Search design

Everything on a competitor, from a single search.

A redesign of search for Klue, a platform where sales teams collect what they know about their competitors. Sellers look things up in it to learn what they're up against. I led the UI design and tested its behaviour in prototypes.

Klue's redesigned search showing results for “Cirrus”: a grid of identical result cards for a company, a product, a board, a battlecard, a sales call, a highlight and a win-loss report, under a search field and four filter buttons, Types, Topics, Companies and Sources.

Project facts

Role
Design lead
When
Q4 20242
Where
Klue, a competitive intelligence platform
Team
Myself and a PM
Delivered
Figma files, prototypes and video walkthroughs
Status
Implemented
Scope
Search in Klue's web app
Precedents
Slite, Dovetail, Drive, Finder, Klue V1, Slack, Notion, Dropbox, Epidemic Sound, Google, Spotify

80%

of searches were for a company’s name, because no one knew what to search for.1

35

content types were in the product, but weren’t being surfaced.

1

consistent, easy-to-read layout for search results.

The problem

Search hid what Klue knew.

The search experience was fragmented and confusing. It returned only one type of content, so users didn't know what to search for. Research showed roughly 80% of searches were for a company's name.1 We needed a way to reveal the depth of knowledge stored in a user's account.

  • Single type.Results came back as one kind of object. The exception was the golden card, a special result that held only links: a stand-in for content search couldn't reach.
  • Bespoke refinement.Refining results was inconsistent and cumbersome. Some drop-downs should have been toggles, some had tabs, some had drop-downs inside them, and even those patterns were duplicated.
  • Visual variations were wide.Results were visually inconsistent: their hierarchy came from how each author formatted them. With no results, search returned something that looked like a result.
  • Different scanning behaviour.The layout and design changed with where you were when you searched, so every page was read a different way.
Klue's old search results for “cirrus”: a masonry wall of long card fragments with yellow keyword highlights, a yellow golden card holding only battlecard links, and a side panel of radio buttons and checkboxes.

The old search. Fragments of cards in a masonry wall, the golden card of links, and a panel of filter widgets.

What I owned

From the first audit to the prototypes.

I led the UI design and defined result anatomy, introduced a reusable component, established interaction rules, and validated behaviour through interactive prototypes.

The solution

Know what a result is at a glance.

I gave every kind of content the same card, so you'd read a result the same way whatever it was. If every result looked different, a seller would spend the search working out the page instead of the competitor. The design came down to four key elements, each a job a seller needed done.

A search result card with its eight parts named by teal leader lines: direct rival, object title, controls, content, data line, type, attribution and date, dashed for the optional ones.

The parts of the card. Dashed lines mark the optional ones.

Underneath, part one: the content model

Shaped by the legacy system.

It had to fit the legacy system: keep its tagging, respect each user's permissions and normalize content for indexing. The data already knew far more than search showed. I spent time mapping the current content model to basic, readable and consistent types that our users would understand at a glance.

Machine-named system values, such as topic.legal.lawsuit and source.chat.slack, joined by lines to the plain filter chips they became, such as Lawsuit and Slack, under Types, Topics and Sources.

From system values to seller words. I mapped technical definitions to common language.

Underneath, part two: the mapping

Every object, mapped into one card.

The card came from the objects it had to show. I took the fields every kind of object had in common: a link, a title and a description, with a rival and a line of data as options. I adhered to the existing content, while leaving flexible space for customization.

A company overview result card beside the full company profile it came from, with arrows tracing its title, description, link and update date to their sources.

A card mapped to its object. The new card holds everything the legacy design did.

Tough calls

Four calls shaped the design.

  • Inline filters.

    I wanted search that felt fast and simple. Inline filters under the search bar were ideal.

  • Short cards.

    I capped the card's height because scanning results was more important than reading in place.

  • The web app only.

    Search in Klue's web app, on a 13-inch laptop screen. Everything else was out of scope.

  • A vision, shipped as smaller features.

    I designed all of search at once, so it had to hold together as a system and still come apart into tickets.

The outcome

It set where Klue's search went next.

Simplified search was more consistent, legible and easy. Klue started putting it into production, from my designs and video prototypes. It hadn't launched by the time I left, but the result card and the filters set the direction of Klue's search for 2025.

Later, I built the redesign as a working search

80%

of searches were for a company’s name, because no one knew what to search for.1

35

content types were in the product, but weren’t being surfaced.

1

consistent, easy-to-read layout for search results.

My take

Better indexing, less surface.

My biggest takeaways were that this project emphasized the need for far better indexing throughout the product, as well as the benefit of reducing product surface/complexity.

Questions? Answers.

  • Why are the names made up?

    The real screens hold Klue’s customers, competitors and content. The pictures are redrawn from my design files with an invented cast: Larkspur Security, its seller, a buyer at Halden Systems, and the competitors Cirrus and Breachline.

  • Where did it start?

    With a sketch. The product manager had drawn a new results page to put in front of users in interviews, and I rebuilt it as a comp. My note on it said the real explorations would need precedent research and design proposals that consider all options for improving search.

  • What did you look at first?

    As many variations as I could find, to feel the direction things were going: search in Slite, Dovetail, Drive, Finder, Klue V1, Slack, Notion, Dropbox, Epidemic Sound, Google and Spotify. I marked what worked and what didn't: a drop-down over a drop-down, a filter with 200 options, newest and oldest as two controls instead of one button.

  • Who found the 80%?

    The product manager who owned the search epic, in Klue's search data. My part was the design built on it.1

Next project

Search prototype.

A search design, built as working software.

  1. Klue's search data, late 2024, found by the product manager who owned the search epic.
  2. October to December 2024. The design hadn't launched when I left Klue.

The screens on this page are redrawn from the design files. The companies, people and content in them are invented.