Accenture flags PDF-to-markdown conversion as a major driver of AI token costs

Accenture’s agentic AI strategy lead, Justice Kwak, told colleagues that internal data indicates non-engineers are responsible for much of the company’s AI token consumption, contradicting assumptions that engineers were the primary drivers. In a leaked meeting recording, Kwak’s remarks were followed by a comment from Stuart Henderson, Accenture’s client group lead, who joked about PDF-to-markdown conversion as a major ‘token chewer.’ Kwak confirmed that Accenture’s own data supports Henderson’s observation.

The exchange highlights a gap between perceived and actual drivers of AI costs within enterprises, where routine document processing habits—particularly converting PDFs into more AI-friendly formats—can substantially increase token usage. Accenture may use these findings to advocate for broader adoption of alternative document formats to curb unnecessary AI spending.