The 70-20-10 Innovation Allocation Rule
Structure innovation into three separate buckets — the 10% bet is where paradigm returns live
Schmidt describes the 70-20-10 rule as the structural solution Google used to resolve the tension between sustaining existing businesses and pursuing disruptive innovation. The allocation: 70% of resources to the core business, 20% to adjacent opportunities, and 10% to genuinely experimental bets. The framework is not about percentages — it is about separation: each bucket must operate with distinct incentive structures, or the 70% bucket's rewards will always crowd out the 10% bucket's risks.
The ROI asymmetry is extreme. Google Brain — the 10% bucket experiment run by a team of ten to fifteen people — produced the transformer architecture and the deep learning stack that the entire AI industry now runs on. Schmidt estimates this generated ten to forty billion dollars in extra profits for Google over a decade. This is the canonical example of why the 10% bucket is where paradigm-shifting returns live, not the 70% bucket where most resources are concentrated.
The framework has a direct implication for AI platform analysis: companies running open-source AI models (such as Meta's LLaMA) are executing an offensive version of the 70-20-10 logic — giving away the 10% bucket's output to dominate the 70% bucket's commercial returns through advertising targeting and feed algorithms. Understanding which AI companies are in which bucket for which incumbent predicts what they will and won't protect commercially.
- Innovation and harvesting cannot share organisational incentive structures — physical and structural separation is required, not just departmental labelling
- The 10% experimental bucket generates paradigm-level returns that dwarf the core business when it succeeds — the ROI asymmetry justifies the allocation even if the 10% fails often
- The best use of the 10% bucket's output can be offensive: giving it away (open source) to dominate the 70% bucket's commercial returns
- Disruptive innovation requires freedom from the constraints that protect the core business — those constraints are the mechanism by which incumbents lose to challengers
- Almost no examples exist of disruptive and sustaining innovation succeeding simultaneously inside the same building
- Audit current resource allocation across the three bucketsMap where engineering, capital, and leadership attention actually flow — not where the budget says they go. Most organisations nominally run a 70-20-10 but operationally run a 95-4-1 because the 70% bucket's incentives crowd out the rest.Pro tipUse headcount and management attention as proxies, not budget lines — the 10% bucket is chronically under-staffed relative to its nominal allocation.
- Establish structural separation for the 10% bucketPhysically and organisationally separate the experimental team from the core business. Different managers, different success metrics, different reward structures. The pirate-flag test: would the 10% team's failure be a genuine career risk for the 70% team's managers? If yes, they are not separated.WarningNominal separation (a different team name, the same reporting line) does not work — the incentive contamination happens through performance reviews, not org charts.
- Set asymmetric success criteria for each bucketThe 70% bucket is measured on revenue efficiency and margin. The 20% bucket is measured on adjacent market penetration. The 10% bucket is measured on learning velocity and paradigm-shift potential — not near-term revenue. These criteria must be explicitly different or the 70% metrics will colonise all three buckets.Pro tipSchmidt's Google Brain example: the team was not measured on advertising revenue. It was measured on research output. The revenue came a decade later.
- Identify whether the 10% output should be given away or protectedOnce the 10% bucket produces a result, determine whether open-sourcing it accelerates the 70% bucket's commercial dominance. Meta's LLaMA strategy is the current canonical example: open-sourcing frontier models commoditises the model layer while Meta's advertising targeting (70% bucket) becomes the beneficiary of industry-wide AI adoption.Pro tipAsk: does widespread adoption of this technology benefit our 70% bucket more than our 10% bucket's exclusivity? If yes, open-source is the offensive play.
Schmidt created Google X as the institutional home of Google's 10% bucket. Its first project, Google Brain, was a team of ten to fifteen researchers operating with complete separation from Google's core advertising business. They were not measured on ad revenue. They pursued deep learning research that no core Google team would have prioritised given existing incentives.
The Mac team was physically separated from Apple's main building and flew a pirate flag — a literal signal of operating outside the main company's rules and constraints. Steve Jobs deliberately created a separate culture, separate incentives, and separate success metrics for the team.
Schmidt developed and applied this framework during his decade as Google CEO (2001–2011) and continued to refine it as Executive Chairman. Google X — the Moonshot Factory — was the institutional embodiment of the 10% bucket. Its first product was Google Brain, run by a small team operating entirely outside Google's core business incentive structure.
Schmidt cites the historical precedent of the Macintosh: Apple's Mac team was physically separated from the main building and flew a pirate flag above it. The separation was not symbolic — it was structural. The same manager cannot rationally risk their core business to run an experimental bet. Schmidt uses this to explain why the Macintosh succeeded where internal Apple projects didn't, and why Google Brain succeeded where Google's core engineering teams (optimising for search) would not have pursued the same research direction.