Growth in the Physics of a Market: Lessons from Gojek, Kumu and Southeast Asia's Super Apps

Widjaja grew up in San Jose, studied political science, and took an investment banking research job, where she taught herself MySQL to tame a 130,000-row spreadsheet. Banking did not embrace the idea, so she left and applied her pattern-matching to Southeast Asia, where her family is from. She cold-emailed Gojek's HR and offered to move to Indonesia, and they took the bet.

Over five years she built the data team from scratch, then fraud and risk, then performance marketing, and finally the growth team. Gojek, now part of GoTo, served about 170 million users across Southeast Asia with more than 20 services, from rides and food to bill pay and movie tickets. Kumu, where she is chief product officer, is a social super app focused on the Philippines that combines feeds, audio, video and multi-seat rooms.

She joined Gojek because the value proposition was clear: Indonesian traffic is so bad that a motorcycle taxi beats a car, and food delivery spares people a two-hour trip. The drivers and restaurants already existed but were hard to connect, and she thought the mobile-app channel was a very solvable problem. The company rented a stadium to onboard 60,000 drivers within a couple of weeks, and when she arrived the company was in a house doing 4,000 orders a day.

Asked why the US has no super app, Widjaja points first to culture. In Southeast Asia people have grown up trusting conglomerates that own the mall, the apartment building and the school, while in America people worry about whether Google already knows too much.

The second reason is device economics. Many people leapfrogged the computer and have only a phone, so when storage fills up they delete an app rather than a photo of their kid. To survive, an app has to be part of a super app, and Lenny adds that such an app only needs to be good enough at everything to get the job done.

Widjaja says the bread and butter at Gojek was doing slightly crazy things that validate a point. To test a subscription, the team added 100 drivers to a WhatsApp group and told them to pitch it to riders for $10; interns then manually granted the vouchers and deducted the money from the driver's balance. It was a Wizard of Oz test that needed no build and validated conversion rates.

For a new onboarding screen, a designer overlaid the proposed flow on a screenshot of the current screen and it went out as an in-app message. Quizzes ran through Typeform, and the team later made an in-app webpage so features could be tested without waiting for a mobile release. The guiding question was what experience they wanted to create and how to manifest it as fast as possible.

Her advice to startups is that an idea without a testable hypothesis is almost useless, because consumers are not always rational. Even with a sample of 30 you should run experiments, since a larger sample changes the precision of a trend but not the trend itself. Better to have 30 data points than zero.

People want to jump from zero to retention, Widjaja says, but users decide based on the events that precede it. For GoFood the barrier was trust in a new restaurant, so the team used Facebook connect data to show food that a user's friends had bought. Users were twice as likely to order from a brand-new restaurant, which raised GMV by solving trust first.

She clarifies that she does not mean retention is unimportant, only that it is too vague a target. She uses an exercise from *Made to Stick*: asking for everything orange in the world is hard, while everything orange on a construction site is concrete and actionable. Retention gains usually come from the step before conversion, where friction or poor copy keeps people from the next action.

Founders often grasp at straws with brand-new features that have no Wizard of Oz test and no data showing users want them. If many people land and do nothing, she suspects copy that should speak to the pain point rather than the solution. Otherwise she looks for the most painfully long conversion path; at Kumu, speeding up the underused, slow friend search raised conversion from 60% to 90% in a few weeks, and she layers such sure wins with riskier bets.

Widjaja offers concrete benchmarks. At scale, a free product should retain at least 60% at its natural frequency period, and a paid product perhaps 20 to 30%. In the earliest days with friends and family, it should be close to 80%, because if people who care about you will not use it, it will not solve the job for anyone else.

She thinks of it in cohorts: week one should land around 60% and then flatten within two to three periods. Gojek showed 60 to 70% in its early days, which told her it would be fine. She is wary of misleading early spikes, citing international launches where only users with accepted payment cards can subscribe, so the first-week surge just pulls forward everyone who could join.

