Data advantage

A data advantage can strengthen with use when new activity produces useful information, a better decision changes the customer outcome, and that outcome attracts more relevant activity. If any link fails, volume alone does not compound quality. The Google court treated scale as one barrier to entry: a new search engine needs users to learn from and needs to be useful to attract them. Judge Mehta observed that Microsoft has run Bing since 2005 and bought Yahoo's search data in 2009, and Bing remains well behind in absolute scale.

The claim also has to be tested, and it can be contested. Google argued that Bing had reached diminishing returns, and its expert concluded that only 2.9% of the quality gap between the two engines came from the volume of user data. The court gave that study little weight, in part because Google had never used the finding for anything beyond the lawsuit: if less data produced the same quality, the company would have stopped storing so much of it. The remedy the court ordered in September 2025 is the strongest evidence of how it viewed the data. Google must share certain search-index and user-interaction data with qualified competitors, so that it does not keep the fruits of its exclusionary conduct. The order excludes advertising data.

The advantage is worth what the decision is worth. Data that changes a decision made millions of times a day has value even when each improvement is small. UPS expects its route-optimization system to cut 100 million miles and 10 million gallons of fuel a year, and drivers on ORION routes already average six to eight fewer miles a day. A saving of that size requires the company to act on the data: the drivers, dispatch systems and rules must change. Data without a workflow to use it is a report.

The practical test is the counterfactual. Ask what a competitor with public information and a good analyst would decide, then measure the gap. If the gap is small or cannot be measured, the company has data but not an advantage.

Zillow shows the failure most managers do not plan for: the data is real, the scale is real, and it answers the wrong question. Then the pandemic froze the housing market and prices began to rise at an unprecedented rate.

The Q3 2021 results show the pattern. Zillow bought 9,680 homes in the quarter and sold 3,032. It wrote down about $304 million of inventory because it had bought homes at higher prices than its estimates of future selling prices, and said it expected a further $240 million to $265 million of losses in Q4. Its own explanation is that higher conversion rates than it had seen before led it to purchase homes unintentionally at those prices. The letter lists a pandemic, a temporary freeze of the market and a supply-demand imbalance as the shocks, and the lesson is not that the data was bad. Knowing what homes had sold for, and what shoppers looked at, did not tell Zillow what the market would pay when it resold months later.

A behavioral dataset fits a problem where the system is stable and feedback is immediate: a search result gets a click within seconds. A market forecast has neither property. Buying at risk turned a data question into a balance-sheet question, and errors were paid for in cash. The company kept the audience and the Zestimate, and it said it would build asset-light ways to help people sell.

Advantages built on data also fail when the law changes what a company may keep or must share, as the Google order shows, and when rivals reach similar quality without matching the volume.