Lumen Trends
Product Design / AI & Data
·
2024-2026
ROLE
Sole Product Designer, concept to release
TEAM
1 PM · 3 data science · 2 backend · 2 frontend · 1 search
TIMELINE
Mid-2024 – 2026
OUTCOME
Closed beta with 59 client accounts and 88% activation, then launched in September 2026
Lumen Trends search: natural-language input on top, trends recommended for the user below
Consumer brands use Talkwalker to monitor social and online conversation, but its tools were built to answer questions users already knew to ask. Lumen Trends set out to surface emerging trends before they peak. The data science team's engine produced statistical scores and LLM-generated output with no interface; my job was to turn it into a product.
Natural-language search replaced keyword-only input
Discovery by category: ranked trends, live results and top influencers in one view
the challenge
In interviews and feedback sessions with clients, the same pain points kept coming up:
Hard to tell what is actually trending, and even harder to tell what is becoming a trend.
Limited reach: teams didn't have access to all the platforms and networks where trends first emerge.
Hard to classify: even once a signal was found, making sense of it was manual.
Slow: teams lost a lot of time finding trends and then analysing them.
Design goal: help brand teams find emerging trends early, understand them quickly and move straight into deeper analysis.
Discovery and feasibility. From mid-2024, what was then called the "Trends module" went through a long discovery phase of client interviews alongside the PM. In late 2024 I produced the first wireframes and tested them against a basic prototype my manager built on the API and OpenAI, which proved the engine could return useful results. Only then did I move into full design in Figma. Some ideas were deliberately left out of the first release, such as letting users upload their own keyword lists.
Prototype flow: search by term or category, the loading steps, then ranked results
Row menu: “Save as topic” and “Hide query” replaced the error-prone “Edit query”; the query behind each trend shows under its name
The pivotal decision: don't rebuild analysis, connect to it. The first versions included a full trend-analysis area inside the product. After reviewing this with the tech and data teams, we agreed it was the wrong bet: the platform already had mature analytics tools, and rebuilding them would duplicate effort and split the experience. Instead of designing analysis, I had to design the bridge between discovering a trend and analysing it in the rest of the platform. The PM and I interviewed Customer Success Managers to learn which widgets they relied on and how they investigated a trend, and I redesigned the IQ app Trends analysis page around that use case.
A deliberately simple beta. It shipped on Talkwalker's existing UI to reach clients quickly. Search started with keywords only; as the data science team's technology matured, I redesigned the input for natural language, so users could ask anything, even paste a full brief, without needing Boolean logic.
What the beta taught us, and what I changed:
Long AI response times read as errors → a dedicated loading experience that made progress visible.
"Save as topic", the main action, was barely used → far more visual priority as the primary call to action.
"Edit query" caused errors and zero results → removed; each trend row got a menu with "Save as topic" and "Show query", which reveals the query under the trend's name (hidden by default) and switches to "Hide query" once it is shown, so users can see why a trend appeared without breaking the search.
Results were hard to analyse and compare → new network visualisations, trend classification and a table supporting more metrics.
A trade-off I didn't win. I proposed an asynchronous search: searches would go into a history, and users could keep working anywhere in the platform and be notified when results were ready. It was vetoed because the backend team didn't have the capacity, so the loading experience was the best solution within that constraint.
Moving to Bento. After the acquisition, I proposed moving Lumen Trends to Hootsuite's Bento Design System and redesigned the next iteration in Bento, on reusable components instead of one-off patterns. This is the version that launched to market in September 2026, shortly after I left.
From search to action: a saved trend flows straight into the platform’s existing analytics
conclusion
Launched to market in September 2026 with my Bento redesign, shortly after I left.
Closed beta with 59 client accounts and 100+ users across teams, including global FMCG and Beauty brands; 88% activation, and a module sold before general availability.
Discovery connected to analysis: trends flow directly into the platform's existing analytics through the redesigned IQ app, instead of living in a separate tool.
Beta issues addressed: perceived errors while waiting, an underused main action and query editing that led to zero results.
Reflection. I'd push harder for a lighter version of async search. The wait was the biggest friction, and the loading screen only softened it.