Building GoChat across five countries.

The roadmap was full of feature requests, but the platform underneath them had a growing cost and reliability problem. The existing third-party provider was charging roughly $300–400K a month, with extra integration work and charges every time we added a capability. At Gojek's scale, those economics mattered to the viability of the roadmap.

Uptime sat around 97%, chat windows could take four to five seconds to open, and messages between customers and drivers sometimes failed to arrive outright.

Engineering saw the case for building in-house. Product and ops weren't initially convinced the migration justified the investment and disruption risk, so I needed to show why it should take priority alongside the features everyone wanted to ship. I made that case with cost and reliability data, including what the failures meant for customers. We migrated without downtime, cutting vendor costs by an estimated $4–5M a year and giving the team control over the product.

There was a second cost hiding in how drivers and customers coordinated. In Indonesia, drivers regularly swapped SIM cards to chase free SMS and call credits, and each new number meant re-verifying by OTP, a per-login cost to Gojek at scale.

Moving everyday coordination into in-app chat reduced drivers' reliance on those SMS and call offers. With less reason to swap SIMs, roughly 80% of drivers no longer needed the repeated OTP re-logins. This was on top of the $4–5M vendor saving.

I hired one PM to own the in-house platform and order-chat experience, including canned messages, translation, stickers and image sharing. A second PM focused on everything beyond a single order: customer support chat, marketing communications through chat (including Jeklin), and customer-to-customer chat for peer-to-peer payments, splitting orders, and sending and receiving parcels.

Most conversations were about confirming an order or letting someone know the driver had arrived. Canned messages cut the need to type those out every time: respectful greetings like "pak" and "mbak" in Indonesia, plus practical lines like "I'm at the pickup," "my order is confirmed" and "I'm at the door waiting." Later iterations layered in ML-personalised suggestions based on each customer's most-used messages, and adoption climbed with every iteration, reaching around 45%.

Around 17% of chat users were tourists, concentrated in Bali, so in-chat translation addressed a language barrier in those bookings. The longer-term aim was to push typing effort down toward zero.

We extended chat into group conversations and peer-to-peer payments, and into new markets: Vietnam, Thailand and Singapore. Each launch needed attention to local constraints. In Vietnam, that meant working directly with government officials, because rollout couldn't start without their approval.

We also explored how chat could support communication beyond a single order. Snippets used a Stories-style format; Shuffle Cards used a familiar swipe interaction that let customers signal which kinds of communication they wanted more or less of.

Chat adoption in Indonesia was around 90%, while roughly 56% of GoFood merchants had no ratings at all. Customers were already comfortable engaging through chat, and a lot of merchants needed a way to build visibility and engagement. I developed the commercial strategy across chat and the wider communications platform around that gap, then, as Head of Product, put it into practice by helping secure a highly contested central Home placement for chat, introducing internal notification billing and communications governance, and building out Jeklin for two-way engagement. These were the foundations for merchant advertising on the platform; the eventual third-party advertising business launched later.

Jeklin was a conversational assistant built for two-way engagement, letting customers signal their own communication preferences. I launched it alongside Snippets and Shuffle Cards as different ways for customers to engage with communications and tell us what they wanted more or less of.

Jeklin achieved roughly 2x the click-through rate and up to 5x the conversion of push and inbox messaging, where conversion meant a customer taking a real action, like making a booking on Gojek or setting up and topping up GoPay. Across the wider communications work, push CTR separately rose from about 1% to 2.5–4% within a quarter. Jeklin later became Gojek's named assistant in Indonesia. My part was the launch.