Many survey results show that AI adoption within companies often remains at the pilot level. A representative example is an MIT finding that 95 percent of pilot AI projects fail to move into real-world deployment and remain pilots.

Against that backdrop, Silicon Valley venture investment firm Andreessen Horowitz (a16z) published a rebuttal saying the MIT statistics are hard to believe, based on internal data and conversations with corporate executives.

The a16z research focuses on quantifying where companies use AI, how they use it, and where ROI is clearly generated. Given that it comes from a company with a position that benefits if AI takes off, it may be useful as a reference. This analysis focuses on enterprise AI and areas where companies create value with AI. Consumer-focused companies were excluded from the analysis.

According to the a16z research, 29 percent of Fortune 500 companies and 19 percent of Global 2000 companies are running AI in production as paying customers of leading AI startups.

a16z stressed that these figures show a scene different from what was seen in previous technology paradigms. Traditionally, Fortune 500 companies have been far from early adopters. Startups often sold products and technology to other startups early on to gain momentum. It was generally common for it to take years for a startup to win deals in the enterprise market.

But the a16z research shows AI is breaking from that formula. OpenAI released ChatGPT in 2022 and turned AI’s potential into an issue in both consumer and enterprise markets. Large enterprise companies also began betting on AI faster and more aggressively than ever, and those results are reflected in this research, a16z said.

From the perspective of paid use, the area where enterprise AI adoption is happening fastest is, as expected, coding, which holds an overwhelming position. Support and search work were also cited as AI use cases spreading in the enterprise market.

According to a16z, coding is the most ideal AI use case. It is based on strict syntax and predictable outcomes, and it can be verified. Anyone can run code and check whether it works.

Support sits on the opposite side of code development. Many organisations perceive support work as boring and complex, and often outsource it to overseas vendors or business process outsourcing (BPO) providers.

AI excels at supporting such work for several reasons. In customer support in particular, most interactions are time-constrained and have a clear purpose, such as handling refunds, so the problems agents must solve are clearly defined. Support teams also face heavy workloads and high turnover, creating a need to train new agents quickly and in a standardised way, which is a favourable condition for AI adoption. Because there are paths to connect to human agents, 100 percent accuracy is not necessarily required for it to be useful.

AI search is also cited as an area spreading in enterprises as various large startups have already emerged.

One difficulty many companies face internally is enabling employees to easily find and extract relevant information scattered across different systems, and Glean is standing out as a leading startup in that environment, a16z said.

From an industry perspective, technology, legal and healthcare were cited as representative sectors that use AI heavily. Looking at ChatGPT alone, 27 percent of business users came from the technology industry, and many early customers of Cursor and Glean were also technology companies.

It is not particularly surprising that technology companies that build technology directly use AI heavily, but legal and healthcare are somewhat different. In the legal sector, projects are long and the industry has been widely known as reluctant to adopt technology. The fact that AI provides clear value is cited as a factor driving the change.

Using legal work as an example, existing enterprise software provided limited value to lawyers. Workflow tools did not improve the efficiency of the unstructured and subtle tasks lawyers mainly perform, but with the emergence of AI, the value proposition that technology offers to lawyers has become much clearer, according to a16z.

In healthcare, unlike existing software, AI is spreading rapidly. Companies such as Abridge, Ambience Healthcare, OpenEvidence and Tennr have seen explosive revenue growth based on specific use cases such as medical scribing, medical information search, or automating complex regulations applied to healthcare delivery and billing.

Technology, legal and healthcare have already emerged as promising areas in the AI landscape, but there is still ample room to go deeper. In legal, for example, there are various types of lawyers, such as in-house counsel, law firms, patent attorneys and plaintiff-side attorneys, and they each have different workflows and requirements. Healthcare is similar, with different types of doctors and medical facilities mixed together, a16z said.

a16z also stressed that it is important to pay attention to where AI model developers are focusing their latest research capabilities. As agent performance improves rapidly, large-scale investment is made in computer use, and research progresses on reliable interfaces for various media beyond text, such as spreadsheets and presentations, it predicted that a large number of new types of startups will emerge before long.