Empirical Validation
BHI scores tested against real-world retention, churn, and engagement data across 42 platforms
Preliminary validation. The correlation analysis uses publicly available retention metrics from SEC filings, earnings calls, CIRP surveys, and industry analyst estimates. Sample sizes are small (n=14 for the primary correlation). Source quality varies. This constitutes evidence of predictive validity, not proof.
B Score vs Annual Retention
The 14 platforms with directly comparable annual retention rates show a strong positive rank correlation. Higher structural lock-in predicts higher retention — with meaningful outliers that validate the instrument's theoretical design.
Informative Outliers
The three largest deviations from the trend illuminate more than they obscure — each validates that BHI measures structural lock-in rather than simple retention.
Retention far exceeds what structural lock-in predicts. Users could leave easily — they choose not to. Product quality drives retention, not escape cost. BHI correctly identifies low structural lock-in.
Zero voluntary departures since 1973. Standards-based network effects may be underweighted in V3. China's CIPS processes < 15% of SWIFT volume despite years of operation.
Some users overcome Apple's structural lock-in due to price sensitivity. High switching costs don't guarantee 100% retention when economic pressure is strong enough.
Social Platforms: Lock-in vs Engagement
DAU/MAU measures daily habit formation — how compelling a product is. BHI measures structural lock-in — how difficult it is to leave. These overlap but are not identical.
Structural lock-in without daily engagement. Professional identities, endorsements, and recruiter access create real switching costs — but the platform serves an episodic purpose.
Extremely high engagement but low structural lock-in. Algorithmic feed creates compulsion but no data lock-in — no social graph, no stored content library, no workflow dependency.
Within-Category Rank Alignment
The most methodologically sound approach tests BHI within categories where platforms share comparable metrics. These within-category rank-order alignments are striking.
Standardized monthly churn data from Antenna. Perfect rank-order alignment (n=3).
Renewal rates and buyer trajectory. BHI rank and retention rank are perfectly aligned.
Retention rates and account trajectories. The two lowest-BHI fintech platforms show by far the worst outcomes.
Full Dataset: 42 Platforms
Every (Platform, B, Observable Metric) pair recovered from public filings, industry data, and analyst estimates. B scores computed in real-time via V3 formula.
| Platform | Category | B | Metric Type | Observable Metric | Source |
|---|---|---|---|---|---|
| Synopsys/Cadence | Infrastructure | 12.97 | Annual Retention | ~99% | High |
| TSMC | Infrastructure | 10.91 | Annual Retention | ~99% | Medium |
| Visa | Infrastructure | 8.82 | Annual Retention | ~99.9% | High |
| Amazon Prime | E-Commerce | 8.40 | Annual Retention | 93% / 98% | High |
| NVIDIA CUDA | Infrastructure | 8.29 | Annual Retention | ~92% share | High |
| Bloomberg Terminal | Infrastructure | 7.21 | Annual Retention | ~96% | Medium |
| AWS | SaaS | 6.85 | Growth Proxy | 17–20% YoY | Low |
| SWIFT | Infrastructure | 6.78 | Annual Retention | ~100% | High |
| MercadoLibre | E-Commerce | 5.87 | Growth Proxy | Record levels | Medium |
| Social | 5.43 | DAU/MAU | ~83% | Medium | |
| Steam | Gaming | 5.32 | DAU/MAU | 52% DAU/MAU; 75% share | High |
| Apple Ecosystem | Tech | 5.22 | Annual Retention | 89% | High |
| Nubank | Fintech | 5.21 | Annual Retention | 94% | High |
| Solana | Crypto | 4.83 | Growth Proxy | 70%+; 83% dev growth | Medium |
| Roblox | Gaming | 4.79 | Growth Proxy | 151.5M DAU (+70% YoY) | High |
| Microsoft 365 | SaaS | 4.73 | Annual Retention | 98% | Medium |
| Ethereum | Crypto | 4.57 | Growth Proxy | 59–68% TVL; 31,869 devs | High |
| Salesforce | SaaS | 4.54 | Gross Retention | ~92% | High |
| Adobe Creative Cloud | SaaS | 4.38 | Growth Proxy | 11.5% YoY ($25.2B ARR) | Medium |
| Social | 4.19 | DAU/MAU | 69% | High | |
| Coupang | E-Commerce | 3.72 | Growth Proxy | 24.7M (+10% YoY) | High |
| Social | 3.70 | DAU/MAU | 16.2% | Medium | |
| Stripe | Fintech | 3.67 | Growth Proxy | 38% YoY ($1.4T) | Medium |
| Snowflake | SaaS | 3.67 | NRR | 126% | High |
| Shopify | SaaS | 3.15 | NRR | ~100% | Low |
| Social | 3.07 | DAU/MAU | ~55% | Low | |
| Shopee | E-Commerce | 3.03 | Growth Proxy | 400M; +40% actives | Medium |
| Revolut | Fintech | 2.87 | Growth Proxy | +59% YoY | Medium |
| Block / Square | Fintech | 2.81 | Gross Retention | >100% | High |
| Google Workspace | SaaS | 2.68 | Growth Proxy | 28–48% YoY | Low |
| Slack | SaaS | 2.52 | Annual Retention | >98% | Medium |
| Spotify | Streaming | 2.31 | Annual Retention | ~96.5% | High |
| TikTok | Social | 2.24 | DAU/MAU | ~57% | Low |
| Social | 2.15 | DAU/MAU | 26% | High | |
| Bitcoin | Crypto | 1.82 | Annual Retention | 70%+ | High |
| eBay | E-Commerce | 1.71 | User Trajectory | −25% from peak | High |
| X / Twitter | Social | 1.29 | User Trajectory | −5.3% YoY | Medium |
| Robinhood | Fintech | 1.16 | User Trajectory | −52% | High |
| PayPal | Fintech | 1.14 | User Trajectory | −2% then recovery | High |
| Netflix | Streaming | 1.00 | Annual Retention | ~78% | High |
| Zoom | SaaS | 0.90 | Annual Retention | 98% NRR; 2.7%/mo online | High |
| Disney+ | Streaming | 0.71 | Annual Retention | ~61% | High |
Methodology Notes
BHI scores and retention rates have different scales and non-linear relationships. Spearman rank correlation is appropriate because it tests monotonic association without assuming linearity. Infrastructure platforms (EDA B≈13.0, TSMC B≈10.9) sit far above the rest of the dataset, creating a non-linear relationship that Spearman handles correctly.
Observable retention = f(BHI) + product quality premium + market conditions. High BHI guarantees high retention. Low BHI does not guarantee low retention — excellent products can retain users despite zero structural lock-in. This is correct behavior for an instrument measuring structural lock-in.
Small sample (n=14 for primary correlation). Heterogeneous metrics across categories. Source quality varies from SEC filings (High) to analyst estimates (Low). Single-evaluator scoring of all BHI parameters. Cross-category comparisons mix metric types. Full validation requires 30+ platforms with standardized metrics and multiple independent evaluators.
Correlation does not establish causation. BHI may capture a latent variable correlated with retention rather than causing it. The theoretical model (structural lock-in → retention) is plausible but not proven by correlation alone. Longitudinal validation — tracking B-score changes and subsequent retention changes — would provide stronger evidence.
Full methodology: Methodology · Explore scores: Observatory · Academic paper: Paper