From open-ended search
to structured research.
The evidence, decisions, trade-offs, the direction that didn't work, and what I'd measure next time.
Marketers came with a business question, but AdSpyder handed them a search box: no competitor context, dead ends when data was missing, and campaigns optimised by hand. Paid users fell to about 5.
Rebuilt the product around intent: guided research projects, competitor evidence before tracking, recovery paths for empty results, and an Agent Hub that watches campaigns.
A clearer starting point, fewer dead ends, and one place to optimise campaigns. Sales and outreach also contributed to growth.
Contents
AdSpyder helps marketers discover, analyse and learn from advertising activity across 15+ platforms.
Find what is working in the market → understand why → turn that intelligence into a decision.
Senior Product Designer
2023 to 2026
Engineering, data and business stakeholders
Research, UX, UI, prototyping, post-launch iteration
I looked at what happened
before the search.
I reviewed the existing product experience and spoke with about 10 active users and 3 to 4 churned users: why they came, where they struggled, why searches failed, and why they stopped.
- 1Users start with a goal, not a query: understand competitors, validate ideas, find winning creative, explore a market.
- 2Context was scattered: business information, keywords, competitors and intent lived in different places.
- 3Users had to do the synthesis: AdSpyder exposed intelligence, but people still had to connect the dots.
From ‘search is difficult’ to ‘users don't know how to translate their business question into the right research setup.’
Organised around
the project.
Three decisions,
and what they ruled out.
Start with intent, not search.
The first question changed from 'What do you want to search?' to 'What are you trying to understand?'
Trade-off. More setup up front. Less uncertainty later.
Make the project the unit of research.
Goal, business context, keywords and competitors are captured once and reused across every research activity, instead of rebuilt for each search.
Collect context progressively.
My first version asked for everything before showing anything. The final flow asks for the minimum, shows value, then asks for more when it helps.
Trade-off. Less complete context at the start, in exchange for value sooner.
A single intent-driven entry point that branches based on the user's context, not a dashboard that assumes product knowledge.
Search-first entry as the primary experience, and adding more filters to an already uncertain starting point.
The first version asked for
too much, too soon.
Collect goal, domain, business context, keywords and competitors before delivering any value.
The product became more helpful, but onboarding became heavy. Users were asked for a lot before they saw what it would do for them.
The goal wasn't to collect less. It was to ask for the right information at the right moment.
Make competitor intelligence
actionable.
The question it answers: “Why is this competitor relevant to me?”
- 1Impact badges on each suggested competitor
- 2“Review keyword impact” action
- 1“Start tracking”, with what tracking unlocks
- 1Change alerts on tracked competitors
No data should not mean
the end of the journey.
Users could request a fresh crawl and get notified when data arrived, or explore adjacent keywords that already had ads.
- 1Market signals
- 2Adjacent keywords with “Research” actions
- 3Competitor fallback
From finding insight
to acting on it.
Marketers connected their Google Ads account to AdSpyder. Optimising it meant reviewing keywords, locations, audiences, creatives and click fraud separately, by hand, every time.
One Agent Hub, with purpose-built agents per area, instead of a generic AI assistant.
- 1Insight banner with three detected issues
- 2“Deploy agent”
- 1Prioritised actions with impact
- 2Apply and review controls
AI recommends.
People decide.
Show what needs attention, explain why, and let people decide how much control to give it.
Higher-impact actions need explicit approval. Lower-risk rules can be automated. Any rule can be reviewed, edited, automated or kept manual. An activity log records what happened, why, what the agent recommended, what the user approved and what was done.
- 1Each optimisation listed with its effect
- 2“Review individually” option
- 1Rule conditions
- 2Approval setting
What changed,
and what it can't prove.
Monthly sign-ups, Nov 2023 to Aug 2024, as reported publicly by AdSpyder's CEO.
Retention from AI-powered analytics that turned AI insights into clear next actions.
Start from the research goal, not the search model.
Projects keep the context across activities.
Guided progressively, not all up front.
The base for keyword, competitor and market intelligence, and for the Agent Hub.
Since then, AdSpyder has shipped ad generation for text and image ads, and keyword and title optimisation, which I designed.
The redesign was one part of a wider growth effort. Sales outreach and campaigns ran in the same period, so the growth is not credited to design alone.
What I'd do
differently.
Validate the minimum context users will give before they see value, and set quantitative baselines earlier.
Activation
Sign-up → project creation → first search → first useful result
Retention
Users returning to continue research or monitor competitors
Recovery
No-data searches → alternative or crawl request → return
Conversion
Project creation → meaningful research → paid workflow
The biggest change wasn't a new screen. It was changing what the product asked users to do first.
