Strand · Creative analytics SaaS
See what your creative is
made of.
A creative performance intelligence platform that breaks every campaign down to its creative DNA and tells teams what to make next.
What this is
Performance tools tell brands that a campaign worked. Almost none tell them why: which hook, format, or talent actually moved the number, and what to make next.
Strand reframes creative analytics around a single question: if performance is the outcome, what were the inputs? Every asset is tagged across five creative dimensions, each dimension is correlated against performance, and the result is a ranked, readable answer rather than another dashboard.
Self-initiated concept. I owned the full scope: research synthesis, information architecture, interaction design, and the data visualisation system.
The problem
Eight conversations, three recurring gaps
I interviewed growth marketers and creative strategists at DTC brands and agencies. Different titles, same story: the data exists, but it never turns into a decision.
Teams can see CTR and ROAS. They cannot tie those numbers back to the creative choices that produced them. "It did well" is not a brief.
Fragmentation
Spend in one platform, creative in another, results in a spreadsheet. Nobody owns the full picture, so insight dies on the way to the next brief.
"By the time I've stitched it together, the moment's gone."
Attribution
Teams see CTR and ROAS, but can't tie them to creative choices. "It did well" is not a brief. It's a shrug with a number attached.
"I can't tell my team what to make more of."
Speed
The monthly report explains what worked after the next campaign already shipped. Learning never compounds, so teams relearn the same lesson every quarter.
"We relearn the same lesson every quarter."
Who this is for
Growth leads need to know which creative choices move the metric and by how much. Creative strategists need a brief that is likely to win, not a coin-flip.
The core concept: Creative DNA
Every asset is tagged across five dimensions: hook, format, talent, message, and CTA. Strand correlates each attribute against performance to surface lift, how much a given choice moves the metric versus the account average. That single primitive turns a wall of metrics into a ranked, readable answer.
Built around a loop, not a report
The information architecture follows the work: see performance, find the cause, turn it into the next brief. Five surfaces, one continuous loop.
Overview
- KPIs
- Trend vs baseline
- Segments
Creative
- Variant compare
- Lift ranking
- Tests
Creative DNA
- Attribute lift
- Next-brief recipe
- Drivers
Audience
- Segment reach
- By creative
- Overlap
Signals
- Risk feed
- Coverage
- Auto-rules
Connect
Spend, creative and results land in one place.
Tag DNA
Each asset is decomposed into five attributes.
Read lift
See which choices move the metric, and by how much.
Build recipe
High-lift attributes assemble into a next brief.
Ship and learn
New creative goes live and feeds the next read.
Three rules I held to
One hero metric: lift
Raw KPIs describe. Lift decides. By making a single comparative number the spine of the product, every screen answers "is this better, and by how much?" instead of leaving users to do the math.
Diverging bars over ranked lists
A list ranks; it does not show direction and magnitude at once. Bars from a center axis encode both at once: what helps, what hurts, and how strongly, in one horizontal read.
Signals informs, never gatekeeps
Creative teams distrust tools that block their work. The brand-safety layer surfaces risk ranked by exposure and keeps a human in the loop. Nothing is auto-paused. Trust was a design requirement, not a feature.
The product
Designed dark, dense, and decisive.
Analysts live in this product all day. Every screen is built for high information density without visual fatigue: clear hierarchy, restrained color, and one accent that only ever means signal.
The decisions
Four places where the interesting part is not the output but the choice that produced it.
Decision 01
Lift as the hero metric
The problem
Raw KPIs give absolute numbers. A 3.9% CTR is meaningless without knowing whether 3.9% is good for this account, this format, this audience.
What I did
Made lift, performance relative to account average, the primary number on every screen, with raw KPIs available one level down.
Why
Comparative numbers are actionable. Absolute numbers require mental math the product should do for the user.
Decision 02
Diverging bars over a ranked list
The problem
A ranked list of attributes shows which ones matter but collapses the signal into a single direction. It cannot show that UGC format lifts CTR +34% while static format drops it 21%.
