About 92% of large companies are achieving returns on their investments in artificial intelligence, and the same percentage are increasing their AI investments. But what does it take for startups and early-stage companies to get to this point?
That’s a critical question, according to Sukwoong Choi, a postdoctoral scholar at MIT Sloan. “AI utilization is tied to startups’ products and services. It’s more directly relevant,” he said.
In a new paper, Choi and his co-authors find that firms need to be ready to make a significant investment in AI to see any gains, because limited AI adoption doesn’t contribute to revenue growth. Only when firms increase their intensity of AI adoption to at least 25% — meaning that they are using a quarter of the AI tools currently available to them — do growth rates pick up and investments in AI start to pay off.
The paper was co-authored by Yong Suk Lee, Taekyun Kim, and Wonjoon Kim.
Here are three things companies should know about investing in AI.
1. It takes time to see gains.
The researchers surveyed 160 startups and small businesses in South Korea about their use of AI technologies such as natural language processing, computer vision, and machine learning. Of the firms included, 53% were in technology-related fields (namely software, pharma, and mobile computing), and 54% had adopted AI to some degree.
The survey was administered to companies created before 2015, as these firms were founded before AI adoption generally took off in South Korea. (A footnote in the paper points to “an explosion of interest in AI” in the country after Go master Lee Sedol lost four of five matches to Google DeepMind’s AlphaGo program in March 2016.)
Among the firms surveyed, the correlation between AI adoption and revenue growth followed a J-curve: slow and steady at first, then substantial. The turning point was an intensity of AI adoption of 25%. For firms with AI intensity below 25%, annual revenue growth was essentially zero; for firms above the 25% threshold, growth approached 24%.
“There’s a disruptive power for AI. With lower utilization, it’s harder to make a profit,” Choi said. “When you’re in those early stages of AI adoption, you may need some time to obtain the payoff to using AI.”
2. Invest in complementary technology.
Several factors can influence a firm’s embrace of AI, the researchers found. For example, firms that are smaller and/or were founded by CEOs with prior entrepreneurial experience are more likely to adopt AI intensively. Larger firms or spinoffs from other companies are less likely to adopt AI at that level, though lab-based spinoffs are an exception.
One of the most influential factors, though, is adoption of complementary technology — namely, big data capabilities and cloud computing. The former contributes to better AI outcomes through more mature data collection and management, while the latter provides the computational power necessary to run complex analyses. Both help firms drive growth from their investments in AI.
This finding came as little surprise to Choi and his co-authors. For decades, investing in one type of technology has driven the adoption of other technologies. Examples abound: Better operating systems led to better software, faster modems made computer networks possible, and IT infrastructure supported the growth of online selling.
“Complementary technology makes it easy to adopt new technology such as AI,” Choi said. “To adopt and utilize AI effectively, and to get the payoff at earlier stages in your investment, you need the technology and the skills that go with it.”
3. Encourage smart investment.
The pivotal role of complementary technology points to one key takeaway from the paper, Choi said. To support AI adoption, it’s not enough to have access to the technology — you also need the infrastructure that supports it. “When you make that easily available, you can accelerate AI adoption,” Choi said.
Related Articles
The second consideration is how closely AI is tied to a company’s core product or service, he said, and how that impacts the company’s research and development strategy.
Internally focused R&D helps a company build “absorptive capacity” — in this case, AI know-how — that positions it to more intensively adopt and use AI technology. This is helpful for firms that need to protect their proprietary algorithms as intellectual property, or for firms working with sensitive data sets they’d rather not allow a third party to process.
On the other hand, if AI is a complement to the work that a firm is doing but isn’t the core focus of that work, firms can turn to external resources, Choi said. Large language models, such as OpenAI’s ChatGPT, are a good example of this: They’re readily available, widely used, and constantly being refined.
“It’s important to ask, ‘Is there a point solution for the AI work I’m trying to do?’” Choi said. “If your area of work is more systematic, then you don’t necessarily need an internally focused R&D strategy. You can license something that’s already available.”
