Tiktok Swiggy Uber Case Studies

The previous post showed how Netflix solved a billion-dollar churn problem without asking users for anything. Three more products faced variations of the same challenge -- users arrived, but did not do what the product needed them to do. Each solution came from the same place: find the specific friction causing the drop-off, and remove it.

TikTok: A 900% Lift From One UI Nudge

When TikTok launched, the dominant video experience on mobile was YouTube: horizontal video, tap to play, related videos in a list below. TikTok proposed something different -- vertical short-form video consumed by swiping up.

The problem was that this behavior did not exist. Users had no intuition for swiping up to get more content. They would watch one video, rewatch the same video, and leave.

Product managers observed the pattern in new user sessions. The fix was a nudge: after a video plays, the bottom of the next video peeks up just enough to signal that more content is one gesture away.

The engineering decisions behind that small animation were not trivial:

• Does audio from the next video begin playing while the previous one is still visible?
• Does the pane snap back if the user partially swipes but does not complete the gesture?
• How much of the next video is revealed before committing to the transition?

Each of these touches user experience in ways that can irritate or delight depending on execution.

The result: average time spent per user per day in China went from 5 minutes to 52 minutes. A 900% increase in daily engagement from a UI nudge that a user barely notices.

That shift made TikTok a global phenomenon. Instagram Reels and YouTube Shorts both came after, copying the vertical format and the swipe behavior that TikTok normalized.

Swiggy: 70% of App Opens End Without an Order

Approximately 70% of app opens on food delivery platforms do not result in a transaction.

If a crore of users open Swiggy on a given day, 70 lakh of them scroll for 10 to 40 minutes, get overwhelmed by choices, and close the app without ordering anything. Every one of those sessions is a cost without revenue -- infrastructure, data processing, server load, all consumed without a rupee earned.

On top of that, the marketing spend that brought those users back -- notifications, ads, re-engagement campaigns -- was wasted on a session that produced nothing.

The metric product managers track here is app opens to transactions completed. Features like "What to Eat" on Swiggy exist entirely to close that gap: surface the right options before the user has to search, and more sessions end with an order placed.

The window to convert a session is short. A user who has already spent 30 minutes deciding and gone nowhere has mentally switched to something else. They are not coming back.

Uber: Cutting Abandonment From 80% to 40%

Uber faced the most concrete version of this problem. 80% of users who opened the app to book a ride did not complete a booking.

The cause was not a shortage of drivers. It was the destination input field.

Ten to fifteen years ago, mapping data in India was imprecise. Typing a destination returned multiple results with similar names, none clearly distinct. Users would start typing, fail to identify the right pin, get frustrated, and give up -- then flag down a cab on the street. Uber was losing the majority of its potential rides to a bad text input experience.

The solution was suggested destinations. Uber analyzed historical trip data for each pickup location: where users from that spot typically went, filtered by time of day and day of week. From home on a weekday morning, the top suggestion would be the office. Friday evening, a frequent restaurant or a friend's area.

No typing required. One tap.

After launch, the 80% non-booking rate fell to 40%. Roughly half the users who had been abandoning now completed a booking. The revenue impact was immediate.

ExpandPM and developer collaboration: the pattern behind every product improvement in this series

A/B Testing: When the Right Answer Is Not Obvious

The Uber case introduced a second concept: A/B testing.

The team did not know in advance which ranking algorithm for suggested destinations would produce the highest booking rate. They built multiple algorithms -- each with a different hypothesis about what predicts where a user wants to go -- and served them to different user segments simultaneously.

The winning algorithm became the default. The others were dropped.

This is the value of A/B testing: it decouples confidence from certainty. You do not need to know in advance which solution is better. You build the infrastructure to test both, define the metric that matters, and let user behavior tell you.

In some cases, the best algorithm for Mumbai performs differently in Delhi. A/B testing at geographic granularity lets teams serve the right experience for each context rather than forcing a single global answer onto everyone.

For developers, the practical implication is architectural. A feature that will be A/B tested needs configuration-driven behavior from the start. A hardcoded single implementation is a rewrite every time you want to test a variant.

Building for experimentation is not over-engineering. It is acknowledging that user behavior is the final answer -- and the code needs to accommodate that.

The next post covers what happens when product decisions affect more than the obvious user -- and why mapping every stakeholder before you build is what separates products that scale from ones that break unexpectedly.

The Essentials

1. TikTok's swipe-up nudge lifted daily engagement from 5 to 52 minutes -- a 900% increase from a single behavioral insight applied through careful UI engineering.
2. 70% of food delivery sessions produce no transaction. The metric is app opens to orders placed. Features that reduce decision time directly improve this ratio.
3. Uber cut abandonment from 80% to 40% by suggesting the most probable destination upfront. Bad input UX was costing the majority of potential rides.
4. A/B testing is the tool for navigating uncertainty. When you do not know which solution is better, test both, measure what matters, and let user behavior decide. Build for configurability from the start.

Further Reading and Watching

• What is A/B Testing? (Optimizely) -- clear overview of how A/B testing works and when to use it
• How Uber Builds its Products (Uber Engineering Blog) -- behind-the-scenes look at the data and engineering decisions powering features like suggested destinations
• How to Build Habit-Forming Products (Nir Eyal, YouTube) -- the behavioral loop behind TikTok's engagement and why small nudges compound into massive retention numbers