Original source: Hyperight AB
This video from Hyperight AB covered a lot of ground. 2 segments stood out as worth your time. Everything below links directly to the timestamp in the original video.
If your organisation rebuilds its data infrastructure for each new project, it is effectively paying multiple times for an asset it already owns — and leaving the compounding returns on the table.
Data Assets Generate Compounding Returns at Zero Marginal Cost, Schmarzo Argues
A single customer dataset — purchase history integrated with loyalty records — can be deployed independently by sales, marketing, call centres, and product teams, each extracting distinct dollar value without incurring the cost of acquiring or rebuilding the underlying data. What this exposes is a structural asymmetry unique to data: unlike physical capital, which depreciates with use, the same dataset generates successive returns while its cost basis remains fixed. A retail example in the transcript illustrates a 2.5% improvement in promotional effectiveness for sales, followed by a separate 2% customer acquisition gain for marketing, both drawn from identical source data.
The structural issue here is that most organisations still treat data as a departmental resource rather than a shared economic asset — meaning they systematically undercount the returns already embedded in data they have already paid to collect.
"Marketing is going to achieve that ROI, that increase in value, at zero marginal cost — they're going to use the same data set that's already been set up and used by the sales organisation."
AI-Trained Models Appreciate With Use, Compounding Improvements Across Entire Fleets Overnight
Tesla's autonomous vehicle system, Schmarzo argues, demonstrates a fundamentally different economic logic for AI assets: every Tesla runs its autopilot in shadow mode even when not engaged, continuously learning from the driver, and each night that learning from roughly one million vehicles is aggregated in the cloud and pushed back to every car in the fleet. The result is that a single insight gained by one vehicle is effectively shared with 999,999 others at no additional cost — a compounding dynamic that, applied as a daily one-percent improvement over a year, yields a 38-fold gain.
The real question is not whether AI models can learn, but rather whether organisations understand that reusing analytic modules across use cases means any improvement to one module ripples through every application it serves — turning conventional depreciation logic entirely on its head.
"Any improvement that I make in an analytic module that is used by any other application or use case — any improvements to that analytic module ripple back through so that every one of those use cases just got more effective, and they got it at a marginal cost equal to zero."
Summarised from Hyperight AB · 18:58. All credit belongs to the original creators. Streamed.News summarises publicly available video content.
