For those who remain skeptical of applying AI to venture capital, consider what it has allowed Connetic Ventures to accomplish. The early-stage firm has just three full-time investors, but its internally developed AI system has enabled Connetic to analyze more than 30,000 start-ups and give it the confidence to invest in 220 of them to date.
The system, dubbed Wendal, “screens out 90 percent of companies [that apply, freeing us to] meet with 10 percent and invest in maybe 10 percent of those,” partner Chris Hjelm told Venture Capital Journal. “We’ve done some analysis, and Wendal can do the work of roughly 250 analysts, so we can keep a team of our size as we scale and get more assets under management and just tweak the algorithm.”
Connetic, based in in Covington, Kentucky, is among a handful of venture firms applying AI to deal sourcing, operations and due diligence with the aim of increasing productivity, freeing employees from repetitive tasks and unlocking insights from proprietary data sitting on hard drives. Other firms leading the AI charge include Alumni Ventures, AngelList, EQT Ventures, Headline, H/L Ventures and M13. (Read our September cover story about how AI is transforming the venture industry.)
How it started
When Kyle Schlotman and Brad Zapp founded Connetic in 2015, their aim was to fund promising start-ups that had been overlooked because they were based in the Midwest instead of more popular VC hubs like Silicon Valley. The duo raised about $5 million for their debut fund in 2015, and then closed on $23 million for their sophomore fund in 2018.
Hjelm joined the firm with Fund II and quickly concluded that he and his two partners could be a lot more effective if they leveraged the kind of quantitative analysis he had used at a previous job doing high-frequency trading for a wealth management firm.
Connetic initially brought in Joe Wendt, a former manager of business intelligence and application development for the Cincinnati Reds, to build the base system of Wendal over the course of roughly eight months. The firm now employs a small team of data scientists who continue to improve the algorithm as the first companies it tracked succeed and fail.
Wendal lives on Microsoft’s cloud computing platform Azure, which hosts all of its bot and machine learning algorithms, according to Hjelm. When founders apply, they are asked to fill out a short list of questionnaires and complete a proprietary behavioral psychology quiz called TeamPrint. The quiz asks founders to respond either, “yes, this describes me,” or “no, this does not describe me,” to 175 adjectives. Based on a combination of scoring methods developed in-house, each founder is then assigned an archetype, examples of which include operator, closer and crusader.
After acquiring the consulting practice and proprietary tools of an industrial psychologist who built an early version of TeamPrint, Connetic decided to focus on people and behavioral psychology to identify good investment bets, Hjelm explained.
“Wendal paints this really robust behavioral profile of the start-up and the founders,” he said. “We measure all kinds of crazy stuff like leadership, extroversion, drive, all the way down to toxicity. Covid tripled our dealflow and then as more people went through it and we invested in founders, people started to say, ‘Oh, this is a real thing; these people aren’t crazy.’”
While TeamPrint can make up as much as 45 percent of Wendal’s final recommendation, founders’ applications are also evaluated on their responses to short expository questions about their companies, as well as the pitch deck and financial information they submit.
Wendal compares company information and founder results to more than 365,000 proprietary data points that Connetic has gathered from more than 33,000 start-up submissions since 2019. (See a dashboard generated by Wendal on Connetic’s website.) Previous company outcomes are used to determine the fit for Connetic’s portfolio and the likelihood of success. Based on Wendel’s recommendation, a candidate either proceeds to the due diligence process or doesn’t.
As access to Wendal expands to other VC firms and is used for later-stage deals, Connetic plans to integrate customer testimonials and other market data to the algorithm, said Hjelm.
Wendal remains “in closed beta,” but once it becomes available to other firms, likely in three or four months, Connetic will allow information about prospective deals to be shared among users, he added.
In the next six months, Connetic plans to integrate a large language model (LLM) similar to ChatGPT, where users can type a question and receive a written response. The LLM would allow Wendal to streamline the analysis of pitch decks and other uploaded documents, as well as provide an interactive means for founders to get advice while uploading their data and before submitting their application.
For example, founders could ask the algorithm how much capital they should be raising based on the financial data they have uploaded, Hjelm explained. Once a centralized database has been built and the model has been trained, Wendal will be able to give an appropriate answer. Connetic is also working on an LLM feature that would enable Wendal to complete a company overview after reading all uploaded documents.
After some initial pushback, founders have become more receptive to Connetic’s AI screening process. “They know from the time they meet us they’ll get a decision quickly, but also the odds of getting an investment are so much higher because we know what we’re looking for to a certain degree,” Hjelm told VCJ.
The automated screening process also promises to remove biases that distort VCs’ response to founders. Hjelm cited “a transgender founder who cried and said, ‘This was the first time I felt like I was given a fair shot.’”
It’s also not lost on Connetic that machine learning tools often reflect biases of whoever programmed them. The firm has gone to great lengths to ensure that Wendal contains as little bias as possible by validating its results for fairness across age, race and gender, Hjelm said.
One way of doing this is by comparing the demographic breakdown of applications to that of founders that Wendal recommends to confirm they align, he explained. “So far, they’ve been within half a percent with any demographic we look at.”
While demographic data is an optional input, Hjelm says that roughly 70 percent of applications include that data, which he says “is good enough” to measure Wendal’s biases.
Connetic briefly spun Wendal out as a standalone start-up and raised a seed round before rolling it back into the holding company. That was largely based on Connetic’s decision to raise a 1940 Act fund, a pooled investment vehicle required to be run by a registered investment adviser under the Investment Company Act of 1940. The firm wants Wendal to be a part of a fund it plans to launch this winter.
“We’re going through the RIA and broker networks to market our fund for a variety of reasons, not least of which is there’s a real learning curve around venture capital, particularly in respect to the illiquidity of assets,” Connetic’s head of RIA, David Ross, told VCJ.
With Wendal, Connetic will be able to “offer in-line fee structure with other mutual funds,” Hjelm added. “We can do what a team of 50 can do with a team of three, and that’s why we can offer this low-fee, public-facing fund, and that’s a big reason why we’re doing it. And because we’re in Kentucky and it’s hard to raise private funds here.”
While Connetic plans to utilize Wendal in its upcoming fund, Hjelm said that the firm has an even bigger goal. Ultimately, Connetic aims to create a network of 10 to 15 like-minded VCs that would co-invest, share dealflow and embrace a data-driven selection process for portfolio companies.
“The ideal state is if we can get this in the hands of the Angel Capital Association and have all angels use it,” Hjelm said. “Then we have every deal in America in our system and can lead all the pre-seed and seed rounds for all the ones that meet certain thresholds.”