Konga Search Service

Bringing the tool behind every shopper search in-house

The search box Konga shoppers use runs on this tool, and so do the rules that decide what it shows.

I was the lead designer on Konga’s in-house search and merchandising platform. It replaced a third-party search tool paid for in US dollars, and brought search rules and product recommendations, which had run on separate services, into one interface. Because it followed a standard search-management layout, merchandisers moved across at launch without retraining.

Lead designer, 2024 to early 2025. I worked with search engineering, merchandisers and business stakeholders across search, recommendations, onboarding and settings.

Why the old tool had to go

Cost

Konga earns in naira but paid over $75,000 a year, in dollars, for a third-party search tool. It was also an outdated version, so the dollar price bought less than the current product offered.

Integrations

The integrations engineering wanted sat in a pricing tier Konga wasn’t paying for. Product recommendations ran on a separate service, and voice and image search had nowhere to go.

Daily use

Shoppers searching for a phone got accessories first, and staff testing the site saw the same. The outdated interface was hard for merchandisers to work in, so fixing results was slow.

Before and after

The third-party tool

Cost

A yearly subscription in US dollars, for a business earning in naira

Daily use

An outdated interface that was hard to work in

Integrations

The integrations engineering needed were in a tier Konga didn’t pay for

What’s next

Recommendations on a separate service, and no place for voice or image search

Konga Search Service

Cost

No vendor subscription for search, and over $75,000 a year avoided

Daily use

A layout merchandisers already knew, with no retraining at launch

Integrations

Konga’s own platform, so integrations are an engineering decision, not a pricing one

What’s next

Search and Recommend in one workspace, with voice and image search ready to switch on

Released, and received well

The engineering manager who led the build announced the release of Konga Search and credited the group who built it. I was part of that group.

“We were spending over $75,000 a year on a third-party search tool, but our customers still struggled to find what they needed.”

From the release announcement. Within weeks, shoppers were describing a “seamless experience”.
$75K+A year in third-party search costs avoided

It was the work of a team.

The trade-off: building in-house meant a period where relevance was still maturing. Putting the merchandising controls first meant the team could close that gap itself instead of waiting on engineering for every change.

A familiar layout, on purpose

I didn’t design a new kind of tool. Merchandisers already knew how search-management dashboards are laid out, so I followed that standard layout and spent the design effort where no layout could help. That meant no retraining at launch, and engineering could build against a pattern everyone already understood.

A KSS settings screen with coloured boxes marking the product rail, module menu, application and index context, index summary and actions, tabs, grouped settings and setting detail
The layout, broken down. The product rail, module menu, application and index context, index summary, tabs and grouped settings follow the standard pattern, so anyone who has used one of these tools knows where things are.

One workspace for search and recommendations

Search rules and product recommendations used to live in separate tools. I mapped how they could share one product rail, one set of components and one way of writing rules. Recommend has its own models (complementary, alternative, trending, looking similar) and its own rules, written the same way as search rules.

Voice and image search weren’t part of the first release, but I gave them a place in the rail and in index settings, switched off, so engineering had a home for them when the time came. Personalization got the same treatment.

Left: the Recommend module beside Search, with a 4-step start and recommendation models. Right: voice and image search toggles in index settings
Designed ahead of the build. Left, Recommend sits beside Search with its models and a 4-step start. Right, voice and image search toggles live in index settings, off by default, under search behaviour.

A component library for the build

I broke the interface into reusable parts, so engineering could build a piece once and use it everywhere. The library also gave design and engineering the same names for things, which mattered when we talked through complicated search logic.

The rules table: a search field, a count of rules, and a rule row with its query condition and pinned products, whose IDs are blurred
Rule summary chips: query contains, category page and date conditions
Searchable attributes in a drag-to-reorder list, each marked ordered or unordered
The pin field: search items to pin, with a position stepper
Index actions: New, Add records, Manage index, copy and refresh
The onboarding stepper: import data, select data to display, searchable attributes, configure relevancy and implement search
Building blocks from the KSS designs. The rules table, rule chips, attribute rows, the pin field, index actions and the onboarding stepper. Product IDs are blurred.

The cases a layout doesn’t show

Following a familiar layout settles where things go. It doesn’t say what happens when there’s nothing there yet, when an action can’t be undone, or when 2 rules pull in different directions. I worked those out with search engineering, one case at a time.

Empty state: No synonyms yet, with an explanation and an Add Synonyms button
Delete index dialog: type DELETE to confirm before the index, its records, rules and synonyms are removed. The index name is blurred
Empty state for query suggestions, with an example of suggestions and a button to create the first index
Partial export: choose settings, synonyms or rules to export. The index name is blurred
Advanced setting: pinned items must match active filters to be displayed
Use cases for recommend rules, including hiding out-of-stock products and keeping to the viewed category
Mapping what can go wrong. Empty states say what to do first. Deleting an index means typing DELETE. Export lets you pick settings, synonyms or rules. Pinned items must match a shopper’s active filters, so a pin can’t override a filter. Recommendation rules cover out-of-stock products and category limits.

Reviewed case by case, tested with merchandisers

Design reviews went through the search logic case by case: how rules combine, what a pin does under a filter, what shoppers see when nothing matches, and which actions need a confirmation step.

In testing, I sat with merchandisers while they worked through their usual tasks, to check the new tool was easier to use than the one it replaced. At launch they moved across without retraining.

From a rule to a search result

Merchandisers write rules here. Each one has a condition, such as what the query contains, and a consequence, such as pinning a product to a position. Those rules decide what shoppers see in the search dropdown and the results behind it.

Rule summary chips: query contains running shoes, with a filter and a context, a category page rule and a date rule
The pin field with a position stepper
Konga search dropdown for Samsung: suggestions, products and matching brands
Existing interface

The rule and the dropdown are shown side by side to explain the relationship. They aren’t the same search.

What I took from it

When a business is paying $75,000 a year for a tool that isn’t working, the useful design is the one people can use on day one. Following a known layout freed time for the parts no layout covers.

Designing the parts with engineering, rather than handing over screens, kept a year-long build moving and gave us a shared way to talk through complicated search logic.

Shoppers got better results because merchandisers could finally fix them. Work behind the scenes showed up on the storefront.

Get in touch

Email me

Open to design roles. I’m in Lagos, and my hours match the UK and Europe, with a few hours of overlap with the US.

You made it this far! Want to feed a giraffe, or help feed a baby elephant?