Dan Diacchio of EY attended an on-site AI Agenda event hosted by The Information, and one phenomenon is worth noting.

Data from consulting firm EY shows that since enterprises began using AI, they still have not achieved significant revenue growth or cost reduction.

"I have not yet encountered a client who says they plan to cut AI spending, but quite a few clients will lament, 'I simply cannot see where the return on investment (ROI) is right now,'" said EY executive Dan Diacchio during a conversation with Jessica Lessin at The Information's event.

This view deserves serious attention. Complaints of this kind from enterprises about AI have persisted for one to two years.

This also explains why a large number of software and AI companies have chosen to cut prices in order to persuade enterprise customers to use AI tools.

Over the past year, service providers such as Anthropic have widely popularized usage-based billing models, which has also made enterprises especially sensitive to the question of AI returns.

This billing model pushes up AI bills. For example, Uber's CTO pointed out that Claude-Code can cause AI budgets to spiral out of control.

On one hand, enterprises have begun to use AI tools more prudently; on the other hand, they are turning more toward open-source models to reduce the fees they pay to Anthropic and OpenAI.

Even so, Diacchio said that current enterprise AI spending is still "driven by enthusiasm, rather than evidence of actual returns."

EY conducts a large number of CEO surveys. He said that only one in ten enterprises "can truly prove on the income statement that AI has generated a return on investment."

He added: "There is no doubt that everyone will say AI significantly improves organizational productivity." But from the perspective of a chief financial officer, "productivity must translate into one of two outcomes: either increasing revenue or reducing organizational costs."

Some companies, such as Block and Cloudflare, have already used AI as a reason for large-scale layoffs.

Block executive Owen Jennings said that thanks to AI investment, the company's gross profit has "significantly accelerated"; enterprises can deliver products to customers without maintaining their previous headcount scale.

Drawing on the approach of Snowflake's chief information officer, Jennings said Block treats layoffs as a kind of "forced driving force" to push employees to improve work efficiency with the help of AI tools.

Jennings said on stage to our colleague Amir Efrati: "Without such a powerful forcing mechanism, it is very difficult to drive change." (This kind of rhetoric from American companies may also be part of the reason why American voters and ordinary workers generally dislike AI.)

Anthropic, Blackstone, and other Wall Street institutions established a joint venture called Ode, aimed at helping enterprises implement AI. Ode's CEO Chris Taylor agrees that "returns on AI investment need time to appear."

He told our reporter Cory Weinberg at the summit that Ode's customers are only just beginning to see returns from AI investment; the income statement itself is a lagging indicator.

"An AI project must complete launch and implementation and produce business impact, and after that a period of time must pass before the effects show up in income statement metrics. The entire cycle is very long."

Some software companies choose to tightly lock in customer data, but Atlassian does the opposite.

Atlassian focuses on software for developers and team collaboration. In February this year, it launched a feature allowing customers to use external AI chatbots and code agents such as OpenAI and Anthropic to access work records, transcripts, and other data stored within the Atlassian platform.

Atlassian's chief product and AI officer, Tamar Yehoshua, said there was internal disagreement over whether to launch the feature.

But the decision proved correct, and this feature drove more active customer use of Atlassian products.

The industry widely fears that large model labs will seize the market of traditional software vendors, but Yehoshua disagrees. This judgment is also the confidence behind the company's willingness to open up this feature.

She said at the event: "Whenever a technological change occurs, people always think that the underlying foundation model vendors will inevitably expand upward and eat up the entire market. But this has never happened in history, and I firmly believe it will not happen in the future either."

Yehoshua, who worked at Google for more than eight years, gave an example: in the past, everyone worried that Google would use its search monopoly to take over the entire consumer application market, but that did not come true.