A June 24 report by 404 Media cites alleged internal meeting audio from Accenture indicating that the company’s internal data links a large portion of its AI token consumption to non-engineering staff workflows. In the recording, Justice Kwak, Accenture’s agentic AI strategy lead, says: “We’re seeing from some of the data internally at least that it’s actually not our engineers that are driving the token consumption. It’s a lot of the non-engineers that are doing some of those behaviors [...] you were talking about.”
What was said in the exchange
Stuart Henderson, Accenture’s client group lead, interrupts and jokes that he hopes Kwak didn’t just convert a PDF into images and then into markdown files. Henderson comments, “I’m learning that’s one of the big token chewers,” and asks, “Turning PDFs into markdown: is that right?” Kwak responds that Accenture’s own data shows that pattern.
Why this matters
The exchange highlights how routine document-preparation workflows can materially influence costs in generative AI deployments. Converting PDFs into images and then into markdown — a workflow mentioned in the recording — can multiply the amount of text or tokens submitted to language models, which in turn raises consumption and expense.
Reporting and limitations
The account comes via a 404 Media piece based on the leaked audio; the published excerpt does not include specific numerical data or detailed breakdowns of token counts. The recording focuses on the observed relationship between user behavior and token consumption rather than providing precise metrics.
Takeaway
The discussion underscores that AI-related spending is affected not only by model choice and engineering practices but also by how non-technical staff prepare and submit documents. Such internal findings may prompt organizations to review document-handling and conversion practices to reduce unnecessary token use and associated costs.
