Rebuilding TikTok Ads Measurement to Correct Systematic Undervaluation
Snapshot
Product Type
B2B ads measurement platform (analytics / data insights), 0→1 from concept to launch
My Role
Lead Designer, end-to-end — framing, research, system design, GTM
Timeline
2023–2024
Core Challenge
When the data that proves TikTok's value is also data advertisers don't yet have a framework to believe, which do you solve for first — credibility or comprehension?
What Was Actually Broken?
Last-click works well for intent-driven ads like search, but breaks down for discovery-based platforms like TikTok.
Last-click attribution failed on TikTok for three reasons:
View-through influence
Users often see a TikTok ad, don’t click, and convert later.
Longer conversion cycles
TikTok ads often create intent early, with conversions happening days or weeks later.
Multi-touch journeys
Conversions usually result from multiple ads across campaigns and channels, where TikTok is often an early touchpoint rather than the last click.
The Structural Failure
As a result, most TikTok-driven value was invisible under last-click attribution. A post-purchase survey showed 79% of TikTok-driven purchases were invisible under last-click attribution. Applying last-click logic here doesn't produce an incomplete picture; it produces a actively misleading one.
For advertisers, this led to:
  • Undervaluation of TikTok
  • Lower confidence in spend decisions
  • Trust erosion
Why this was hard (0→1 measurement):
  • No industry benchmark for view-through attribution; all metrics were new.
  • Advertisers had no mental model to interpret or act on the insights. We were asking advertisers to accept a new view of reality on our terms alone.
Design Judgement
The temptation in a product like this is to lead with impact: surface the attribution gap as a headline number, let the delta make the case, and drive adoption through emotional urgency. The risk is that in a high uncertain measurement context, a misread metric that leads to a wrong budget call is harder to recover from than an advertiser who simply needed more time to understand the data.
So the core judgment was: comprehension is a prerequisite for trust, and trust is a prerequisite for behavior change. The action layer — next-step recommendations, optimization prompts — wasn't deprioritized. It was sequenced. MVP focused on establishing methodological credibility; action-oriented features were introduced in staged rollout only after early adoption validated that the measurement foundation was trusted.
This directly impacted the design goals and strategies, research sprint, and MVP prioritization.
Product scope
We launched three core Attribution Analytics features globally, each explaining a different dimension of TikTok’s value in an understandable, reliable, and actionable way:
1. Performance Comparison (Merged 2 concepts)
Introduces comparative attribution metrics.
Shows how much value last-click attribution misses by comparing click-through and view-through conversions (CTA vs VTA) across funnel events and attribution windows.
2. Time to Conversion
Introduces time-based attribution metrics
Shows when conversions happen over time, highlighting the difference between click-through and view-through influence (CTA vs VTA).
3. Touchpoint to Conversion
Introduces path-based attribution metrics
Shows how multiple campaigns contribute to conversions across a journey, rather than through a single interaction.
Design goals and strategies
To make correct product decisions under high uncertainty:
What we optimize for
Interpretability
can advertisers understand the data?
Reliability
is it stable and decision-safe?
Actionability
can advertisers decide what to do next?
How we design toward it
Key strategies: Metrics/dimensions; granularity/aggregation; visualization; in-product education; insight-to-action
Explore broadly before committing to one direction
Given the high uncertainty, we explored before designing deeply.
6-week research sprint (Design + PM + DS+ Research)
10
concepts explored in parallel
33
rapid iterations
65+
validated hypotheses
Focus: comprehension and value across 50 metrics & 14 dimensions, visualization, education, actionability
Execution choices
  • Used real advertiser data in 60+ interactive prototypes
  • Prototyping balanced speed, data flexibility, visualization capability
Outcome
By the end of the sprint, all concepts moved from ambiguous to concrete; improved on interpretability, reliability, actionability
Decide what to ship using a 5D prioritization framework
After exploration, we had more strong concepts than we could ship.
How we decide what to ship: a 5-dimensional framework
Interpretability
Desirability signal
Reliability
Desirability signal
Actionability
Desirability signal
Data Quality
Integrity constraint
Data Scalability
Integrity constraint
All ten validated concepts were evaluated across five dimensions.
Outcome
  • Selected highest-scoring concepts to ship as three core MVP features
  • The remaining concepts weren't abandoned — but their desirability and data integrity gaps meant surfacing them prematurely would hand skeptics a reason to doubt everything. We built an internal platform where client-facing teams could work with those features through real sales conversations, pressure-test the data, and build the evidence base before any public release.
Design Intervention
What Changed and Why It Mattered
Anchoring All Metrics to Conversion Revenue
The Old Approach
Prior advertiser-facing data showed conversion counts in isolation — useful for analysts, but disconnected from the budget decisions that actually mattered.
The New Framing
We reframed every attribution metric around revenue impact: not "how many view-through conversions happened," but "what is the revenue value of the conversions last-click missed."
This made the attribution gap legible to budget owners, not just measurement teams, and gave advertisers a defensible number to bring into internal spend conversations.
Designing for Different Advertiser Sophistication Levels
Challenge
KA advertisers needed multi-dimensional attribution depth; SMBs needed simplified roll-ups they could act on without a measurement specialist; mid-market needed a balance of both.
Design judgments
Optimize for broad adoption first, and build product depths for power users over time.
Excution
  • Consistent information architecture across features
  • Defaults set to surface the most decision-relevant, least ambiguous view for each feature
  • Additional granularity — deeper breakdowns, extended attribution windows, more dimensions — was available but not foregrounded, surfaced progressively as users signaled readiness.
Outcome
Power users could get to the depth they needed; less sophisticated users weren't overwhelmed before they'd established a baseline understanding. The design didn't try to serve every segment equally at once — it sequenced the experience so the same product could work across all three.
