The Cross-Domain Uniformity Paradox

A 16-iteration directed exploration of the Falcon Telecom & Media synthetic warehouse, focused on what the SUBSCRIBER_KEY join across billing × viewership × tickets × lifecycle events actually reveals.

Executive Summary

The most useful finding from this exploration is a meta-finding: the Falcon Telecom & Media warehouse describes a customer base where cross-domain participation is universal but cross-domain intensity is demographically random. Every one of the 100,000 subscribers streams; 96.9% buy live event tickets at some point; engagement levels are uniform across segment, age, credit, tenure, plan technology, and acquisition channel. The only meaningful demographic discriminator is plan type — Enterprise Fiber subscribers are 47% more likely to never buy a ticket than Prepaid 5G subs (3.45% vs 2.35%) — a small but interpretable corporate-vs-personal account signal.

The four strongest null findings challenge widely-held cross-domain narratives:

• No carrier ↔ platform parent affinity. Comcast subscribers do not over-stream Peacock (Peacock is actually their lowest platform at 19,138 sessions vs Tubi at 19,676). All five prospect carriers' subscribers distribute streaming uniformly across all 12 platforms within a 2.8% spread. Vertical integration produces no detectable streaming preference here.
• 5G access does not drive 4K HDR streaming uptake. 5G plan subs and 4G LTE subs split streaming quality identically (15% / 22% / 54% / 8% across 4K HDR / 720p / 1080p / SD).
• Demographics do not predict genre selection. Drama is 19.4–19.6% of viewing across all 5 segments — including Business Mid. Family does not over-index on Kids content.
• Tenure does not predict churn. Subscribers who churned in 2024–25 have identical average tenure (60.4 mo) to those who didn't.

The two strongest positive findings:

• The COVID Q2 2020 cross-domain substitution is a precise, measurable phenomenon — ticket-buying subscribers collapsed 91.3% (9,263 → 804) while streaming subscribers grew 41.8% (25,292 → 35,874) and view minutes grew 52.7%. By Q2 2021 ticket buying recovered to 12% above the Q2 2019 baseline; streaming returned to baseline trajectory after the spike. The same population substituted streaming for live events — clean substitution, not a structural shift.
• Heavy-tailed spending exists but is demographically random. The top 5% of spenders (4,404 active subs) generate 16.7% of total ticket revenue — a 360× per-capita gap vs the bottom 5%. But these super-spenders are evenly spread across all 5 segments (4.83–5.06% concentration in each). The "VIP cohort" exists but does not have a demographic signature.

Methodology

Mode & Scope

Directed AutoExplore session. Theme: cross-domain Subscriber 360 — what does the SUBSCRIBER_KEY join across FACT_BILLING × FACT_VIEWERSHIP × FACT_TICKET_SALES × FACT_SUBSCRIBER_EVENTS reveal? 70% of iterations focused on the theme; 30% probed adjacent dimensions (carrier, plan, platform, geography, content, time) for unexpected connections.

Confirming Standard

One query is a hypothesis. A finding requires confirmation from a second angle. Strong null findings require effect sizes inside the noise floor across multiple cuts. All percentages are computed against population baselines, not raw counts.

Data Constraints

Several 4-way LEFT JOIN queries (e.g., a full subscriber-presence matrix across all 4 facts) timed out after 30 seconds. These were redesigned as per-fact aggregations joined post-hoc. The EXISTS pattern proved expensive on million-row fact tables; COUNT(DISTINCT SUBSCRIBER_KEY) per fact is efficient and was the workhorse.

Statistical Notes

For uniform-distributed counts at ~19,000 sessions per cell, the sqrt-counting noise floor is ±138 sessions = ±0.7%. Effect sizes below 1% are not distinguishable from noise. Several "candidate findings" (e.g., event-type × age slight orderings, acquisition channel slight differentials) were below this threshold and treated as null.

Findings

1The Cross-Domain Uniformity Paradox

Cross-domain participation in the Falcon Telecom & Media warehouse is essentially universal — but cross-domain intensity is demographically random. Across 14 different demographic and behavioral cuts tested, only one (plan type) produced a meaningful discriminator. Every other cut — segment, age band, credit band, tenure, acquisition channel, carrier, plan technology, region — returned spreads inside the statistical noise floor.

Cross-Domain Participation

Demographic Cuts That Showed No Differentiation

2No Carrier ↔ Platform Parent Affinity

The single most-tested cross-domain hypothesis in vertical-integration narratives — that subscribers of carrier X over-stream platform Y when X owns Y — fails completely in this data. Comcast (NBCUniversal) subscribers should over-stream Peacock; they don't. Charter subscribers (who distribute Discovery+) should over-stream Discovery+; they don't. T-Mobile (Apple TV+ bundle); AT&T (legacy WBD relationship) — none of these show detectable affinity.

Comcast Subscribers' Streaming, by Platform

Peacock is Comcast subscribers' lowest-streamed platform — though within statistical noise (sqrt-counting noise floor ±0.7%, observed deviation 1.2%). The same null pattern repeats for every prospect carrier across the 60-cell carrier × platform matrix.

