---
type: "synthesis"
sources: ["spine"]
tags: ["data-strategy", "moat", "contradiction", "flywheel"]
id: "cd-proprietary-data-moat-debate"
sourceVaultSlug: "hbr-seg-spine"
originDay: 1
articleStem: "hbr-seg-spine"
sourceUrl: "(unified vault: 9 sources)"
sourceTitle: "HBR — Strategic Spine — value thesis & how much to bet"
---
This is the corpus's sharpest head-to-head disagreement, and a downstream agent should surface it whenever data strategy comes up.

**A096 (Barney & Reeves): proprietary data is a WEAK moat.** Three mechanisms dissolve it: [[concept-functional-data-equivalence]] (different datasets yield the same patterns), the [[concept-data-saturation-point]] (1B ≈ 50M once the pattern is visible — [[contrarian-bigger-data-better]]), and [[concept-ai-strategy-inference]] (rivals reverse-engineer your strategy from public outcomes). Add [[concept-ai-first-mover-disadvantage]] ([[claim-early-movers-train-competitors]], [[quote-first-mover-training]]) and being early looks like a penalty ([[contrarian-first-mover-penalty]]).

**A047 (Prasad): proprietary data is among the STRONGEST moats.** [[concept-data-flywheels]] deployed in real operations create compounding proprietary data, massive switching costs, and lock-in ([[contrarian-proprietary-data-moat]] is exactly the claim Prasad rejects — see John Deere).

**Reconciliation.** They are less opposed than they appear. Both agree the *model* is not the moat. A096's caveats concede that private fine-tuning and closed feedback loops *can* break first-mover disadvantage; A047's flywheels are precisely such closed loops embedded in unique workflows. The synthesis: generic proprietary data is weak (A096 is right about static datasets); *dynamic, workflow-embedded, closed-loop* data can be strong (A047 is right about flywheels). A004's [[question-competitive-compression]] is the same debate in marketing form: how fast does the arbitrage window close? See also [[cd-ai-is-never-the-moat]].