Finance chief Dan Durn is turning Adobe’s finance organization into an early proving ground for agentic AI—using autonomous software agents to forecast results, scan contracts, and even answer hundreds of thousands of emails.
The push mirrors Adobe’s broader strategy around agentic AI. For customers, the company lets them choose models, combine them with their own data and Adobe’s, and point agents at specific business outcomes.
Inside finance, Durn groups AI use into three buckets: forecasting, anomaly detection, and general productivity.
For forecasting, AI uncovers patterns and signals in data that would be difficult for humans to detect quickly, he explains. Anomaly-detection agents flag performance that’s unexpectedly strong or weak—“things that can get lost in the sea of data”—so finance can intervene faster, he says.
However, Durn says the best examples now sit in productivity, citing three use cases:
1. Extracting information from PDFs
One of the most developed use cases involves “containers” of information—collections of PDFs such as investor transcripts, quarterly reports, and analyst research. Finance teams use Adobe’s PDF Spaces to load documents into a shared digital workspace and use an agentic AI assistant to surface themes, insights, and messaging cues in minutes rather than hours.
2. Cutting contract review time in half
Adobe is also using agentic AI to overhaul contract reviews across finance and procurement functions including revenue assurance, contract operations, product fulfillment, and vendor management. Instead of finance professionals combing through every clause, an AI assistant scans thousands of contracts, highlights provisions relevant to each function, and flags non-standard terms.
The system has cut review time roughly in half, speeding individual reviews and allowing teams to query the entire contract repository—for example, identifying which contracts include auto-cancellation features or foreign-exchange adjustment windows, Durn says. Adobe built its first prototype by April 2024 and began onboarding teams in January 2025.
3. Automating “common” inboxes
“In 2025 alone, the system auto-responded to about 300,000 emails across 19 inboxes, saving more than 5,000 hours of manual work and freeing teams to focus on more complex issues,” he says. The tool took about six months to build; beta teams began using it around August 2024, with full rollout in January 2025.
The payoff, he stresses, isn’t headcount cuts but the ability to scale more efficiently as Adobe grows.
Durn traces these finance use cases to Adobe’s long AI journey and a bottom-up idea pipeline. The company has invested in machine learning and AI for more than a decade, initially to understand customer usage patterns and embed intelligence into products—work that laid the groundwork for generative and agentic AI.
Many of the best applications come from “reaching down into the organization” and asking employees where AI could remove friction or make their jobs easier, he says. There are more ideas than capacity, so the team prioritizes those with the greatest impact.
When deciding whether to green-light AI investments, Durn focuses on organizational velocity—the ability of back-office functions to keep pace with faster product innovation. If finance doesn’t adopt AI, he argues, it risks becoming a “rate limiter of growth.”
The actual spend is modest, he adds; much of the work involves change management and process redesign layered onto Adobe’s technology.
For his own workflow, Durn relies on AI primarily for insight generation. Ahead of earnings, his team loads pre-earnings research reports, Adobe filings, and peer transcripts into an AI-powered workspace to surface themes and likely investor questions.
Scripts and Q&A preparation are then run through models with guardrails to test whether messaging addresses those themes and to ask, “If I were an investor, what are my key takeaways?”
He sees it as a useful check on clarity and consistency—using AI to validate instincts and sharpen how Adobe communicates with the market.
