Helping sellers find out why sales drop, and fix it in one click.
Sellers had access to performance data, but struggled to understand what needed attention. I designed a feature that turned product signals into actionable recommendations, helping sellers address issues affecting sales and refunds.
BUSINESS GOALS
Reduce avoidable refunds and help sellers improve product performance
ROLE
Product Designer — decision logic, concept exploration, validation, MVP scoping, and UI/UX design
KEY RESULTS
−10% out-of-stock refunds within 6 months of launch
Problem
Sellers had the data, but not the answers
Based on past research, we know that sellers could track product performance, but struggled to identify why sales declined or which issues needed attention. In the meantime, they tend to forget to update stock or store hours leading to avoidable refund issues. This created two connected problems:
Seller side
Identifying issues like low stock, outdated prices, or poor product performance wasn't straightforward, making it harder to know what to fix first.
Bukalapak side
7.4% of transactions were refunded, about Rp 68B (~US$4M) in monthly GMV. 49% of those refunds were caused by avoidable seller issues.
The opportunity wasn't to build another analytics dashboard, but to help sellers understand what was wrong and take action.
Strategy
Sellers didn't need more charts. They needed a diagnosis.
We also noticed that for some sellers, especially the inexperienced ones, interpreting charts was difficult. Instead of building another analytics dashboard, we focused the MVP on 10 products needing attention, prioritizing declining sales and stock issues.
The goal was simple: help sellers identify which products needed action, without making them interpret more data.
The logic
The UI looked simple. The logic underneath was the hard part.
On the surface, the feature is a list of products with a recommendation attached. But a recommendation is only useful if it’s the right one. A seller with low impressions needs a different fix than one with low checkout conversion.
I worked with the data and product team to map each possible problem to a specific action. The idea is to find root causes for sales decline and find the solution. One example is: Low impressions → Visibility issue → Recommend increasing exposure.
The challenge wasn't just identifying underperforming products. It was deciding which problems we could reliably detect and turn into useful actions.
Working with the product and data teams, we mapped available signals to potential recommendations. Three considerations shaped the direction:
Relevance: Does the signal point to a problem sellers can address?
Reliability: Is the available data sufficient to support the recommendation?
Feasibility: Can we deliver it within the MVP without significant backend changes?
Rather than trying to diagnose every possible issue, we narrowed the scope to recommendations that could be supported by existing data and communicated clearly.
First concept
My first concept aimed high.
With the logic in place, I designed the most complete version I could. For each declining product, sellers would see:
Two recommendations, so they could choose how to respond
A full performance breakdown (views, favorites, sales, and conversion at each step) so they could see exactly where buyers dropped off
A sales funnel graph (Informasi Penjualan) showing the whole journey from visitor to payment
The idea was that sellers would be able to see the story of their product, not just receive an instruction.
Validation
Testing and engineering both pushed back, from different directions.
From sellers. We tested the concept with four sellers. Sellers understood the idea, but the details caused friction:
“Low” meant different things. Some saw 3% conversion as high, others saw 6% as low.
“Low stock” depends on category. 50 units is low for food but plenty for electronics.
Too many steps. Every extra step to reach a recommendation meant fewer sellers would act.
From engineering. When I reviewed the concept with developers, they flagged that the product-level statistics and the sales funnel graph both needed major backend work that wouldn’t fit the timeline.
So I had a design that sellers found confusing and that we couldn’t build in time. Before cutting anything, I needed to decide what the feature was really for.
Product scoping
I asked one question of every feature: does this change what a seller does?
If it did, it stayed. If it only added depth, it moved out of the MVP. That question made the trade-offs clear: What stays: the diagnosis and the one-click action. Those things were the point.
Simplified
Concept V1
Two recommendations per product
MVP
One recommendation, shown directly
A clear action beats a choice. The flow went from 2 clicks to 1
Reused
Concept V1
Full statistics inside the feature
MVP
Link to the existing Product Statistics page
The numbers already existed. Rebuilding them would have added weeks, not actions.
Embedded
Concept V1
New standalone feature
MVP
Entry point on the daftar barang page
Sellers already look at their products there, so I met them where they were.
Defaulted
Concept V1
Custom definition of "low" per category
MVP
One default threshold per signal
No single number satisfies every seller. A consistent default let us ship and let us know where to refine.
Deferred
Concept V1
Sales funnel graph
MVP
Moved to a later release
It needed a heavy BE work and it just added more info without adding additional impact.
Final design
One click from “something’s wrong” to “here’s the fix.”
In the shipped version, a seller opens their product list and sees which products are declining. Each one shows a diagnosis in plain language and one clear action. No dropdown to open, no menu of options to compare.
Impact
Within six months, refunds from out-of-stock fell by 10%.
Within six months of launch, refunds from out-of-stock issues fell by 10%. Sellers who were nudged to update stock and pricing stopped selling items they couldn’t deliver, which was the most avoidable source of refunds in our original data.
I didn’t have access to adoption / sales data after I moved on, so I can’t claim an effect on sales. If I were continuing, I’d track [he share of sellers who act on a recommendation and sales for diagnosed products vs. a comparison group.
Reflection
What I learned
The simplest design holds a lot of complexity.
My job as a designer is to explore the complexity and bring simplicity to users. Elegant solutions require a lot of exploration and iteration.
Knowing what not to build.
By focusing on clear recommendations and reusing existing functionality, we could preserve the core value while reducing complexity.