Venture capital data habits are still too weak, former Spotify VP says

Venture capital data practices remain too dependent on instinct and after-the-fact review, according to Henrik Landgren, Spotify’s first vice president of analytics. In a Fortune commentary, Landgren said many investors still lack the data culture and infrastructure needed to judge startups with the same discipline that helped Spotify scale.

Landgren said he led Spotify’s data strategy under Daniel Ek before the company had entered the U.S. market, when streaming music was still early. His task, he wrote, was to shift the company away from static Excel files and toward systems that could track user behavior in more timely and useful ways.

The early tools were slow by current standards. Landgren said Spotify used Hadoop, and analysts could spend hours waiting to learn whether a query had been built correctly. Even so, he said the system let Spotify study where users clicked and where they spent time, then bring that information into product and business decisions.

That transition took years, Landgren wrote, but he linked it to a more precise way of running the platform as Spotify grew past 20 million subscribers. The central lesson, in his account, was that judgment improved when leaders had evidence to test it against.

What does Spotify's former analytics lead say VCs get wrong about data?

Landgren said he entered venture capital about half a decade later and found that many investors had not applied data to their own decision-making in a serious way. He described Stockholm’s startup boom after Klarna and Spotify, when investors often relied on stories about charismatic founders and fear of missing out on the next major company.

He did not argue that venture investing can be reduced to a formula. Landgren wrote that reading founders and markets is partly an art, but he said that same instinct can hide mistakes when firms remember the wins and give less attention to the failed bets.

He cited Deutsche Bank research saying nearly two-thirds of venture deals do not return the capital originally invested. Landgren said many firms keep post-mortems on their misses, but those records often do not influence the next investment decision.

Landgren pointed to Bessemer Venture Partners’ public anti-portfolio as a rare example of a firm openly listing famous opportunities it missed, including Apple before its initial public offering. He said the broader industry tends to use data to support decisions after they have been made rather than to shape them beforehand.

Why does data struggle to change VC decisions?

Landgren attributed part of the problem to how venture firms are organized. In his view, data teams often sit outside the core investment group, with limited power to affect whether a deal happens.

At his first venture firm, Landgren said he helped build a system called Motherbrain, named after the Nintendo character. Before large language models became widespread, he said the tool looked at signals such as traction, product use and web traffic to identify potential investments earlier.

Landgren said colleagues later estimated that more than $100 million in the fund’s investments had been influenced by Motherbrain, a figure also reported by Crunchbase News. He said attempts to improve tracking at portfolio companies exposed another problem: many startups did not have the systems needed to capture useful performance data.

That gap matters more as investors and startups adopt AI, Landgren argued. He said companies that could not track their own signals are now adding large language models to systems that were not designed to supply them with reliable data, while venture firms are making similar moves.

Landgren’s conclusion was that founders and investors need a shared factual view of company performance. Without better data foundations, he argued, AI tools are unlikely to solve the basic problem of decisions made on partial information.

This story draws on original reporting from Fortune.