The chai-wala has better retention than most apps because he has turned a purchase into a habit: the same cue (the 4 pm break), the same place, the same order he remembers, and a tab you settle at month-end. Indian shopping apps kept only about 4% of new Android users after 30 days through 2025, according to AppsFlyer; a constant churn rate that loses 96% in a month leaves essentially nobody after a quarter.
- ~4% day-30 retention for Indian shopping apps on Android through 2025 implies about 10.7% daily churn.
- 66 days is the mean time for a daily behaviour in a stable context to become automatic (Lally 2010).
- 9.4% of users become permanent in a habit model, against almost none under constant churn.
- Four levers: a time cue, a fixed place, a remembered order and a month-end tab.
- 34% vs 19% loyalty-card completion with a two-stamp head start (Nunes and Drèze).
Habit research by Lally, Wood and Neal shows why the tea stall wins: a behaviour repeated daily in a stable context becomes automatic in about 66 days on average, after which people repeat it even when the product is no better.
Each of his four levers has a measurable app equivalent, and the rest of this post builds them one by one.
India drinks tea on a scale no app can match: about 840 grams per person per year, and roughly 80% of the country's tea production stays at home. Most of it is bought or brewed as a daily ritual, and a large share is sold one glass at a time at roadside stalls.
Nobody at a tea stall runs a retention dashboard, yet the regulars come back every day for years. This post treats the chai-wala as a retention system, models it with the maths of habit formation, and turns each of his levers into something a product or marketing team can build.
How bad is app retention compared with a tea stall?
Very. An app that keeps 4% of users at day 30 is losing about 10% of the remaining users every single day, while a tea stall regular who has kept coming for months is, almost by definition, losing very few.
The AppsFlyer data on Indian shopping apps reported by CXOToday puts day-30 retention at about 4% on Android across the second, third and fourth quarters of 2025, before it fell a further 41% by mid-2026 even as installs grew 85% since 2024. Convert that to a daily hazard rate, assuming churn is constant:
- Survival under constant churn. \( S(t) = e^{-h t} \), where \( h \) is the daily probability of leaving.
- Solve for \( h \) from day 30. \( 0.04 = e^{-30h} \Rightarrow h = -\ln(0.04)/30 = 0.107 \) per day.
- Project forward. \( S(60) = e^{-6.4} = 0.2\% \), and by day 90 it rounds to zero.
Real apps do better than that projection, because churn is not constant: the least interested users leave first and the survivors are stickier, the pattern Fader and Hardie modelled, which I covered in the leaky lota.
But the reason churn falls over time is the interesting part. In the people who stay, a habit has formed. That is the chai-wala's whole business, and one of the world's biggest retail brands was built by copying it.
In 1983 Howard Schultz, then marketing director of a small Seattle coffee-bean seller called Starbucks, went to Milan for a housewares trade show. Starbucks roasted and sold beans; it did not serve drinks. Walking the city, Schultz found about 1,700 espresso bars, which the company history describes as a third centre of Italian life after work and home.
His bosses were more interested in selling beans, so he left, visited hundreds more bars in Italy, and in spring 1986 opened his own espresso bar, Il Giornale, in Seattle. In 1987 he bought the six-store Starbucks chain for about $4 million and put its name on his cafés. The product he copied was not the coffee. It was the daily ritual around it.
Encyclopedia.com: Starbucks Corporation (International Directory of Company Histories)
How long does it take for a habit to form?
About 66 days on average for a simple daily behaviour, with a range from 18 days to an estimated 254, in the best-known field study.
Ninety-six volunteers each chose an eating, drinking or exercise behaviour to do once a day in the same context, such as after breakfast, and tracked how automatic it felt over 84 days; 82 gave enough data. Automaticity rose along a curve that flattens out, reaching its plateau after a mean of 66 days, with individual times ranging from 18 days to a projected 254. Missing one day did not set people back measurably, but many missed repetitions lowered the plateau. The limit: self-rated automaticity, a small sample, and behaviours people chose themselves.
Read the paper (free)Automaticity in Lally's data climbs quickly at first and then flattens towards a plateau. I'll model that with an exponential curve rising to a ceiling and use it to model retention. Assume:
- Automaticity grows as \( A(t) = A_{\max}\left(1 - e^{-kt}\right) \), and habit counts as "formed" at 95% of the ceiling (my assumption), so \( e^{-k \cdot 66} = 0.05 \Rightarrow k = \ln 20 / 66 \approx 0.045 \) per day.
- The daily chance of quitting starts at the app benchmark, \( h_0 = 0.107 \), and falls as the habit strengthens: \( h(t) = h_0 e^{-kt} \). This link between habit and churn is my modelling assumption.
- Cumulative hazard. \( H(t) = \int_0^t h_0 e^{-ks}ds = \frac{h_0}{k}\left(1 - e^{-kt}\right) \).
- Long-run limit. \( H(\infty) = h_0/k = 0.107/0.0455 = 2.36 \), calculated from the 4% day-30 retention and Lally's 66 days with the assumptions above.
- Share who stay for good. \( S(\infty) = e^{-2.36} \approx 9.4\% \).