Widjaja's process starts with constraints, the physics of the market, product, model and channels. For Gojek that meant Indonesian consumers and drivers, a mobile app, a per-order fee, and channels including push notifications, Facebook ads and the real world, where a driver in a jacket was effectively a moving advertisement that drove word of mouth in each new city. Step two checks whether your loops, funnels and inputs fit these physics, changing one small thing at a time.

The team then noticed an underused lever: drivers. To grow the GoPay e-wallet, they built a small service that checked if a rider had ever topped up and messaged the driver to offer extra money for taking the rider's cash and depositing it into their wallet. Drivers made great salespeople because they had a captive customer, and the feature made up 60% of GoPay acquisition once released.

She advises starting from what already works and attacking the biggest constraint, since you are not a wizard who can move the universe. At Kumu the job is harder to measure than a transaction, so the team adds deliberate friction, such as questionnaires to join a community, to tell who really wants friendship rather than entertainment.

Widjaja says that early growth unlocks were often copy. Because many people did not know what a digital wallet was, Gojek put users' virtual account numbers on a picture of a credit card, so they understood it worked like a debit account and could top up at an ATM. Top-ups rose just from sending that image.

For a direct-to-consumer subscription company she advises, the top cancellation reason was fear of commitment, yet the app only let users cancel or resume. Adding a pause button gave a temporary answer to a temporary problem and reduced churn that was hard to win back. Lenny notes that Airbnb's snooze feature for listings worked the same way.

Widjaja says her popular Reforge post on why most analytics efforts fail came from frustration, and that the main culprit is instrumentation. Teams track metrics for their OKRs but do not use data to find insights, treating it like entertainment rather than news. Real news changes what you do.

A measurement is an observation, such as power users making four times more bookings, and it lacks context. An insight explains why and tells you how to act, such as power users being more likely to use free shipping on high-GMV baskets, which suggests a marketing change. This requires putting properties on events so you can segment behavior, form hypotheses and test them.

She can tell a weak team from its instrumentation spec: many events with one or no properties. For the Gojek map event she would capture the number of drivers visible, pickup location, city, coordinates, surge pricing, minimum fare and voucher code. That reveals, for example, that users seeing two drivers convert less than those seeing five, and where that happens.

For tooling, she suggests Google Data Studio or Metabase for early teams, CleverTap for in-app event tracking, Amplitude and Segment as teams grow, and Eppo for experimentation. She advises thinking through why someone would or would not press each button and mapping the experience.

At Gojek growth emerged to fill gaps the core product team was too busy to cover, like choosing SMS providers for one-time passwords or telling drivers when a customer was brand new, since new users did not know whether to wait or walk to the driver. Growth product managers were later embedded into teams and became PMs owning parts of the product. She believes the first growth hire should work on a known gap rather than model everything from scratch.

A separate growth team can work if product-market fit is so strong that the core team is scrambling, acting as the cleanup crew. Her team was heavy on statistics graduates, because working at that scale is a numbers game. Her hiring bar is intuitive statistics, including sampling and selection bias, since measuring the wrong thing is the worst outcome.

She cares less about experience than whether a candidate will focus on the right opportunity over the flashiest. Candidates get a take-home case on experiment design with about four hours of work over five days, and she likes hearing that they had to Google things. Gojek's first real growth experiment was a Python script using Twilio to text drivers about low acceptance rates, which lifted acceptance by 2%.

Widjaja cofounded the nonprofit Generation Girl with former Gojek colleagues after repeatedly hearing comments about not looking like an engineer. It offers free classes for girls aged 12 to 17 and college students, and is happy if girls conclude they do not like engineering, since that is agency without social pressure. Several thousand students have gone through its clubs and classes, and it works with Indonesian tech companies to pair students with engineers.

Its newest initiative, Class, is a free site for teachers, built through the MIT Solve program, to help them teach newer concepts. Because every teacher reaches thousands of students, she sees this as high leverage. She says below 18% of computer-science graduates are women, partly because girls get fewer introductory STEM classes and feel behind, and she cites Carnegie Mellon introductory classes that helped women graduate at similar rates.