In a venture capital firm, you want different talents that will enrich the investing team, such as a person from industry, say, mixed with people from the finance world, and perhaps people with a legal or public policy background.
You may even want an automaton that crunches numbers.
"Motherbrain" is the name that Henrik Landgren, operating partner, and his colleagues at venture capital firm EQT Ventures have given to the computer program that they increasingly turn to in order to get an early read on potential investments.
Motherbrain uses convolutional neural networks, or CNNs, the most popular form of machine learning, to review time-series data about companies to help guide where the firm should invest. The technology has seriously improved EQT Ventures's ability to scope out deals early in the pipeline, Landgren said in an interview with ZDNet.
"Much of it is just learning what is the best process we should pursue, what should we assess and in what order," he explains, "to increase the capacity of how many companies we can look at."
EQT Ventures is part of a much larger organization, the 25-year-old private equity firm EQT, which is run as a partnership and based in Sweden. The Ventures arm has raised €566 million in funds to invest. EQT, which has raised roughly $50 billion for investments over the years, is rumored to be contemplating a listing on the Swedish stock exchange, according to a recent Reuters report.
Being a young outfit -- EQT Ventures was only formed three years ago -- the team is expressly focused on sourcing deals. There are millions of companies the world over, explains Landgren. With just thirty team members, and even with access to EQT's broader workforce of 540 people, the search for prospects can be a daunting task.
With Motherbrain, the firm has assessed 10,000 companies. Much of the leverage is just knowing what firms a human should look at.
"Of all the 'Nays' we do," he says, referring to the deals ultimately passed on, "99% of the cases, we could cut out that time earlier and assess a new company instead."
While it's too soon to speak of how Motherbrain has improved the investment "yield" for EQT Ventures, Landgren is convinced the program is helping to avoid his team missing opportunities.
"There are companies we would not have found, and companies we would not have prioritized, without Motherbrain," says Landgren, who speaks of the computer program at times as if it is a new associate on the team. Landgren himself joined in 2016 from Spotify, the streaming music company, where he was vice president of analytics. He started computer programming when he was six, but at some point realized, "I was not going to be a coder all my life; I'm more interested in the uses of technology."
One investment helped by Motherbrain is the German software virtualization company AnyDesk, which makes tools to remotely view a desktop.
"Thanks to Motherbrain, we saw that earlier. It was a company that wasn't even interested at the time in investment, they were really crushing the product," he reflects. "We met the guys at the right time so we could build a relationship with them; without Motherbrain, we might have seen them much later, and so we might have gotten in way too late."
Other investments are an interesting assortment -- you can check out the full list on the Ventures website. There are many in the enterprise market, such as Acerta, which is developing AI for the auto industry; and Bimobject, software for "building information management" in construction. There are some in consumer, such as Natural Cycles, which offers women something called "digital contraception"; and some that may lie somewhere in-between, such as Mental Canvas, which makes a novel drawing app that lets one sketch drawings that turn into 3D wireframe models.
Because, again, EQT Ventures is young outfit, one might suspect they had a sparsity of training examples for Motherbrain. After all, the company has only had about 40 investments so far, and it has had only one exit, a sale of the Finnish games maker Small Giant Games, to the US firm Zynga, in late December. That means that typical signals for the neural network are less prevalent than they would be with an operating track record that is longer.
But that hasn't stopped Landgren and the team from supplying lots of information to Motherbrain to fuel insights, he says. "For every company, there is time series of performance data for the last several years, and we can overlay the relationships of who has invested in it, how did all their companies do," he explains.
"You don't have to have exists, you have things such as follow-on rounds [of funding] and huge valuations, and you use that to feed the algorithm." As a result, "we have enough test data now with an objective function," the measure of success in a neural network that lets one train the system.
"We are continually trying out new methods, continuously adding new algorithms to Motherbrain, for different types of signals, layering in test mining methods, for example."
Landgren notes a recent staff discussion about possibly adding graph network analysis to Motherbrain. "Just last week, we talked about a very recent paper to combine graphs into convolutions, that was super cool," he says, referring to a post on Medium by one Tobias Skovgaard Jepsen, an outside researcher.
An immediate result of Motherbrain's involvement is the division of labor: humans will spend more time on some tasks that demand their sense of uncertainty, after Motherbrain has shown where that uncertainty lies.
- 'AI is very, very stupid,' says Google's AI leader (CNET)
- How to get all of Google Assistant's new voices right now (CNET)
- Unified Google AI division a clear signal of AI's future (TechRepublic)
- Top 5: Things to know about AI (TechRepublic)
"Everyone will get better as we get this data back," says Landgren. "We can see how we can completely automate a lot of the work we do, and spend our time as humans on stuff we should do be focusing on, such as relationship building."
And what of their new associate, the machine: Can EQT Ventures trust AI to not lead them astray? Asked if he can interpret why Motherbrain brings them to a given investment, Landgren is frank, admitting, "We can't really say exactly for sure."
Part of knowing is building the system, constantly, with new capabilities, being intimately involved with its progression as a program, he suggests.
Another part is just seeing how it performs until a certain confidence level is reached.
"It's about trust to a large extent," he concludes.
"If you have the algorithm flagging companies and you feel a good portion of the time you agree with its decision, that this is a good company, or this other one is not, then you build up trust, then that is enough for the team to say, I can trust this to be a good ranking or prioritization," he says of Motherbrain.
And, as he points out, at the end of the day, "If you hire a human, it will take you time to build up trust, it will take time until this person has shown me, enough times, that they're delivering what I expect."
Previous and related coverage:
An executive guide to artificial intelligence, from machine learning and general AI to neural networks.
The lowdown on deep learning: from how it relates to the wider field of machine learning through to how to get started with it.
This guide explains what machine learning is, how it is related to artificial intelligence, how it works and why it matters.
An introduction to cloud computing right from the basics up to IaaS and PaaS, hybrid, public, and private cloud.