- Interpretation. Same terrible first week, but once the behaviour is anchored to a daily cue, about one user in eleven becomes permanent instead of almost none.
See the data as a table
| Constant churn (app benchmark) (%) | Churn falls with habit (Lally curve) (%) | |
|---|---|---|
| Day 0 | 100 | 100 |
| Day 7 | 47.2 | 52.6 |
| Day 14 | 22.3 | 32.9 |
| Day 30 | 4.0 | 17.3 |
| Day 60 | 0.2 | 11.0 |
| Day 90 | 0 | 9.8 |
| Day 180 | 0 | 9.4 |
| Day 365 | 0 | 9.4 |
Source: My model: day-30 retention of 4% from AppsFlyer via CXOToday; habit rate from Lally et al. (66 days to plateau) · Both curves start with the same daily churn; the second lets churn fall as automaticity rises.
The sensitivity check
The sensitivity check is instructive:
- Habit forms twice as fast (\( k = 0.09 \)): the permanent share rises to \( e^{-1.18} \approx 31\% \).
- First-day churn halves (\( h_0 = 0.054 \)): it rises to \( e^{-1.18} \approx 31\% \) too.
The two levers are interchangeable in this model: make the early experience less leaky, or make repetition more frequent and consistent so the habit forms sooner. A product used once a week gets roughly one-seventh the repetitions of a daily one, so its habit forms far more slowly, if at all.
The chai-wala wins on frequency before he does anything clever.
Will your users form a habit before they churn?
What are the chai-wala's four retention levers?
A fixed cue in time, a fixed place, a remembered order and a month-end tab. Each has research behind it.
1. The cue: same time, every day
Habits are triggered by context, not by intention. David Neal, Wendy Wood and colleagues showed this with stale popcorn: people who habitually ate popcorn at the cinema ate about as much week-old popcorn as fresh when they were in a cinema, even though they rated it poorly, while in a meeting room the stale popcorn went largely uneaten.
Their habit was attached to the place, not the taste. The 11 am and 4 pm chai breaks are exactly that kind of cue: the time of day triggers the walk to the stall before any decision is made.
Two field experiments gave cinema-goers fresh or week-old popcorn. People with a strong popcorn-at-the-movies habit ate similar amounts of stale and fresh popcorn in a cinema, despite disliking the stale popcorn, while people without the habit ate much less stale popcorn. In a meeting room showing film clips, freshness mattered for everyone. In a second study, eating with the non-dominant hand broke the habit and reduced consumption of stale popcorn. The limit: a single US campus setting and a snack, so effect sizes for other products will differ.
Read the paper (free)2. The place: always on the way
The stall sits on the route between office and metro, bus stop and home. Habit research calls this a stable context, and Neal's hand-switching result shows its flip side: make the behaviour slightly harder to perform and the habit weakens.

3. Your order: remembered without asking
Ek cutting, kam cheeni, said once, becomes the default forever. Remembering the order removes a decision and signals that you are already part of something.
Joseph Nunes and Xavier Drèze measured the value of that head start with a car-wash loyalty card. Of 300 customers, those given a 10-stamp card with 2 stamps already filled completed it 34% of the time over nine months, against 19% for an 8-stamp card starting empty, though both needed eight more washes.
They called it the endowed progress effect. A remembered order is endowed progress: the relationship is visibly underway.
4. The tab: pay at month-end
Many regulars run a khata and settle it on payday. That separates the pleasure of the tea from the pain of paying, which behavioural economists call decoupling.
Drazen Prelec showed its force at MIT: in a sealed-bid auction for sold-out Boston Celtics tickets, students told to pay by credit card bid more than twice as much on average as those told to pay cash. A monthly tab is the street version of a credit card, and it also locks the customer in: you do not switch stalls while you owe this one.
| Chai-wala lever | Mechanism | Evidence | App or brand equivalent |
|---|---|---|---|
| 4 pm chai break | Time cue triggers the behaviour | Lally 2010: daily, same-context repetition builds automaticity in ~66 days | Attach to an existing daily routine; send at the user's own usual time, not yours |
| On the way | Stable context, minimal friction | Neal 2011: habits fire in their context; friction breaks them | Home-screen placement, saved login, one-tap reorder |
| Remembers your order | Default plus endowed progress | Nunes and Drèze 2006: 34% vs 19% completion | "Your usual" button, pre-filled cart, progress already started |
| Month-end tab | Decoupled payment and lock-in | Prelec: card bidders paid more than twice as much | Postpaid, autopay, subscriptions billed after use |
| ₹10–20 a glass, several times a day | High frequency, low stakes | More repetitions per week means faster habit formation (Lally) | Design a small, frequent use case before the big one |
What is a regular worth to the chai-wala?
A daily regular is worth several thousand rupees a year to a single stall (with the example inputs below), at near-zero acquisition cost, which is the unit economics most consumer apps never reach.
Assumptions (my estimates, since stall prices vary by city): ₹12 a glass, 2 glasses a day, 300 working days a year, a gross margin of 50% on milk, tea leaf, sugar and gas.
