Data Science Insights¶
Validated findings from BigQuery analysis of all three collectibles products. Each finding has been through multiple waves of cross-validation. These are the numbers that should drive product decisions.
The 10 Governing Numbers¶
These are the metrics that define the collectibles business. Every product decision should reference at least one.
| # | Metric | Value | Source |
|---|---|---|---|
| 1 | Revenue concentration (top 10%) | 79-87% across all products | F003 |
| 2 | Cross-product spend multiplier | 95x (3-product vs 1-product users) | F001 |
| 3 | Weekly S-segment churn | 41-46% across all products | F002 |
| 4 | Marketplace discovery (cheap packs) | 83% never find marketplace | F013 |
| 5 | W0 conversion improvement | 4x (2.9% → 12.3%) via positive EV | F011 |
| 6 | Challenge retention effect | +11.8pp (93.9% vs 82.1%) | F004 |
| 7 | NBA XL whale decline (2yr) | -49% (353 → 180 quarterly unique) | F005 |
| 8 | NFL seasonal revenue swing | 5-8x between peak and trough | F008 |
| 9 | Disney L-tier loss (2yr) | -75% (456 → 113) | F006 |
| 10 | VIP deposit match ROI | 2.7x (self-funding) | F015 |
Findings by Theme¶
Pipeline & Funnel Health¶
The whale factory: how users move from signup to whale status, and where the pipeline breaks.
- F002 — Universal Pipeline Pattern: All products share identical structural patterns. Whale-dependent revenue, shrinking pipeline, every segment losing ~50% headcount over 2 years. Disney is worst (75% L loss, 51% M→S regression weekly).
- F013 — Marketplace Discovery Broken: 83% of cheap-pack buyers never discover the marketplace. $50+ pack buyers discover at 46%. The marketplace creates whales; packs create one-time buyers.
- F022 — $50 Packs as Marketplace Gateway: $50+ pack buyers discover marketplace at 46% and spend $1,025 avg on crossover. Users with 4+ marketplace transactions in first 7 days become L/XL at 2-5x base rate.
- F010 — First 30 Days Determine Everything: Disney users who transact in first 30 days retain at 7x. NFL at 6.4x. After 30 days, activation probability drops sharply.
- F011 — W0 Conversion 4x Win: Positive expected value pricing lifted NBA signup-to-purchase from 2.9% to 12.3%. Sustained 10 months. Single biggest product win of 2025.
- F006 — Disney Pipeline Most Broken: 75% L-tier loss, 51% weekly M→S regression, 2.5% M→L upgrade rate (half of sports products). Pipeline bleeds users before trading can activate.
Whale Economics¶
Who the whales are, how they spend, and why they leave.
- F003 — Revenue Concentration: Top 10% = 79-87% of revenue. Top 1% = 33-50%. Universal across products and time periods.
- F001 — Cross-Product 95x Multiplier: 1,998 three-product users generate $46.7M aggregate spend. Critical caveat: ecological fallacy — these users were different BEFORE adopting multiple products. The 95x is a descriptor, not a causal lever.
- F005 — XL Whale Decline is Seasonal: NBA XLs: 353 → 180 quarterly unique (-49% over 2yr). The platform no longer recovers from the off-season (Q1 2025 +10% recovery, Q1 2026 +1% recovery). Remaining whales spend harder (+29%).
- F021 — Disney Whales Are Committed: 87.4% of Disney top 500 are net depositors ($6.8M net invested). The whale conviction problem is solved — the pipeline-to-whale path is what's broken.
Engagement Levers¶
What drives retention and engagement — and what doesn't.
- F004 — Challenges: Correlation Not Causation: Challenge participants spend 20x non-participants, but matched-control analysis shows challenges prevent churn (+11.8pp retention), they don't cause spending. The 20x is self-selection.
- F012 — Challenges as Universal Engine: NFL outperforms NBA: 30% participation vs 14%, 25x multiplier vs 20x. NFL off-season: challenge-active users survive at 92.9% vs 63.6%. Challenges are load-bearing retention infrastructure.
- F007 — Disney Trading Engine: Trading is Disney's engagement driver (218x spend multiplier at 50+ trades). But 70% of traders start 90+ days after first purchase — too slow for the pipeline that bleeds users weekly.
- F014 — Best Pals Content Spike Failure: 5,617 buyers, 98.1% never traded, $18 avg spend, churned. The 1.9% who traded: $5,408 avg, 48 still active. Content spikes without trading bridges = wasted acquisition.
- F015 — VIP Deposit Match 2.7x ROI: Self-funding reactivation lever. Should be permanent part of the off-season toolkit.
- F016 — IRL Events 317-Day Reactivation: 99% signup completion rate at events. One user reactivated after 317 days dormant. Physical experiences convert at rates no digital channel matches.
- F018 — Mass Common Burns: Zero Impact: 160x more Commons burned vs Rares. Zero differential marketplace lift. 2.5M Common listings = 45 months of inventory. Stop mass Common burn programs — they're burning noise.
Product Operations¶
Seasonal patterns, cadence, revenue mechanics, and reactivation signals.
- F008 — NFL 5.8x Seasonal Swing: Jan playoff peaks declining YoY ($2.48M → $1.92M → $1.32M). Off-season loses 39-41% of L/XL population in 4 months. The off-season is where the product contracts permanently.
- F009 — NFL Whales Migrate, Don't Leave: NFLAD $5K+ whales spend MORE on NBA during the NFL off-season than on NFLAD during the NFL season. Jul 2025: $2.1M on NBATS vs $254K on NFLAD. The off-season problem is a product problem, not a customer problem.
- F017 — Monthly Drop Cadence 2.3x Revenue: Shift from weekly to monthly tent-pole drops: 2.3x revenue per drop, 131% more returning buyer revenue, 2.4x variance reduction. Scarcity discipline works.
- F019 — Buyback Distortion 70.1%: Buybacks consumed 70.1% of Q2 2025 gross marketplace revenue ($6.95M of $9.9M). Always use net revenue in analysis.
- F020 — Lapsed User Reactivation Signals: 19,631 lapsed users (90+ day gap) reactivated in 2025. 64% viewed pack drop page in the 7 days before returning. 83% bought a pack on return. Pack drops are reactivation events. 9,676 lapsed $1K+ users still log in and window-shop.
How to Use This Data¶
For Product Decisions
Every feature proposal should reference which finding it addresses. "This feature improves F013 (marketplace discovery)" is a complete justification. "This feature is cool" is not.
Correlation vs. Causation
Many of these findings show correlation (challenge participants spend 20x more). The matched-control analyses isolate causation (challenges add +11.8pp retention, not spending). Use the causal numbers for projections, the correlational numbers for targeting.
Data Access
Source data lives in BigQuery production_sem_open.* tables. David Wang owns the semantic layer. For ad-hoc queries, see the Data Infrastructure page.