Original source: Bill Schmarzo
This video from Bill Schmarzo covered a lot of ground. 3 segments stood out as worth your time. Everything below links directly to the timestamp in the original video.
The difference between an AI model that becomes more valuable over time and one that silently corrupts your decisions may come down to a single architectural choice made at the design stage.
Tesla's Fleet Learning Model Reframes AI Assets as Appreciating, Not Depreciating
Most machine-learning models built inside organisations quietly become liabilities: they drift, decay, and produce wrong outputs once the engineer who built them has moved on — a phenomenon Schmarzo terms 'orphaned analytics'. Tesla's architecture inverts that logic. Each of the roughly one million vehicles on the road continuously learns from every mile driven, uploads those learnings nightly to a central cloud, and receives back the aggregated intelligence of the entire fleet — meaning a lesson learned by one car is instantly inherited by all others. Compounding one-percent daily improvements, Schmarzo calculates, yields a 37.8-times gain in value over a year.
What this exposes is a structural failure in how most organisations treat analytical assets: they engineer for the initial problem, not for perpetual reuse. The Tesla model suggests the real competitive advantage lies not in building models, but in building systems that make models self-sustaining.
"You can engineer these things so they continuously learn and adapt — they appreciate in value the more they are used, not depreciate."
Precision Data Modelling Could Break Healthcare's Law of Diminishing Returns, Schmarzo Argues
The law of diminishing returns — where each additional dollar spent on maintenance, healthcare, or marketing yields progressively less improvement — is not an immutable ceiling but a symptom of decision-making based on averages, according to Schmarzo. His proposed corrective, which he calls 'nano-economics', uses data science to build granular profiles of individual humans and devices, capturing propensities, behaviours, and risk scores precise enough to direct resources at the exact point of highest impact. Applied to COVID-19 vaccine distribution, he argues, the failure to identify and prioritise the individuals most likely to catch and spread the virus — defaulting instead to population-wide averages — directly extended the pandemic and cost lives.
The structural issue here is that average-based policy is computationally convenient but economically wasteful; the gap between what precision targeting could deliver and what blunt averages actually achieve grows wider the more heterogeneous the population.
"We have caused this pandemic to extend itself because we haven't taken a very focused data science approach to understanding who are the people most at risk."
Data's Zero Marginal Cost of Reuse Creates an Economic Multiplier That Accounting Cannot Capture
Unlike physical assets, data never depletes, never wears out, and can be applied to an unlimited number of use cases at zero marginal cost — a property that, Schmarzo recounts, his research team at the University of San Francisco concluded had no precedent on any corporate balance sheet. That structural peculiarity produces what he calls an 'economic multiplier effect': a single customer point-of-sale dataset can simultaneously improve promotional effectiveness for the sales team, reduce acquisition costs for marketing, and lower churn in the call centre, each application generating independent returns from the same underlying asset. The real constraint, he argues, is not data scarcity but data siloing — organisations that hoard or fragment data destroy the very multiplier that makes it valuable.
The real question is not how much data an organisation owns, but how many use cases it can route through the same dataset — a framing the IMF, he notes, describes as building a 'collaborative value creation platform'.
"Data silos are your enemy. Data you can't share or won't share — that's the great destroyer of economic value."
Summarised from Bill Schmarzo · 55:49. All credit belongs to the original creators. Streamed.News summarises publicly available video content.
