This article first appeared in Sifted’s Up Round newsletter, sign up here.
The time has come — as it does with every investment hype cycle — for VCs to start wondering whether the billions they’ve poured into AI over the past two years will ever pay off.
Companies which raised big earlier in the AI boom will soon need to raise funding again — and VCs are having a good think about what metrics they’ll need to see to merit further investment.
“The valuations are high, some of the companies have fantastic traditional sales and revenue metrics, but you don’t know necessarily how much of that is experimentation on the part of customers, how much of it is really real and going to retain and some of the companies don’t have any metrics,” Suranga Chandratillake, general partner at Balderton Capital, recently told me.
It’s easy to be dazzled by big annual recurring revenue figures, says Francesco Ricciuti, an associate at Runa Capital, who’s focused on deeptech — even though customers might not be sticking around. “We saw companies getting to $1m in revenues in less than a year, which is unseen for SaaS companies, right?” he tells me. “If anything, you need to be more diligent, because that’s a shiny number that could sort of get into your brain, and you would not see many worrying things like high churn, high customer acquisition costs.”
“People are just facing the truth that many of these applications were not needed or not ready to scale,” adds Ricciuti.
So, where are VCs placing their next AI bets?
Back to basics
Michael Treskow, partner at London-based VC firm Eight Roads, says his AI strategy is about, well, not AI. “The ultimate question for us is, ‘Do customers like the product?’” he tells me. “Hopefully most of us are a bit more nuanced in how [we] evaluate this than just to say, ‘Oh, okay, you have AI, check the box.’”
Ricciuti puts it another way: Red flag: if they brand themselves as an AI startup; Green flag: if they don’t.
Of course, AI needs infrastructure. Ricciuti thinks the boom in data centres opens up plenty of opportunities to make money from hardware — things like liquid cooling or providing energy for the data centres. Within software, he says Runa’s looking for “all the old SaaS metrics,” like customer acquisition cost, lifetime value of your customers, churn, the health of customer acquisition and margin.
He’s especially excited by AI in robotics and autonomous vehicle applications — like UK AV startup Wayve’s use of end-to-end AI (a method where the model learns the steps of a function from start to finish). “Maybe that’s what we needed to see these technologies finally come to the real world,” he says.
Treskow says he’s avoiding companies that are essentially building their business around OpenAI’s API.
Instead, he’s looking for companies in heavily regulated industries where you can’t just use ChatGPT — or “any vertical that has a proprietary set of data that is not obviously publicly available.”
He’s also interested in financial services and healthcare — highly regulated industries where “you need to understand where the data comes from”. “It needs to be dealt with in a compliant manner….That creates added layers of complexity that probably justify a separate product,” he says.
Treskow points out that we’re moving more towards AI agents that aren’t just answering questions for you anymore, but performing tasks based on information. “For that to be possible, you need integrations into systems,” and “healthcare is an interesting one for that, where you have your electronic health record systems; you need an integration with that, the data needs to be HIPAA compliant,” he says, referring to health records privacy laws in the US.
“That’s at least as a starting point for saying, ‘Okay, you can see a world where this is not just going to be consumed by a large language model,’” he adds.
As for what we’re not going to see so much anymore? “Many of these big rounds that we’ve seen are not replicable; there’s a bunch of people that wanted to get early into these companies because they believe these companies would do the Mistral-like path — raise every six months at increasing valuations. And if you enter that seed in Mistral now, you could sell secondaries after a couple of years and get a nice, fat profit. That’s a strategy; you don’t need to wait for an IPO, right? But I think it’s increasingly hard to find now. Many of these things have already happened,” says Ricciuti.
I’d love to hear from you, VCs: what’s your strategy for investing in AI right now? Send me a note.
This article first appeared in Sifted’s Up Round newsletter. Want more stories like this? Sign up here.
Read the orginal article: https://sifted.eu/articles/vc-ai-trends-investment/