35G Access Doesn't Drive 4K HDR Streaming Uptake

Telecom marketing routinely promises that 5G enables higher-resolution streaming. In this data, 5G plan subscribers and 4G LTE subscribers split streaming quality choices identically:

Identical to two decimal places. Stream quality selection is statistically independent of network technology in this warehouse.

4COVID Q2 2020 Cross-Domain Substitution

The single cleanest cross-domain story in the dataset is the COVID lockdown response. The same population that bought live-event tickets in Q2 2019 was streaming in Q2 2020.

Quarterly Cross-Domain Activity, 2019–2021

Three precise observations:

1. Ticket buying collapsed 91.3% Q2 2020 vs Q2 2019 (9,263 → 804 distinct buyers).
2. Streaming surged 41.8% in subs and 52.7% in minutes the same quarter.
3. By Q2 2021 ticket buying had not just recovered but exceeded the 2019 baseline by 12.4% — suggesting pent-up demand. Streaming returned to its pre-COVID trajectory after the surge (i.e., the elevated level was temporary).

Compared with the schema description ("8% of normal volume" for tickets and "+45% surge" for streaming), the actual numbers are slightly more extreme — 8.7% of normal for tickets, 41.8% surge for streaming subs.

5Heavy-Tailed Spending, Demographically Random

Lifetime ticket spend per active subscriber follows a heavy-tailed distribution with a 360× spread between the top and bottom 5%:

Top 20% generates 48.1% of total ticket spend; classic heavy-tailed concentration but not as extreme as 80/20.

The surprise: When you look at where the top-5% super-spenders live demographically, they are perfectly distributed across the customer base. Each segment has approximately 5% of its members in the top ventile:

0.23 percentage point spread — perfect uniform distribution.

6Enterprise Plans Skew Slightly Non-Purchaser

The only demographic discriminator detected in 16 iterations is plan type:

Enterprise Fiber subs are 47% more likely to never buy a ticket vs Prepaid 5G subs (3.45% vs 2.35%). The acquisition channel data confirms: Business Direct channel is the lowest-spending channel ($4,448/sub vs $4,591 for Dealer) — within 3.2% but consistent with the corporate-account hypothesis.

7Tenure Doesn't Predict Churn

Standard telecom churn modeling assumes a tenure curve — newer subscribers churn at higher rates ("honeymoon falloff"); long-tenure subscribers are sticky. This dataset rejects that assumption.

Identical to one decimal place. Combined with the near-uniform churn-reason distribution (Price 15.4% leads narrowly across 8 reasons in the dashboard build) and uniform churn-by-carrier rates, this strongly suggests the warehouse models churn as a memoryless process — independent of subscriber attributes.

What We Didn't Find (Beyond F2/F3/F7)

For completeness — patterns we hypothesized and tested but found no evidence for:

• Family segment over-indexing on Kids genre — 4.91% Family vs 4.77% Business Mid; effectively flat.
• Older audiences skewing toward Drama / News — Drama is 19.4–19.6% across all 6 age bands.
• Concert audiences skewing younger than WWE/UFC audiences — within 1% in transaction-share rankings.
• Acquisition channel × ticket spend differentiation — 3.2% spread; Business Direct slightly low, otherwise flat.
• Tenure × ticket buying intensity — 3.48 to 3.51 transactions per active sub across 4 tenure bands.
• VIP propensity skewed by credit band — uniform across Excellent / Good / Fair / Poor / No Check.

Recommended Actions

1. Reframe the demo narrative. Stop pitching this dataset's "demographic insights" — they don't exist beyond F6. Lead with cross-domain volume capability, the COVID substitution event (F4), and the heavy-tailed spending pattern (F5).
2. For prospect-specific demos, run the carrier-platform null test live. Showing F2 — "we tested whether your carrier subscribers over-stream your platform and the answer was no" — establishes credibility faster than presenting only positive findings.
3. Build a behavioral super-spender cohort definition. 4,404 subs in the top 5% drive $66.6M (16.7%) of ticket revenue. Define them by spend behavior, NOT by demographic profile, and add a "Super-Spender Cohort" view to the existing kit dashboards.
4. Document the COVID Q2 2020 substitution as a reference event. Add a "scenario library" doc explaining how the same SUBSCRIBER_KEY join produced cleanly visible substitution effects, to support prospect questions about black-swan modeling.
5. Do not promise demographic-based churn prediction on this dataset. Churn correlates with nothing measurable here. If pressed, demonstrate methodology only.
6. For the one real demographic signal (F6 — Enterprise plans), add an "Enterprise vs Consumer plan profile" comparison view to the Plan & Carrier Mix dashboard. It's small but real.

Limitations

• This is synthetic data. The uniformity findings reflect the data generator's design, not a real-world phenomenon. On a production warehouse, the same exploration would likely surface meaningful differentiation.
• Pre-churn behavioral signals not tested. A behavioral early-warning analysis (declining viewership/spend in months prior to churn) would require window functions on per-subscriber per-month aggregates. Given the universal null pattern, the expected result is also null, but it remains untested.
• Time-of-day patterns not tested. No exploration of intraday or day-of-week patterns in viewership or ticket purchasing.
• Rolling churn rate trends not tested. Churn count grew 7% YoY 2024 → 2025 (per dashboard data), but the rate adjusted for base growth was not formally tested.