AI hallucinates new crop of financial measures

By Pranav Kiran

Artificial intelligence is sparking some creative finance. Companies from Accenture ACN to Oracle ORCL are using inventive and unconventional metrics to showcase how they're benefiting from the mad rush into large-language models. The efforts, like many similar ones that have come before, are ostensibly designed to help investors understand rapidly evolving business plans, but often wind up causing distortions.

Most of the AI attention nowadays is trained on the gigantic cost side of the ledger. To justify the expenditures and the exuberance, entrepreneurs and CEOs have started charting their revenue prospects. Various strategies are taking shape, some of which complicate the numbers more than others. Sam Altman’s OpenAI partly charges based on the amount of ChatGPT usage, which should in theory produce a relatively more straightforward sum each quarter, while Salesforce CRM bills clients on a subscription basis.

Many established companies are experimenting with new formulations to present AI-related income. Salesforce leaned on its conventional approach to specify on Wednesday that it had generated $440 million of annual recurring revenue from agentic AI in the latest quarter. IT services provider Accenture reports new bookings in the area. Photoshop maker Adobe ADBE talks about “AI-influenced ARR” and “AI-first ARR.” Others cite remaining performance obligations, a measure of how much in bookings could be realized in a reporting period, or classify some revenue as coming from "AI-native" customers.

These inventive, non-GAAP methods have an uncomfortably familiar feeling. Online discounter Groupon GRPN trotted out "adjusted consolidated segment operating income," ACSOI in the vernacular, in its initial public offering prospectus years ago and Tesla used to emphasize its "positive core operational cash flow." Before WeWork collapsed, the shared-office renter pioneered "community-adjusted EBITDA," an even more novel riff on an already artistic measure of profit.

The danger now is that companies are hallucinating ways to paint a rosy picture, even if there may be reasonable explanations for doing so. Standard revenue methodology didn't tell the whole story about cloud computing. Lumpy contracts, varied pricing tiers and upfront cash payments for subscriptions warranted supplemental ways of totting it all up. The industry and its backers eventually settled on ARR, which annualizes recurring revenue to smooth things out.

There's no agreed set of principles yet in AI, however, evidenced by the myriad approaches being used. The danger is that by leaning into the hype with a potentially inflated and hard-to-compare portrayal of the technology's prospects will only make it harder to discern the financial realities.

CONTEXT NEWS

Salesforce said on October 15 that it expects to record more than $60 billion of revenue in financial year 2030, which would be a more than 58% increase from 2025.

The software developer also noted that it had generated $440 million in annual recurring revenue from its generative AI and Agentforce businesses in the quarter ended July 31.