Scrappy growth tactics and analytics foundations for consumer startups

Executive overview

Most growth efforts fail because teams track measurements instead of insights, and bet on unproven features instead of fixing the constraints they already know about. Crystal Widjaja built growth at Gojek — Southeast Asia's largest super app — by combining statistical rigour with do-things-that-don't-scale scrappiness.

The core framework: map the physics of your market, product, model, and channels first. Then find the single biggest constraint. Then fix it with the smallest possible intervention.

Copy, instrumentation, and leverage point identification beat new features almost every time.

Why super apps work in Asia but not the US

• In Southeast Asia, conglomerates are trusted to own multiple life domains — mall, apartment, school.
• Phone storage is scarce; users will delete apps before photos, so one app must do everything.
• Most households skipped desktop computing entirely — the phone is the primary computer.
• In the US, abundant storage and cloud backup mean users tolerate many single-purpose apps.

Scrappy testing at small scale

• Run experiments even with 30 users — trend direction holds; only precision improves at scale.
• Validate features with Wizard of Oz methods: WhatsApp groups, screenshots overlaid with Figma mocks, Typeform surveys, manual backend operations.
• Gojek tested subscriptions by having drivers collect cash on rides and texting an intern who manually applied vouchers.
• If you can't design an experiment, the idea isn't ready — untested hypotheses are not actionable data.
• Do things that don't scale to validate the value prop; then find what scales intuitively.

Growth physics framework

• Step 1 — map the physics: identify your market, product, model, and channels. List every lever you actually have, including ones you've never used.
• Step 2 — fit loops to physics: check whether your growth loops and funnels work within the existing constraints. Avoid changing multiple variables at once.
• Gojek discovered drivers were an under-used acquisition channel: a script checked whether a matched customer had ever used GoPay; if not, it messaged the driver to pitch a top-up for a cash bonus. That single lever drove 60% of GoPay acquisition.
• Change one thing at a time. The universe isn't changing fast enough to justify betting on five variables simultaneously.

Where growth investments pay off

• Copy first: if users land and do nothing, they haven't experienced the product — change the copy to name the pain, not the solution.
• Shorten the longest conversion path: find the most painful user journey to conversion and cut steps. Kumu's search speed-up moved conversion from 60% to 90%.
• Solve the step before conversion: users don't convert because of missing trust, momentum, or context — not because the product is wrong. GoFood doubled purchase rates from new merchants by showing which friends had ordered there.
• Pause buttons over win-back campaigns: AB InBev's top cancellation reason was "too much beer." Adding a pause option recovered churn that reactivation emails could never fix.
• Stack sure wins against big bets. Run an experiment on every speculative bet.

Retention benchmarks

• Free product: 60% week-one retention, then it must flatten. Gojek hit 60–70% in early days.
• Paid product: 20–30% week-one retention.
• Friends and family stage: must be ~80% — if people who care about you won't return, the product isn't solving the job.
• Beware early international cohorts: high week-one numbers can mean you've already exhausted every possible subscriber, not that growth is accelerating.

Why analytics efforts fail

• Measurements are not insights. "Power users book 4x more" is an observation. An insight answers why and tells you how to act differently.
• Insight structure: observed fact + contextual properties + hypothesis + test = causal representation.
• Good instrumentation attaches rich properties to every event — not just the event name. Bad instrumentation has hundreds of events each with zero or one property.
• Example: a map-load event should include number of drivers visible, pickup city, latitude/longitude, surge status, current minimum fare, and voucher presence. Without those, you can't diagnose why users didn't book.
• Track user journeys and experiences, not KPIs. If seeing a metric doesn't change what you do, it's entertainment, not data.

Analytics tooling by stage

• Early / single data warehouse: Google Data Studio (free) or Metabase (open source).
• In-app mobile event tracking: CleverTap.
• Larger scale analytics: Amplitude.
• Data piping / ETL reduction: Segment.
• Experimentation: Epo (automates the decision-making flow).

Hiring and structuring a growth team

• Don't hire a growth person and expect them to figure out what to work on. Know the gaps first; hire someone to close a specific one.
• First growth hire profile: statistically intuitive, knows sampling, understands selection bias. Willing to take an intro stats course if needed.
• A growth hire who measures things wrong and optimises the wrong area is worse than no growth hire.
• Interview method: give a take-home case study (four hours over five days). Look for random sampling, experiment design, first-principles reasoning, and willingness to Google and cite sources.
• Growth as a cleanup crew works when there is undeniable product-market fit and the core team is overwhelmed with foundational feature work.
• At Gojek, growth PMs embedded in product teams, synthesised gaps, and eventually became owners of specific product areas.
• Avoid tool sprawl: a six-month integration project produces no growth. Bias toward quick, hacky interventions first.

Measuring emotional products

• For transactional products (Gojek), success is a completed job — easy to instrument.
• For social products (Kumu), success is a felt connection — harder to observe. Use artificial friction as a signal: users who fill out a detailed questionnaire to join a community genuinely want that community. Users who drop off may want entertainment instead.
• Design hand-raisers so the product reveals which job users are actually trying to do.