Gojek Consumer Ads

The Problem: We weren't invited

In late 2022, GoFood was undergoing a complete redesign. This was a massive initiative - GoFood generated the majority of Gojek's revenue. But here's the problem: the ads team, responsible for $11.3 million in revenue, wasn't included in the redesign.

We were literally going to be an afterthought.

"I couldn't let ads become a band-aid on a redesigned product."

The existing state was broken

• Banner ads occupied prime real estate but had low engagement
• Overload of information and busy design compromised readability and perceived value
• Users didn't consider carousel banners to be useful and often skipped them
• Too many banners, often considered irrelevant and not personalised, leading to banner blindness

The Fight: Writing our way in

So my fellow designer Shreya (my counterpart on merchant tooling) and I wrote a strategic document - 'Principles for embedding APP inside GoFood ads.' We didn't wait for permission. We wrote it, aligned it with Maji, our design manager, and presented it directly to Julee, the GoFood VP.

Three strategic principles

The document worked.

We got embedded in the redesign with clear responsibilities through the DARCI model. We secured 70% capacity allocation for our designers. This wasn't given to us - we fought for it by showing the revenue risk.

"Because we got involved early, ads weren't retrofitted - they were native to the new experience. We protected the $16 million revenue stream and set a precedent that ads must be considered in every major product decision."

Managing the chaos: Stakeholder alignment

Getting embedded was just the start. I had to align with business about relevancy metrics, with designers about which touchpoints we'd own. Each conversation required different framing - revenue for business, user experience for product, technical feasibility for engineering.

Aligning goals across teams

The Solution: Masthead Ads

We designed a premium ad format that would solve the banner blindness problem while respecting user experience.

Design principles for Masthead

Early prototypes to make the case

I quickly mocked up prototypes to share with leaders to make the case for a new ad inventory. The goal was to show, not tell.

The hard part: Making it scalable

As I iterated on UI, challenges emerged:

• Merchants had near-infinite requirements
• Guidelines had to work in both Indonesia & Vietnam
• Had to preserve the "wow factor" while being systematic

Iterations started to get messy, as I realised we needed a system to capture all merchant use cases

Scoping merchant & brand use cases

Worked with PM, biz and ad sales to consolidate Merchant / Brand jobs-to-be-done and define:

• Slot structure for different use cases
• CPM model that worked for various advertiser sizes
• Use case taxonomy (brand awareness, promotions, new launches, etc.)

Creating visual & copy guidelines

Worked with Creative Design + UXW on copy and design rules that could:

• Encompass all merchant/brand JTBDs
• Stay cohesive with Gojek branding
• Be internationalisable (Indonesia & Vietnam)
• Maintain quality standards at scale

Rollout & governance

Rollout meant debugging, working with devs and biz, handling device-specific edge cases. But the real work started after launch.

We debugged on test devices and alpha builds to ensure visbility across devices and screen sizes. The real challenge was a consistent view experience across Android, which had a lot of manufacturers with different specs for notch, cutouts etc.

Masthead ad governance

Merchants would come back to us with feedback and requests. I had to create a workflow to implement changes in the ad sales & design process:

• Process to take in feedback from merchants and ad sales
• Review mechanism with design, product, and business
• Update and version Masthead guidelines
• Communicate changes back to all stakeholders

Impact in GoFood

Most successful ad inventory - Masthead became the most successful ad format and was rolled out across the entire Gojek consumer app, not just GoFood.

What I wanted to do next

Success doesn't mean we were done. I had a roadmap of improvements:

UI only goes so far

Most gains came from personalisation and targeting, not just visual design. I explored how Masthead could integrate into offline and social media channels to create a cohesive brand experience.

Technical constraints we hit

• Limited to 3rd party PNGs due to tech maturity constraints - harder to customise and A/B test variations
• Image only - no support for voiceover or video
• Accessibility gaps - Carousel is an inaccessible format. Wanted to add accessibility in ads creation tools and at design system level

What did I learn?

• How to present to big decision makers and get people on board with a vision
• When to step up - sometimes you don't wait for permission, you make the case
• When to bring others along - collaboration across teams is what makes systemic change stick
• Strategic thinking matters - The document we wrote wasn't just design work, it was business strategy

Banner CTR Experiment: 60% Uplift

Using automation tools and code-based methods to improve ad performance for GoFood merchants

What's the problem?

We make about $100k+ USD per month from thousands of self-serve merchants who run banner ads on GoFood.

Self-serve banners creation (legacy process) were inflexible PNGs super-imposed on top of each other that merchants would choose. We'd noticed less customer engagement with self-serve banners vs. sales-assisted banners (more customizable). As a result, merchants would churn and not renew.

Sales-assisted banners (left) vs Self-serve banners (right)

Why didn't users click on banner ads?

