How to scrappily hire for, measure, and unlock growth | Crystal Widjaja, Gojek and Kumu

Widjaja values watching customers use the actual experience over trusting an idea that only looks good in a model. A small practical test can expose a mistaken assumption. This selection does not use her statistical claim that tiny samples preserve the same trends; small samples still leave uncertainty.

Rather than jump from signup to retention, she asks what a person must do or believe immediately before converting. In her food example, the barrier was trust in a new merchant. Identify the local obstacle instead of treating a broad outcome as a specific problem.

Widjaja says retention is not specific enough by itself. A more concrete frame helps a team generate useful interventions. Translate the broad metric into a particular user, moment and barrier before choosing a feature to build.

She asks where people eventually convert and which route is painfully long. Her Kumu search example focuses on an existing behavior made slow by the product. People already attempting something can provide a clearer improvement opportunity than an entirely new speculative feature.

Her advice is to begin with the existing product and the largest constraint rather than imagine the whole system must change. Fixing one point can improve a working path. The lesson is prioritization grounded in observed behavior, not a promise that every bottleneck has a simple fix.

A database observation, such as power users booking more, does not yet explain what to change. Widjaja looks for context that supports a specific hypothesis about why a behavior happens. Interesting numbers only become useful when they affect a decision and can be tested.

She adds properties to events so behavior can be segmented, for example by customer type and basket context. That makes it possible to form a more precise hypothesis. Collect only appropriate data with needed permission, and avoid treating a correlation as established causation.