- Revenue per regular per year. With these example inputs, \( 12 \times 2 \times 300 = ₹7{,}200 \).
- Gross margin per year. With these example inputs, \( 0.5 \times 7{,}200 = ₹3{,}600 \).
- Lifetime value with low churn. With these example inputs, a regular has a 10% chance of drifting away in any year (estimate), so expected life is \( 1/0.1 = 10 \) years, so the undiscounted margin is about ₹36,000.
- Acquisition cost. Close to zero: footfall from the location and word of mouth.
That LTV-to-CAC ratio is effectively unbounded, which is why the business survives on ₹12 tickets.
Compare it with the payback arithmetic in CAC payback and LTV for Indian D2C, where a ₹500 acquisition cost needs months of repeat orders just to break even, or the costs in WhatsApp Business marketing, where every reminder message has a price. The cultural roots of why that glass is tea and not coffee are in Whyblog's why tea is more popular than coffee in India, and if you want to make the drink yourself, Foodblog explains how to brew masala chai.

Where does the comparison break?
In three places, and each one is fair to the app.
- Selection. We only notice the regulars. Passers-by who tried the stall once and never returned are invisible, so the stall's "retention" is measured on survivors. An honest comparison is the app's day-180 users against the stall's regulars, and the habit model above suggests those numbers are closer than they look.
- Category. Tea is a daily physiological habit with caffeine and a social break attached. A shopping app competes with a need that arises a few times a month, so it cannot reach daily repetition honestly.
- Choice. Habit cuts both ways. Neal's stale popcorn shows customers sticking with a product they rate poorly. Retention built on habit alone is fragile the moment the context changes: an office move, a new route, a closed stall.
So, does the chai-wala really retain better than your app?
On the data, yes: Indian shopping apps keep about 4% of users at day 30, implying a daily churn of about 0.107, while a tea stall turns daily repetition in a fixed context into automatic behaviour, which Lally's 66-day curve shows is what makes churn fall. In a habit model, the same bad first month leaves about 9.4% of users permanent instead of almost none, and either halving early churn or doubling repetition frequency lifts that to about 31%.
The chai-wala's four levers, cue, place, remembered order and deferred payment, each have an app equivalent. The one apps cannot copy is daily need, so the practical fix is to find the most frequent honest use case and anchor the product to it.
What's new
- India shopping-app installs grew 85% while day-30 retention fell 41% in two quarters CXOToday, Sept 2026
- Extreme heat and Assam floods cut India's May 2024 tea output by 30%; per capita consumption is about 840 g a year Drishti IAS, Sept 2025
- New research links habit tendencies to other automatic behaviours in 883 UK adults PsyPost, May 2025
What if next?
- Bollywood trailers are A/B tests Studios test their hooks on crores of viewers before the weekend that matters.
- The wedding card is the perfect piece of direct mail Personalised, physical, delivered by hand, and almost never ignored.
- The street vendor's shout beats your headline Five words, repeated, tuned live to the crowd.
Questions people ask
What is a good day-30 retention rate for an app?
It varies widely by category. AppsFlyer data on Indian shopping apps put day-30 Android retention at about 4% through 2025, and it fell further in 2026. Daily-use categories such as messaging and payments retain far better.
How long does it take to form a habit?
Lally and colleagues found a mean of 66 days for a simple daily behaviour done in the same context, with a range from 18 days to a projected 254. Missing a single day did not reset progress.
Why do loyalty cards with free stamps work?
Because of the endowed progress effect. In Nunes and Drèze's car-wash study, cards with 2 of 10 stamps pre-filled were completed 34% of the time against 19% for empty 8-stamp cards, though both required the same eight purchases.
Does paying later make people spend more?
Often, yes. In Prelec and Simester's MIT auction, students paying by credit card bid more than twice as much for Celtics tickets as those paying cash, because deferred payment softens the pain of paying.
Sources
- CXOToday: AppsFlyer report on India shopping apps (Sept 2026; source for ~4% day-30 retention, the 41% fall and 85% install growth)
- Lally et al., European Journal of Social Psychology: How are habits formed (2010; source for the study design and automaticity curve)
- BPS Research Digest: How to form a habit (source for the 66-day mean, the 18–254 day range and the missed-day result)
- Neal et al., Personality and Social Psychology Bulletin: The pull of the past (2011; the stale popcorn experiments)
- USC News: Habit makes bad food too easy to swallow (Sept 2011; source for the week-old popcorn, meeting-room and hand-switching details)
- Coglode: Endowed progress effect (Nunes and Drèze 2006) (source for the 300 customers and 34% vs 19% completion)
- MIT Spectrum: The psychology of spending (source for the Celtics ticket auction and the more-than-double credit card bids)
- Drishti IAS: India's tea industry (Sept 2025; source for 840 g per capita and 80% domestic use)
- Fader and Hardie, Journal of Interactive Marketing: How to project customer retention (2007; source for retention rising with tenure through heterogeneity)
- Encyclopedia.com: Starbucks Corporation, International Directory of Company Histories (source for the story: Schultz's 1983 Milan trip, about 1,700 espresso bars, Il Giornale in 1986 and the $4 million purchase in 1987)

