We Need To Save Venture Capital From Bad Data

By Henrik Landgren

Investing, particularly venture capital, is 50% science and 50% art. The industry relies heavily on charisma and the founders’ “it” factor. That criteria warrants plenty of merit; one shared truth among all my investor colleagues is that the greatest founders of our generation have an unmistakable drive and dedication to their craft that is near impossible to put a finger on.

But here is how the process actually works once the charming visionaries have been identified. When an investor meets a promising founder and decides to take a closer look, they are handed a vast collection of data.

For many now, harnessing AI to sift through what is relevant feels like the only sensible response. The data is massive, but cherry-picked and packaged by the founder — the crux of the information asymmetry problem that underlies the entirety of the VC model.

Today, when pitch decks and company websites can be vibe-coded in a single afternoon, and data slicing is aided by the world’s most powerful AI models, it becomes progressively more difficult for investors to cut through this noise, question what they are seeing, know how to make sense of the data and, more importantly, where to get it.

Is there another way?

During my time as VP of analytics at Spotify, I became something of a data obsessive. The hottest new thing at the time was technology that allowed us to move away from Excel spreadsheets and actually make decisions based on more granular data.

We used software that helped us store every click a user made. This level of detail changed how we operated completely, and it changed how I viewed data: there was simply much more of it inside a company than most of us had realized.

When I moved into investing, it was a shock to me how far behind the industry was, comparatively. Today, as every financial institution scrambles to prove it has an AI strategy, the pressure to do something visible with the technology overrides the desire to do something useful with it. The way most investment teams adopt AI today is, to put it charitably, misguided.

An AI stack is not a checkbox to be ticked off. Using an LLM to cut down the time needed to put together reports or summarize pitches is instantly gratifying, but seldom changes the efficiency of your work in the big picture. Doing the same thing faster is easy; taking the steps to eliminate unnecessary processes and rethink your workstreams is harder.

Garbage in, garbage out

At the core of the AI adoption challenge is the truism that an LLM is only as good as the data that goes into it. The right approach starts before the AI. It starts with the data: payment records, marketing performance, accounting systems and board reports; each adds a layer to your diligence that reflects how a company actually operates, not how a founder or their analysis team describes it.

Investors should aim higher than faster reports, toward a position where a company’s hidden faults and rough edges are visible to them before the founder decides they should be. If you’re buying a house, you want it surveyed independently, not by the homeowner. Investors should do the same for their portfolio. What this means in practice is going directly to the data source; plugging into a company’s financials rather than waiting for a founder to package them for you.

This fundamentally shifts how investors understand risk. Suddenly, instead of a standing start, the analyst can begin at 70%, with specialist time reserved for the tasks that actually require human judgment: understanding the team, reading the room, assessing the “it” factor.

Looking ahead

This is not going to stop investors from investing — quite the opposite. Consider the capital-efficient, high-retention businesses whose growth isn’t considered aggressive enough, or whose vertical has fallen out of VC fashion.

Those companies get passed over completely because the data that could make a confident case for them isn’t accessed fast enough, or at all. Given the tools to reassess their actual risk level, financial institutions can be more confident in funding such businesses.

More immediately, having better data access makes you more competitive as an investor. In the most attractive deals, the kind all VCs want, the advantage goes to whoever reaches conviction fastest. This means the difference between issuing a term sheet a day before your competitors, without spending a week locating and cleaning up data, is deal-making.

Add to it the fact that, in five years, the companies worth funding will look nothing like the companies we fund today: AI-powered hardware, infrastructure and new categories of deep tech will require a reassessment of how we evaluate performance, with traditional income models and historical indicators of success no longer measuring up.

The industry needs to stop asking how AI can speed up the existing process and start asking what a better process looks like. Better data infrastructure is not a technical nice-to-have. It is the precondition for everything else. The only way to fund the next generation of transformative companies well is to build the infrastructure that lets investors see their work clearly. Anything less, and we are just making the same blind bets but faster, with a model that tells us it is confident.

Henrik Landgren is co-founder and CPTO at AI investment intelligence platform Gilion. He was previously Spotify’s first VP of analytics. At Gilion, Landgren leads development on the firm’s AI-powered investment platform, which is transforming investment and business growth intelligence for venture capital, private equity and banking.

Illustration: Dom Guzman