OpenAI show how the company uses its own technology to transform financial processes. Their in-house “DocuGPT” agent now processes over a thousand contracts per month — without the team having to grow proportionally. From hundreds to thousands of contracts per month. In less than six months. With just one additional employee. That's the reality that OpenAI's finance team faced. The solution? No more hiring people, but building an AI agent.
Every enterprise contract that OpenAI concludes contains crucial information: start dates, billing terms, renewal clauses. In the beginning, the process was clear: reading line by line, manually typing over to a spreadsheet, moving on to the next one.
But when the volume doubled (and doubled again), this manual process broke. Wei An Lee, AI Engineer at OpenAI, describes the problem: the team went from hundreds to more than a thousand contracts per month, while only one person was added. It was clear that this approach was not sustainable.
Instead of solving the problem with more people, the finance and engineering teams built a “contract data agent” together. The design principle was simple but thoughtful: take the repetition out of contract review, but keep experts firmly at the wheel.
The agent works in three steps:
1. Data collectionPDFs, scanned copies, even photos with handwritten annotations — what used to be dozens of inconsistent files now flows through one pipeline.
2. Inference with promptingUsing retrieval-augmented prompting (RAG), the system parses contracts into structured data. Important: It doesn't dump a thousand pages in context. It only retrieves relevant information, reasons about it, and shows how to reach conclusions.
3. Expert reviewFinance experts review the structured output, complete with annotations and references for different conditions. The agent highlights what's unusual; people are brought in to review.
What distinguishes this approach from simple data extraction is the reasoning component. The system shows why a particular term is considered non-standard, cites the reference material, and allows the reviewer to confirm the ASC 606 classification.
The result? Data that is ready overnight for validation the next morning. What used to take hours now arrives annotated and ready for review.
The benefits are measurable in concrete terms:
Each cycle of human feedback sharpens the agent, making each subsequent review faster and more accurate.
Wei An Lee summarizes: “This is the only way we can scale like OpenAI scales. Without it, you would have to let your team grow linearly with contract volume. This allows us to stay lean while we go through hypergrowth.”
This architecture now also supports procurement, compliance, and even the monthly end. The same principle applies: automate routine work, hold people accountable for judgment.
Engineers describe it as “manual work already done” — not decisions that have been replaced. Finance teams are still writing the numbers story; the agent makes sure they don't spend their day tedious manual entry.
What started as a fix for contracts has grown into a new way of working in finance. Data parsing runs overnight. Professionals focus on analysis and strategy. Leaders confidently scale with growth, without allowing teams to grow in lockstep.
According to OpenAI, the contract data agent is a blueprint for how AI can responsibly transform regulated, high-stakes work.
This OpenAI example illustrates a pattern we see in more and more organizations: AI agents that don't instead of experts work, but pre experts — by removing the tedious, repetitive work so professionals can focus on where they really add value.
The lesson isn't that you have to build a complex agent tomorrow. The lesson is that the combination of human expertise and AI automation provides economies of scale that are impossible with pure human capacity.
For finance teams, legal departments, and other knowledge workers who work with large volumes of documents: this is where the future of work is moving towards. Not replacement, but reinforcement. Not fewer people, but people who are engaged in work that matters.
Do you want to know how your organization can use AI agents to scale without growing linearly? Take contact contact us for an informal conversation about the possibilities.
.svg)
