framework 02 · enterprise transformation

Decision Dominance

Most bad decisions in a large company aren't made by bad decision-makers. They're made by good people working from models that don't line up.

By · V-Squared AI

Most bad decisions in a large company aren't made by bad decision-makers. They're made by good people working from models that don't line up. The product team's picture of the customer isn't sales'. The C-suite's read of what technology can do isn't engineering's. Everyone is being rational inside a different map.

One map instead of many

Decision Dominance is what you build when the disconnected, competing maps become one. When every group that needs to decide is working from the same shared model of how customers choose, how the business makes money, and how technology gets built and monetized, decisions stop colliding and start compounding. Speed goes up because you're not re-litigating the underlying model every time. Quality goes up because the model is explicit enough to argue with.

The payoff of removing three bottlenecks

Remove the Understand bottleneck and every group sees the same model. Remove the Work With bottleneck and they can move that model between each other without it getting lost in translation. Remove the Align bottleneck and strategy, implementation, and execution point the same direction. Decision Dominance isn't a governance layer you bolt on top. It's what's left once the bottlenecks between groups are gone.

Decision Dominance isn't governance you bolt on. It's what's left when the bottlenecks between groups are gone.
the evidence · earnings calls, july 2026

The learning machine, described from the inside.

Decision Dominance is what you get when every group decides from the same current model instead of from divergent private guesses. Three of the four July calls describe organizations building exactly that, and naming it as the thing they intend to sell.

MicrosoftFY26 Q4 · Jul 29, 2026

Every firm is a learning machine that needs its own learning machine

Nadella's framing was that a company should control its own destiny by building both human capital and token capital, and that the year's goal is helping every organization build a continuous learning loop rather than handing its knowledge to somebody else's model. He was explicit that this should not be an arrangement where a vendor takes the knowledge and the enterprise gets nothing back.

what it confirms

This is the clearest external statement of the mechanism behind Decision Dominance. A learning loop that belongs to the enterprise is what keeps every group's model current at the same time, which is the only way decisions stop colliding.

MetaQ2 2026 · Jul 29, 2026

An allocation decision made from one shared model of return

Susan Li told analysts Meta is demand-constrained rather than opportunity-constrained, and that there are numerous ROI-positive places the company would put compute if it had more, including inside the core business. Capital is being rationed against an internal ranking of expected return.

what it confirms

A ranked queue of opportunities that different groups can argue with is a shared model in the sense this framework means. Whether it produces Decision Dominance depends on whether the ranking is visible to the groups that have to execute against it, which is exactly what an outside observer cannot see and an insider must.

MetaQ2 2026 · Jul 29, 2026

Rebuilding the decision layer itself on one model of the customer

Li described a roadmap toward LLM-native recommender systems, reported healthy scaling laws from continuously pre-training on recommendations data, and said every public Reels and feed post on Instagram now passes through an LLM. Ranking agents shipped an increased number of launches over the half.

what it confirms

The most consequential recurring decision in the business, what to show whom, is being moved onto a single current model of the customer. That is the pattern generalized: not a governance layer bolted on top, but the decision itself relocated onto the shared model.

Sources: Alphabet Q2 2026, Microsoft FY26 Q4, and Meta Q2 2026 earnings calls and releases, plus Amazon Q2 2026 results reported July 30, 2026. Figures are as stated by company executives on those calls. Amazon reported after market close on July 30, so Amazon figures here come from the release and initial call remarks rather than the full transcript.

the wider evidence · measured long before the AI cycle

The shared model has been measured, and it has an effect size.

Decision Dominance claims your decision quality is limited by model divergence, not by the quality of your decision makers. That claim is testable, and it has been tested in the management literature and in large sample firm data for more than a decade.

peer-reviewedsurvey researchcompany filingsmarket dataindependent audit
Brynjolfsson, Hitt & Kimpeer-reviewed2011 · 179 public firms

A 5% to 6% productivity premium for deciding from the model

The authors studied business practices and IT investment across 179 large publicly traded firms. Firms that adopted data driven decision making showed output and productivity 5% to 6% higher than their other investments and technology usage predicted. The effect showed up again in asset utilization, return on equity, and market value. Instrumental variable methods indicated the result wasn't explained by reverse causality.

Brynjolfsson, E., Hitt, L. M., and Kim, H. H. Strength in Numbers: How Does Data-Driven Decisionmaking Affect Firm Performance? ICIS 2011 Proceedings.
what it confirms

Technology investment is controlled for here. The premium isn't for owning data or buying analytics. It's for the practice of deciding from an explicit model. Most companies fund the first two and skip the third.

DeChurch & Mesmer-Magnuspeer-reviewed2010 · meta-analysis, 23 studies

Shared structure predicts performance, not shared facts

A meta analysis of 23 independent studies found shared mental models positively related to both team process and team performance, and the relationship held across measurement approaches. The strongest results came from cognitive mapping and content analysis. Knowledge structure predicted both process and performance. Knowledge content predicted performance alone.

DeChurch, L. A. and Mesmer-Magnus, J. R. Measuring shared team mental models: A meta-analysis. Group Dynamics: Theory, Research, and Practice, 2010.
what it confirms

Groups don't align because they hold the same facts. They align because they organize those facts the same way. A shared model has to be explicit enough to argue with, and circulating a dashboard doesn't do it.

McKinseysurvey research1,200+ managers

Only 20% of organizations think they decide well

Across surveys of more than 1,200 managers at global companies, 20% said their organizations excel at decision making and 37% said decisions were both high quality and fast. Respondents who reported fast decision making were 1.98 times more likely to also report high quality. At organizations with one to three reporting layers, 70% reported high quality decisions, against 45% at organizations with seven or more.

Sources: McKinsey Global Surveys on decision making, reported in Decision making in the age of urgency and Three keys to faster, better decisions.
what it confirms

If speed and quality traded off, they would correlate negatively. They correlate positively, which is what happens when the real constraint is a shared model rather than deliberation time. The reporting layer finding says the same thing from the other direction. Every layer is another chance for the model to change in transit.

Brynjolfsson & McElheranpeer-reviewed2016 · US Census plant data

The practice spreads unevenly, which is what makes it an advantage

Using US Census Bureau data on a large representative sample of manufacturing plants, adoption of data driven decision making nearly tripled from 11% to 30% between 2005 and 2010. Adoption correlated with size, with complements like information technology and educated workers, and with firm learning.

Brynjolfsson, E. and McElheran, K. The Rapid Adoption of Data-Driven Decision-Making. American Economic Review 106(5), 2016.
what it confirms

An advantage requires that a practice be available and unevenly taken. Both were true then and both are true now. The complements finding matters too. Adoption tracks whether the surrounding capability exists, which is why buying the tool without building the shared model produces nothing.

The two Brynjolfsson papers and the DeChurch meta analysis all predate the current AI cycle. A mechanism that was measurable before the technology arrived is a mechanism, not a trend.