Key Takeaways
- OpenAI's B2B Signals report says frontier firms use 3.5x as much AI intelligence per worker as typical firms, up from 2x in April 2025.
- Message volume explains only 36% of that gap, which means usage depth matters more than how many prompts employees send.
- Finance and Insurance ranks first for ChatGPT adoption in OpenAI's industry table, but finance teams still need better measures for workflow depth.
- CFOs should track governed use cases, data exposure, review controls and delegated work instead of treating license deployment as success.
What did OpenAI measure in B2B Signals?
OpenAI's B2B Signals, published May 6, 2026, uses aggregated enterprise usage data to answer a simple question: do frontier firms just use more AI, or do they use it differently?
The answer: they use it differently. Frontier firms (the 95th percentile of AI use) deploy 3.5x as much intelligence per worker as typical firms. That gap has widened since April 2025 (it was 2x then). But here's the key insight: message volume explains only 36% of that gap. The remaining 64% comes from richer context, more complex tasks and more substantive outputs.
OpenAI uses tokens generated as a proxy for intelligence demanded. The implication is blunt: broad deployment ≠ deep use.
For finance leaders, the useful part is the decomposition. OpenAI says message volume explains only 36% of the frontier advantage. The remaining gap is tied to richer context, more complex tasks and more substantive outputs. That moves the finance question from "Who has ChatGPT Enterprise?" to "Where is AI doing work that changes the process?"
Why is access no longer enough?
Access is a starting line, not a maturity measure. A firm can deploy thousands of accounts and still keep AI outside core finance work.
OpenAI frames the first phase of enterprise AI adoption around seats, access and experimentation. That was the easy scoreboard. It told leaders whether employees had tools, not whether those tools changed month-end close, forecasting, procurement review or compliance work.
The B2B Signals data shows why that distinction matters. Finance and Insurance ranks first for ChatGPT adoption in OpenAI's industry table, but it ranks lower on ChatGPT intensity, Codex adoption and API intensity. Broad deployment does not automatically become deep use.
That is why AI adoption metrics for finance teams need to separate reach from depth. Reach asks how many people can use AI. Depth asks whether the work now includes company context, governed data access, documented review and reusable workflows. The first metric is procurement. The second is operating change.
What does depth of use mean for finance teams?
Depth of use means AI handles more substantive finance tasks with context, controls and review instead of answering one-off questions.
OpenAI's task table gives finance leaders a starting point. In Finance messages, OpenAI reports 37.1% writing and communication, 22.4% advice, 18.4% information and 10.8% analysis and calculations. Those numbers do not prove business value. They do show where employees are already asking AI to support finance work.
The deeper signal is in the API examples. OpenAI lists finance and insurance use cases such as data analysis, summarization, extraction, automated expense management, research-summary generation, workflow optimization, policy search and customer support. Those are closer to workflows than chat prompts.
This is where the accounting and finance trust problem returns. As Nexairi covered in our expert call on closing the AI gap in accounting, professional confidence depends on review paths, liability clarity and source visibility. Depth without control is just a faster way to create exposure.
| Metric | What It Shows | What CFOs Should Ask |
|---|---|---|
| Seat deployment | How many employees can access AI tools | Which roles use the accounts weekly for finance work? |
| Workflow depth | Whether AI supports analysis, close, reporting or procurement tasks | Which workflows changed cycle time, review time or output quality? |
| Context quality | Whether users provide policies, files, systems data and business rules | What data can AI use and where is that access logged? |
| Delegated work | Whether teams hand defined tasks to agents for execution | What must a human approve before work leaves the finance team? |
| Governance coverage | Whether controls match the risk of each use case | Which AI outputs need source links, signoff or audit trails? |
How should CFOs measure AI adoption now?
CFOs should measure adoption by workflow evidence: repeat usage, approved data access, review controls, output quality and documented business impact.
A practical scorecard starts with four questions. First, what finance workflows are using AI every week? Second, which systems or documents feed those workflows? Third, who reviews the output? Fourth, what business metric changed after adoption?
That sounds basic, but many AI programs still report adoption as licenses deployed or training sessions completed. Those measures miss the depth gap. A finance analyst using ChatGPT Enterprise to rewrite emails is not the same as a treasury team using an approved workflow to summarize exposure, flag exceptions and produce a review packet.
The measurement model should also account for enablement. OpenAI says frontier firms use AI more heavily for education and learning tasks. For CFO governance, that suggests AI training for finance teams is not a side benefit. It is part of the infrastructure that lets staff move from simple prompting to supervised delegation.
Do finance teams need AI agents yet?
Finance teams do not need agents everywhere, but they should start testing delegated work where review paths are clear.
OpenAI says the frontier advantage is largest in advanced agentic tools, with frontier firms sending 16x as many Codex messages per worker as typical firms. Codex is a coding agent, but the broader signal is not limited to engineering. OpenAI's summary says typical firms use AI to answer questions while frontier firms use it to help execute complex work.
The finance market is already moving in that direction. In a May 2026 collaboration, OpenAI and PwC said they are building AI agents around planning, forecasting, reporting, procurement, payments, treasury, tax and the accounting close.
PwC described the effort as focused on "practical, high-value workflows" under human supervision. That phrase is the standard finance leaders should hold vendors to. If a vendor cannot explain the task boundary, human approval point, data access rule and audit trail, it is not ready for core finance work.
The Firm-Level Adoption Test
The depth gap is a useful warning because it pushes leaders past vanity metrics. A board slide that says 1,000 employees have AI access does not answer whether AI reduced close friction, improved forecast work or made policy review more consistent.
The next quarter should be narrower and more measurable. Pick one workflow that has repeat volume, clear source documents and a defined reviewer. Examples include procurement policy checks, variance commentary, contract intake summaries or dashboard generation. Then measure time saved, exception quality, reviewer changes and any rework caused by AI output.
This is also where credential claims get harder to fake. In our analysis of AI and fractional CFO credential inflation, practitioners warned that AI can make shallow expertise look polished. Depth metrics ask for proof that the workflow changed under supervision.
What should firms change over the next quarter?
Firms should shift from AI availability reporting to a governed workflow scorecard that tracks depth, risk, review and measurable operating change.
Start by replacing one adoption dashboard. Keep seat count, but move it below workflow depth. Add fields for use case owner, data touched, review standard, risk level, business metric and whether the task is assisted or delegated.
Second, map where AI touches client data, employee data and financial records. Depth increases value because AI gets more context. It also raises the control burden because richer context can include sensitive information.
Third, train staff on delegated work. The beginner question "do I need AI agents in finance" has a practical answer: not for every task, and not before governance. But teams do need to learn how to hand off bounded work, inspect outputs and improve the process without relaxing accountability.
This is the real takeaway. Frontier firms are not just using AI more—they're using it to reshape workflows. Finance leaders who want to keep pace need to shift from tracking deployment to tracking whether AI actually changed how work gets done. Stop measuring seats. Start measuring business outcomes. The depth gap is real. The question is whether your firm is on the right side of it.