Comparing sales-assisted vs self-serve banners, we identified key differences:

• Image quality seen as proxy for food quality - Self-serve banners had lower quality images
• Promo information - Users look at details on discounts/cashback etc
• Credibility - Users look for signals like reviews, ratings, to make a decision to order something new

User behaviour creates a vicious cycle

• Discovery stage: User views self-serve banner on the GoFood app
• Consideration stage: Is the banner relevant to what I want to eat? Does it contain information I care about? (promo/discount/rating)
• Decision stage: User ignores banner due to lack of relevance (targeting/algo issue) or lack of information (design issue)
• Low return on ad spend: Merchant spends money to run the banner but doesn't see a return. Merchant pulls the ad.
• Vicious cycle: Fewer merchants run ads, meaning fewer users are shown banners on the app and banners are less relevant

User Research

Card sorting with customer panel

We wanted to get a prioritisation of what customers cared about (carried out in collaboration with UXR team). Key findings:

• Promo details were the #1 factor in click decisions
• Ratings/reviews provided social proof and credibility
• Image quality was seen as a proxy for food quality
• Restaurant name helped with recognition and trust

Exploring solutions and building the case

I worked with PM and the data team to brainstorm and explore numerous ideas to solve the low CTR problem. This included competitor research with Grab and Foodpanda to understand what was working in the market.

The vision: Contextual & personalized banners

My hypothesis was that by making banners contextual to user preferences and behaviour, we could significantly improve ad relevance and CTR. The ideal future state would include:

• Personalized SKU discounts - Banners showing discounts for categories users have ordered before
• Social proof - User-generated content and popularity signals ("2,000+ people ordered this today") to tap into the hive mind effect
• Behavioral targeting - Reducing friction in decision-making by showing relevant offers at the right time

"This was the idealized vision - what we could eventually achieve. Some ideas required tech and data infrastructure we didn't have yet. But I presented them as possibilities to plant seeds for the future, while focusing the immediate experiment on what we could realistically build and test."

After presenting these ideas to product and design leadership, we got the green light to move forward. With bandwidth and alignment secured, we scoped an experiment we could execute immediately while keeping the longer-term vision in mind.

We defined our hypothesis and end state

Hypothesis: If our banners contain the data points that are relevant to our customers' purchase intent then we will see an increase in click through rate on banners

Turning qual insight into measurable metrics

Key questions we had to answer:

• How do you turn qual insight into a measurable metric?
• How do you draw a relationship between a point of information on the banner, versus the actual user behaviour that it causes?

Multivariate testing approach chosen: Same core mechanism as A/B testing, but compares a higher number of variables, and reveals more information about how these variables interact with one another.

Experiment treatments

Worked with PM, Data Science and engineering to define experiment hypothesis and parameters. We designed 6 treatments to test different combinations:

All 6 treatments tested in the multivariate experiment

How do we put this into production?

For the experiment to be statistically significant, there must be a minimum number of daily users it had to reach.

Scale: Total: 6 banners × 2 merchants × 10 cities = 120 banners

Process:

1. Worked with PM and Biz team to get data points
2. Created a spreadsheet and broke down image parameters (merchant name, rating, image URL) for each merchant × treatment
3. Evaluated and used a plugin to auto-generate the banners
4. Transformed the data into a CSV that was parse-able by the plugin to generate the banners

This automation approach saved weeks of manual design work and ensured consistency across all 120 banner variations.

Experiment Results

Treatment 2 and Treatment 5 emerged as winners of the experiment. This meant that our hypothesis of showcasing discunt amount details and information hierarchy improvements were validated.

Counter-intuitive insight

We had some hypotheses around this. Rating affected CTR only when the rating was good. This led to some ethical questions:

• Do we show rating if rating is low?
• Do we optimise for click through rate or user satisfaction?
• What are other data points we can show that provide social proof and credibility apart from rating?

New challenge - developer collaboration

Prior merchant ad solution was built as an MVP with limited functionality. Now the solution had to change to include additional data points as per experiment results.

It was becoming a bit complex to communicate everything as a PRD. So I thought: what can I do as a designer to shorten this process?

Solution: Code-based proof of concept

I created a code-based proof of concept for the engineering team, using internal APIs to demonstrate how the component could work and collaborated as they built it in terms of high-fidelity specs and visual QA.

Impact

• Project and methods received great feedback from cross-functional peers
• Presented case study at design all hands to team & leadership
• New ad format scaled up to 100% merchants and customers
• Increased merchant signup and retention contributed to 92% increase in advertising revenue in 2024

Learnings and next steps

"Tools complement, not substitute well-defined hypothesis and rigorous thinking. In setting up an experiment, you need to account for not just what you want to find, but how you will go about finding it. The work is never done."

Addressing patchy merchant image quality

Even though we increased merchant participation, we created a new problem: image quality was still inconsistent. How do we curate at scale? We still haven't cracked this.

We had some vision for how to solve it. We realized we needed to give merchants feedback in their campaign reports - showing them when poor image quality was leading to bad engagement. That became the next problem we were supposed to solve... but it hasn't been solved since I left Gojek!

Creating a better framework for personalisation & performance

My vision was for banner ads to no longer be PNGs but UI overlays on high-quality merchant images, where copy could be tailored to user personas:

• Social proof for people who need validation
• Discounts for discount hunters
• Ratings for foodies

It would be dynamic and more in line with state-of-the-art advertising that Meta and Google could offer. Of course, we couldn't get there during my time, but that's the vision I drove in the GoFood redesign - planting seeds for the future.