Consistent information architecture across 3 features
Progressive disclosure with additional granularity in 14 dimensions — deeper breakdowns, extended attribution windows, and more
Smart defaults for decision-relevant first views in 14 dimensions
Education as a Cross-Touchpoint System
Challenge
New metrics → no mental model → confusion + misuse risk → erode trust
Design judgments
Interpretability must be designed in, not explained later
Interpretability is the foundation for reliability + actionability
A measurement system works only if interpretability, reliability, actionability are understood consistently across touchpoints
Education builds comprehension, and is especially important for a brand-new data-complex product. At the same time, we (Design + Content Design + GTM) treated education as a structural design problem to drive adoption.
In-Product
Build interpretability and reliability:
  • Contextual narratives
  • Interactive introductions
  • Transparent communication of data quality
Internal Training
Gave client-facing teams the confidence to present unfamiliar metrics
External Content
Established market-level credibility in Help Center content and launch materials
Interactive introductions
Contextual narratives for complex or new concepts
Clear communication of data quality, coverage, and limitations
Help center articles
PR launch marerials
Internal training doc
System-level impact
48 → 84
NPS in client-facing team global training +36 points
Client-facing teams consistently described the product as intuitive, reliable, and actionable — signals that tracked directly back to the three design goals we'd set at the start.
Designing for Actionability When Insights Weren't Enough
Challenge
Advertisers could read the data, but stalled when it came to acting on it. New insights take time to digest→ unfamiliar metrics and cross-checking needed → insights do not naturally translate into action.
Design judgments
Actionability must be designed, not assumed
Insights create value only when they change decisions
Rather than leaving the insight-to-action gap as a future problem, I led a workshop with seven XFN teams to systematically map what actions the data could reliably support.
The Output
17 actionable themes across the MVP features, distilled into two core strategies:
  • Triggering personalized, valuable, and measurable recommendations at the right decision moment
  • Promoting holistic measurement approaches — such as conversion lift studies and post-purchase surveys — when optimization was premature
Most initiatives were added to the roadmap, with several entering development before my departure. The intervention mattered not because it shipped a feature, but because it reframed the product's job from delivering insights to supporting decisions.
Time to Conversion — key iterations and decisions
Final decision
Define conversion timing using a small set of representative time signals, rather than granular breakdowns
Key challenge
Identifying the right metrics, dimensions, and visualizations that explain conversion timing accurately without overwhelming users
Key design decisions
Removed funnel position and event-type breakdowns
Focused on average days to conversion and days to capture 80% of conversions
Anchored conversion timing to revenue to reflect business impact
Prioritized table-first views for advanced users, supported by simplified visuals
Iterations
Ads Path Analysis — key iterations and decisions
Initial hypothesis
Path analysis could explain TikTok’s multi-touch value
Key constraint
Objective-level paths were unstable due to data quality and scalability limits
Parallel validation
  • Direction A: Explored a more stable multi-touch signal (ad frequency)
  • Direction B: Validated whether path sequencing was valuable
Final decisions
  • Keep path sequence metric; shift dimension to campaign-level, anchored paths to conversion revenue
  • Spin off ad frequency journey as a new roadmap feature
Iterations
Impact
Advertisers who adopted Attribution Analytics shifted from using it as a reporting tool to using it as a decision input — bringing attribution data into internal budget conversations rather than simply logging it. That behavioral shift was the primary validation signal: the interpretability foundation had held.
59%
Revenue Penetration
Among key accounts
62%
Revenue Penetration
Among mid-market — above the 50% threshold that signals mainstream adoption
26%
Revenue Penetration
Among SMB
53%
Weekly Spend Growth
Among users during BFCM vs. 28% for non-users
+52%
ARPA Growth
Year-over-year for adopters vs. +22% for non-adopters
Market signal
Recognized as a first-party solution addressing last-click attribution limitations
"Visibility into view-through data made it easier to prove TikTok's value and scale campaigns."
— Power Digital Marketing
What I'd measure next is decision confidence as a behavioral proxy: are advertisers who use Attribution Analytics more likely to increase TikTok budget in the 90 days following adoption, and does that effect persist at 6 and 12 months? The spend uplift is compelling but correlational. The question worth answering is whether the product changed how advertisers make decisions, or whether higher-confidence advertisers self-selected into adoption first.
Trade-offs & Reflection
What I Optimized Against: Shipping four features instead of ten meant the product felt incomplete. Valuable concepts were delayed because their data integrity didn’t meet the bar — a deliberate trade-off to protect long-term user trust.
Where I Preserved Friction Intentionally: For users still forming mental models, one-click optimization would have driven action without understanding. Most recommendations were delivered as guided flows requiring engagement before commitment. The goal was durable behavior change. One-click prompts appeared only when system confidence was high.
What Remains Unvalidated: Whether product-level interpretability can shift organizational measurement strategy. Many advertisers operate on institutional attribution models, and individual analyst behavior change may not translate into budget decisions.
What I'd Do Differently: Actionability was treated as post-launch, despite research signaling it was launch-critical. Elevating it earlier — alongside interpretability and reliability — would have embedded clearer decision support from day one.
I'm drawn to problems where the product has to earn the right to change user behavior — where the design challenge isn't clarity or aesthetics, but credibility. This project required me to treat trust as a system property: something that had to be designed into metrics selection, visualization defaults, educational scaffolding, and sales enablement simultaneously, because a gap in any one layer would undermine the others. Under constraint, I tend to anchor on the most fragile assumption and build from there. What makes my approach distinct is that I treat scope discipline as a design skill: knowing what not to ship, and being able to articulate why, is as important as knowing what to build.