What I did
Bars emanate from a center axis. Positive lift extends right in teal, negative lift extends left in red. Color reinforces sign for accessibility.
Why
Direction and magnitude in one read. The shape of the chart tells the story before the user reads a single number.
Attribute lift · vs. account avg
Decision 03
Recipe as the output of DNA, not a report
The problem
Even when teams understand which attributes lift performance, translating that into a brief requires a separate creative step that often gets skipped.
What I did
The DNA view assembles the three highest-lift attributes into a "recipe", a starting brief the creative strategist can send directly to the team or refine.
Why
Closing the loop from insight to action inside the product is what keeps teams from reverting to intuition the moment they leave the dashboard.
Decision 04
Signals informs; it never auto-pauses
The problem
Brand safety tools that auto-pause creative destroy trust with creative teams. A flag on a live asset costs real money. Pausing it without human review costs even more.
What I did
The signals feed ranks risk by reach exposed and assigns severity, but every action requires a human to confirm. Auto-rules exist only for low-severity, low-reach issues.
Why
Trust was a design requirement. A tool that creative teams distrust gets ignored, which is worse than no tool.
The five surfaces
Scroll to see each connected screen. Each is a stop on the same loop.
"If this existed, I'd stop exporting to spreadsheets."
Growth lead, unmoderated prototype test, session 3 of 8
Step 1Research
Eight interviews with growth leads and creative strategists at DTC brands and agencies. Screened for campaigns with 10 or more creative variants running simultaneously.
Step 2IA and flows
Mapped the full information architecture before designing any screen. The core loop (connect, tag, read, build, ship) determined the navigation structure.
Step 3Prototype tests
Eight unmoderated sessions on a Figma prototype. Primary task: find the winning creative attribute for a given campaign. Median completion time 9 seconds.
The iteration that changed the DNA view
- Initial design
- Attribute lift as a simple ranked list with percentage values only.
- Observation
- Participants could not compare positive and negative lift at a glance. They read the numbers individually instead of scanning the pattern.
- Realisation
- Direction matters as much as magnitude. A ranked list collapses both into a single axis.
- Final design
- Diverging bars from a center axis. Positive right, negative left. Color reinforces sign so the chart reads before the numbers do.
What I decided not to build
Two decisions where the trade-off is worth naming, including the argument against my own choice.
A global alert counter I could not keep accurate
The obvious design was a badge on the nav counting open signals. I cut it. Signal severity is assessed by a scanning service that runs on a delay. A counter that could show stale numbers on a live asset is worse than no counter.
The honest caveat: in a real product, the right call depends on how fresh the scanning service can be. If latency is under a minute, the counter is worth building. At 15-minute polling, it is not.
Multi-workspace and permissions
Agencies running campaigns for multiple brands need workspace separation and role-based access. I designed a single-workspace product because it is the shape that lets the core attribution model be evaluated without the complexity of permissions obscuring it.
If Strand were a real product, permissions would be the next sprint. The data model supports it. The UI does not yet.
The outcome
What the concept set out to prove
As a self-initiated concept there is no production data. The metrics above come from an unmoderated prototype test. They measure whether the idea is legible, not whether it sells.
The test question was binary: can a growth lead look at the DNA view and correctly name the attribute driving performance? All eight participants could, and they did it in under 15 seconds. That is the only thing this number measures.
What this work does not prove. Whether the tagging taxonomy holds across verticals beyond DTC. Whether the attribution model survives a real account with messy, overlapping campaigns. Whether the recipe output actually changes what gets briefed. These are the questions a real product would answer.
Why I built it anyway. The domain, creative performance analytics, is real, the problem is real, and the UX challenge of making attribution legible at a glance is exactly the kind of information-dense design problem I want to work on. This is the portfolio piece that proves I can think in that space, not that I have solved it.