The publication reportedly obtained audio of a meeting at consulting giant Accenture, where employees were told they were seeing a “token increase in spending” driven by pointless tasks like converting PDFs to presentation slides. Those uses are actually starting to burn more tokens than the work being done by the technical staff. “At least we are seeing from some data internally that it is not actually our engineers who are driving the token consumption,” an Accenture employee said in the audio, according to 404 Media. “That’s a lot of non-engineers who are dealing with something like this.”
While it may seem a little absurd to think that employees who focus on BS-ing through presentations would be using more tokens than their colleagues in engineering, PDFs are an exceptionally inefficient way to feed information into AI systems. Depending on the tool and document, a model may have to extract and interpret not only the text but also the layout, images, charts, and other visual elements of each page. File format may be the boss’s best friend, but that may change once the token bill is collected.
It’s not at all difficult to figure out how Accenture found itself in this position. according to a Report published by Financial Times Earlier this year, a Fortune 500 consulting company went out of its way to encourage people to use AI tools, even taking steps to track and tie down logins made by employees. Incentivize their chatbot usage. In a memo to employees, the company reportedly said that advancing in the company would require “regular adoption” of AI, hinting that use of the technology would not be optional.
Of course, Accenture isn’t the only company to use some shockingly non-critical metrics to promote the use of AI. Big tech companies like Meta and Amazon started leaderboards to track which employees in the company are spending the most tokens and encourage them to add themselves to the list. This, of course, just burns tokens to motivate people to do menial tasks with AI that they could otherwise do themselves.
This approach is silly but OK in the “free money” era of AI, where costs were not tied to token usage. But as major AI labs move toward IPOs, they have shifted to a usage-based model for pricing. This has resulted in some ridiculously high bills. This is less sustainable, and now companies are starting to ask their employees to cut back on the use of AI rather than increase it.
Strange that a company that tells other businesses how to work more efficiently and effectively failed to see the very obvious and predictable token crash. There’s probably nothing to read there.
<a href