August 13, 2020

NGP turns to AI to find hot startups — and shares part of the tech

Nokia's venture capital arm, NGP Capital, is sharing some parts of the data system it created to help find promising startups faster.

Maija Palmer

4 min read

Bo Ilsoe, partner at NGP Capital

Nokia’s venture capital arm NGP Capital has created a machine learning system to find promising startups — and has made some parts of it open source for others to try as well.

It is part of a move to make VC investing more data-driven, rather than a run on gut instinct, says Bo Ilsoe, partner at NGP Capital and the force behind the project.

“VCs have been quite arcane in the way we work, but now that has started to change as more data becomes available,” he says.


NGP has spent 18 months building an AI system to help it sift through potential investment targets faster, and is willing to share part of the system for free. NGP Capital isn’t giving away access to the AI system — known as Q — as such.  You would have to plug in your own data and algorithms. “The secret sauce is how you use the data and what algorithms you apply. We’re not giving that away,” says Ilsoe.

But they are sharing the pipeline management and customer relationship management tool they have built, hich can serve as a first stepping stone for VCs wanting to build their own AI-investment system. Anyone can download it here and begin trying it today.

“We wanted to see what people do with this and start more of a dialogue around data-driven investing,” says Ilsoe.

Human vs machine

Using AI to find investment targets is not particularly new. European VCs like InReach Ventures, Blossom Capital and EQT, as well as SignalFire in the US, have been using AI tools — rather than networks and scouts — to source deals for several years.

VCs have been quite arcane in the way we work, but now that has started to change as more data becomes available.

But these tools have typically been highly custom-made and expensive. InReach spent more than £3m and three years to build its system. Ilsoe is hoping that by making its platform — called Q — available, it will widen the take-up of data-led investing.

NGP Capital has spent 18 months developing the platform and Ilsoe estimates that now around 30% of the deals NGP evaluates have been surfaced by the system. Eventually, he expects this to go up to 90%.

“It has brought companies to our attention that we wouldn’t normally have looked at. Maybe it was in an area adjacent to where we would normally invest. But you see a strong growth spurt at a business and start thinking about ways it might feed into your strategy.”

Get in on the deal early

Unlike, say InReach Ventures, which uses its system to find startups at the very earliest stage before anyone else, NGP typically invests at the series B round stage and beyond, when companies are starting to scale up and expand. Ilsoe says the trick is to catch companies like this just at the right inflection point.

Most companies' funding rounds are preemptive these days and it is important to be in there early.

The machine learning system is typically scanning for things like increased news flow, employee growth, even relatively “informal” data such as Glassdoor ratings and tweets. There are around 300 parameters tracked by the system, and NGP is keeping tabs on some 700,000 companies globally in this way.

“It’s not the machine that does the deal — that is still down to humans. But it helps us get in front of the funding round. Most companies' funding rounds are preemptive these days and it is important to be in there early,” says Ilsoe.


More or less bias?

As for whether an AI-based investment-finder would make things more or less biased, Ilsoe is unsure. On the one hand, a system that surfaces companies based on their performance rather than the network and personality of the founder might help VC’s go beyond the Old Boys Network.

On the other hand, AI has been shown in many cases to replicate bias, as was the case when an Amazon hiring algorithm chose only male candidates.

“It is something we have talked about and looked at a lot,” says Ilsoe. “One thing we have tried to do is to make the system explainable, so it is transparent how certain fields are weighted.”