sent only when you slip, keeps you going longer than money, streaks or a coach alone. Every stage after this tried to prove it wrong.
2. Talking to users
I asked twelve people trying to gain or lose weight three things: the last time you fell off, what happened and who noticed? Who would you hate to disappoint? Would you give them the job of keeping you on track?
The person you’d hate to disappoint isn’t the one you’d give the job
Only 3 of 12 would hand it to the person they named. What changed: who Snitch asks for.
Wives, dads, sisters and kids were all a firm no: love turns into policing, comparison or resentment. The people who said yes picked a gym buddy, an older brother and a former trainer: someone whose respect they want, not someone who worries about them. So Snitch asks for exactly that: a mentor, a coach, a training partner, the sibling who’d give you a hard time.
Logging food is what breaks
7 of 12 fell off around food tracking. What changed: there isn’t any.
Logging every almond, hitting a macro target while sick, weighing pasta, logging client dinners on the road. The tracking itself was the thing people quit. I had planned to add meal photos; this is why I cut them.
Life happens first, then the quitting
9 of 12 fell off after a real event: a newborn, a breakup, an injury, a cruise, the holidays.
Nobody quit because they stopped caring. They quit because something happened and the app carried on as if nothing had. So the plan is judged week by week, you can pause for up to two weeks, and a miss with a good reason can use one of two passes a month.
Streaks punish the moment you need help
A 45-day streak reset by one wedding killed one person’s motivation outright.
A streak turns one slip into a total loss. SnitchDog keeps a streak because people like watching it grow, but it’s yours alone: witnesses never see it, and it never decides whether they hear from Snitch. Only this week’s plan does.
Guilt makes people delete apps
Red “over budget” warnings and nagging notifications were the most common reason to uninstall.
An algorithm flashing red doesn’t care that you’re crying on the couch. Snitch never comments on the number or on food. When you slip, the consequence is a person checking in, not a push notification telling you you’ve failed.
Often, nobody notices
3 of 12 said literally nobody noticed they’d stopped.
Baggy clothes, living alone, hiding habits well. That silence is the gap SnitchDog fills: someone who will notice, on purpose.
Some want a professional, not a friend
4 of 12 would rather be held accountable by a trainer, a coach or someone neutral.
A paid trainer felt safer than burdening a friend, and less shameful than someone they see at dinner. That points at a later version where the witness is a trainer or coach, which is also a channel: one coach can bring many clients.
Who I won’t serve
For someone in eating disorder recovery, the numbers themselves were the trigger.
Scale photos and a watching witness are the wrong tool for that person, and no copy change fixes it. SnitchDog is for adults with a weight goal, and it says clearly at sign-up who it isn’t for.
3. Sizing up the market
Every tool below solves part of the problem. None pairs a private coach with people you respect.
Food and calorie trackers
MyFitnessPal, Lose It, Cal AI. Excellent data, but logging is the chore people quit, and a missed day costs nothing.
Coaching programs
Noom, WW. Real structure and psychology, but the lessons start to feel like homework, and engagement falls off sharply after the first month.
Money on the line
stickK, Beeminder, DietBet, HealthyWage. Stakes work in the short term, but they give people a reason to cheat and a reason to quit once the money stops.
Group challenges
BetterTogether, Strive, leaderboard apps. Fun for competitive friends, but public numbers and comparison are exactly what several people I interviewed said would hurt them.
Community feeds
Strava and its many imitators. Lots of people watching, nobody responsible. When a hundred followers see you stop posting, each assumes someone else will say something, so nobody does. It’s the bystander effect, and the same reason people put in less effort in a group than alone (social loafing). Kudos feel good, but they don’t notice when you’re gone.
AI enforcers
Forfeit and Overlord. I found Overlord, a Y Combinator–backed agent, after my first build. It can watch your location, screen time and spending, block your apps and charge your card. That treats the user as the opponent, and money gives people a reason to cheat: DietBet needs code words, video audits and referees to keep bets honest.
Where SnitchDog sits
One goal, proof you can’t fake, a coach who follows up, and a few people you respect who hear only if you slip. Overlord goes after every habit; I’m narrowing to people who keep falling off a weight goal. That niche is easier to explain, easier to find and easier to win.
4. Scoping the MVP
I wrote a requirements doc with a wireframe for every screen, and gave each requirement an id that the code still points back to.
Every feature then faced one question: does it help prove the bet? If it did, it went in. If it was a good idea that didn’t, it went on the roadmap. If it gave people a reason to cheat or quit, I cut it.
