Generative AI startups like Stability AI, Mistral and Poolside may have normalised the $100m seed round over the past 12 months — but in trying to rival big tech and build large models, they have already missed the boat.
That’s the view of Alan Holland, founder and CEO of Keelvar, an Irish startup building an AI-powered system for automating company procurement that counts huge companies like Coca-Cola, Samsung and Adidas among its clients.
“You’re too late to that,” he tells Sifted. “By the time you get there, the existing vendors will have moved on again.”
Holland says that while the release of ChatGPT has created a lot of interest in the sector, the hype around generative tools has drawn investor attention away from solid businesses that are solving “less sexy” problems with AI.
“It’s so bloody basic”
“The most successful investments will be in businesses that are keeping their heads down and really solving genuine business challenges, and that have happy customers,” he says.
“It’s so bloody basic. It is, like, page one of business school.”
Holland says those VC funds that missed out on investing in OpenAI have lost sight of this basic lesson. Now they are trying to make up for that loss by backing would-be competitors to the Silicon Valley giant.
“[They want to] say: ‘We’ve got a horse in that race.’ I would say that is just a recipe for waste.”
Holland concedes that “procure-tech” — products for streamlining the complicated process of enterprises buying goods and dealing with global supply chains — might not be the “sexiest” industry. But it is a genuine pain point for a lot of companies around the world.
He could argue that the proof is in the pudding: Keelvar was able to run for six years without raising any VC funding before raising a Series A to accelerate growth in 2018.
“It’s actually a very healthy thing for a business to only spend the money you’re earning. I think there is a danger for businesses that you get hooked on the drug of VC money,” he says.
Keelvar, meanwhile, has doubled its revenue year-on-year since 2019 and is targeting a return to break even in 2024.
“It compels you to be very customer-centric because the one thing you don’t want is to have customers churn on you. Our culture became very customer-centric by necessity,” he says.
Selling to enterprise
Holland was previously a research fellow at University College Cork’s AI faculty, but he says a more valuable lesson was watching his parents start their own business.
“They set up their own small chemical manufacturing business when I was a teenager, so I was a witness to how to get a business off the ground,” he says.
“It showed me what it takes, and that you have to be the chief salesperson. You might have these advanced degrees in AI but you’ve just got to get out and start selling stuff.”
In terms of advice for other AI companies targeting the enterprise market, Holland says that the most important lesson he’s learned is not to over promise.
“Key decision-makers need to have real confidence in your team, and if you consistently deliver, you’ll have no issues,” he tells Sifted.
AI is like sports
While Holland is sceptical of companies that have raised large sums of money without having identified a clear problem to solve, he is bullish about businesses using “narrow” AI to solve complex problems.
For Keelvar, this involves using AI to automate tasks within the procurement process, like identifying new vendors, recommending the best suppliers and making bids.
“All of these various challenges require intelligent systems and different subfields of AI,” he says.
And while large language models (LLMs) are able to provide responses on an extremely broad range of topics, GenAI fever has clouded the fact that AI is actually a very diverse field, Holland says.
“Artificial intelligence is as broad as human intelligence. There’s so many different aspects to it and I think that’s just been lost in so much of the coverage: the richness of the field of AI,” he tells Sifted.
“Don’t get me wrong, AI is going to just become more and more powerful, but it’s other parts of AI that people haven’t heard of yet that will get bigger and stronger and more impactful.”
For Holland, fields such as search optimisation, machine vision and recommender systems as subfields will all be growing in the coming years.
“You hear ‘AI’ mentioned all the time. To people who’ve worked in AI for years, it’s a bit like hearing someone say: ‘I’m good at sports.’,” he says.
“But you might be good at swimming and no good at badminton. AI is just the same.”