That cut food logging (the chore people quit), money stakes and app blocking (reasons to cheat), public feeds and group challenges (a crowd where nobody feels responsible), and every habit except weight. These are the rules that stayed:
No money, ever
The stake is a person, not a payment.
A person you respect, told only when it matters, can’t be gamed, and doesn’t vanish when a payment plan ends. In Volpp’s trial, weight came back once the payments stopped.
Witnesses see whether you showed up, and nothing else
Never your weight, your photos or your chats.
Nothing you type is forwarded to them, and they don’t get a dashboard, because a dashboard is a license to police.
Silence is the good outcome
Keep your word and your witnesses never hear from Snitch.
The best week is one where nothing happens. A witness who gets a message every week stops reading them.
Weekly, not daily
A missed morning costs nothing on its own.
You promise a number of weigh-ins and workouts per week. Only the week as a whole decides whether a witness hears anything.
You can’t leave quietly
Deleting the app is not a way to dodge the message.
Removing a witness or deleting your account tells every witness, unless you reached your goal. Otherwise quitting would be the easiest way out.
Consent and a floor
Witnesses opt in; the app is for adults, at a healthy pace.
Either side can end it at any time. No goals below a healthy weight, no aggressive pace, and a sign-up that says plainly who this isn’t for.
5. Building it
I built in small slices, each on its own branch, merged only once it worked. The first slice was the whole loop, done thinly: the app, the server, and the rules that decide when a witness hears. Everything after that made the loop better.
The hard part runs while the app is closed. The server checks every plan every 15 minutes, so a missed week gets noticed even if you never open the app again. Building it changed parts of the spec:
Workouts are checked, not claimed
A yes-or-no question became a geofence around your gym.
The spec asked “did you do it?” after each session, the easiest answer in the app to fake. Now a workout counts when your phone stays 30 minutes at the gym you pinned. The phone works out “arrived” and “left” itself; no location trail leaves it.
One witness became up to three
Each is messaged alone, and none is told who the others are.
A crowd lets everyone assume someone else will check in. Because no witness knows who else is watching, each still acts as if it’s down to them. Witnesses don’t need an app either: the invite opens Telegram.
The pause made the first version
Planned for later, built now, because that’s where people fell off.
You can pause for up to 14 days, and a miss with a good reason can use one of two passes a month.
The scale reads itself
Snap the scale and the number fills in.
This was on the roadmap too. Claude reads the photo on the server and only the number comes back; the photo isn’t kept. Offline, the phone reads it itself.
Ember became Snitch
A flame can’t keep watch, so the coach became a dog.
The coach started as Ember, a paper flame. It became Snitch, a German Shepherd with a whistle, and the app became SnitchDog. You pick how Snitch talks at sign-up, up to a drill sergeant, but it never comments on the number.
Limits on cost, noise and access
Caps on AI spend and witness messages, and data locked to its owner.
Each plan has a daily cap on AI spend, and witnesses get at most three messages a day. Every table is locked to the account that owns it, and a test checks that one account can’t reach another’s data.
6. Testing it
Automated tests walk through every screen on the iPhone simulator: sign-up, the plan, weighing in, witnesses, the coach, pauses and deleting an account. They caught bugs no spec would have, like Snitch telling someone it had granted the same pass twice.
Next is TestFlight, Apple’s beta channel, and a concierge test: ten people for four weeks, with me reading every message to a witness before I let any of it run unwatched. My mom and my brother will be the first witnesses.
These are the numbers I’ll judge it by. They’re guesses going in, not results.
North star: kept at week 8
The share of people who start and are still keeping their weekly plan in week 8. Most weight-loss apps fall below 10% retention by six months, so this is the number that matters.
Activation: a witness says yes
The share of new users whose witness accepts within 72 hours. This is the riskiest step in the funnel. If fewer than half get there, the witness is a wall, not a feature.
The thesis: does telling someone work?
After a witness is told, the share of people back on plan the following week. If escalations lead to uninstalls instead of comebacks, the core idea is wrong.
Guardrails
Uninstalls within 48 hours of an escalation, witnesses who step away, a short “does this make you feel worse?” check-in at weeks 2 and 6, and any sustained loss faster than a healthy pace.
What’s next
- Ship to TestFlight and run the concierge test.
- Trainers and coaches as witnesses. 4 of 12 people wanted a professional, and one coach can bring many clients.
- Marketing to witnesses: the parent or friend who wants someone they love to get healthy is often more motivated than that person is. Give them the invite to send.
- Other habits, once weight works. The same proof model could cover sleep or reading.
Up next
When I give up on Netflix, I open YouTube.