OpenAI and PwC collaborate to reimagine the office of the CFO
Key Points
- AI agents for core finance workflows
- OpenAI finance as customer zero
- Focus on governance and cost visibility
Summary
OpenAI and PwC are partnering to build and deploy agentic AI for core finance workflows (planning, forecasting, reporting, procurement, payments, treasury, tax, accounting close). The collaboration focuses on production-ready implementations—using OpenAI Finance as "customer zero"—and combines OpenAI models and tools (Codex, Workspace Agents, Skills, Connectors) with PwC's finance transformation, controls, and implementation expertise.
Key Points
- Target workflows: procurement, contract review, spend tracking, forecasts, reporting, exceptions and month/quarter close support.
- Technical building blocks: Codex for tooling and scripts, Workspace Agents to operationalize workflows, Skills/Connectors for approved processes and system context.
- Operational emphasis: governance, human-in-the-loop controls, runtime controls, visibility into AI usage (token consumption, projected spend) and enterprise cost governance.
- Production learnings: OpenAI reports 5× contract processing with Codex and 200+ investor interactions handled via IR-GPT during a fundraise.
- PwC contribution: helps move prototypes to production with controls, change management, and implementation in real enterprise environments.
Practical guidance for engineers
- Start from customer-zero patterns: reuse OpenAI finance agent designs and telemetry models where applicable.
- Integrate Connectors and Skills to enforce approved processes and to pull verified enterprise context from ERP, contract systems, and reporting tools.
- Implement human-in-loop checkpoints for policy exceptions, approvals, and high-risk decisions; log decisions for auditability.
- Add runtime controls and monitoring: track token consumption, API usage, exception rates, and projected spend; expose dashboards for finance owners.
- Use Codex to automate repetitive code tasks (dashboards, spend-tracking, exception management) and iterate quickly with tests and feature flags.
- Define success metrics early: throughput (e.g., contracts/hour), error/exception rates, time-to-close, and cost per workflow.
Next steps
- Prototype one high-value workflow (procurement or contract review) with end-to-end Connectors, Skills, and human approvals.
- Instrument usage and cost telemetry, then iterate controls and guardrails before scaling across other finance areas.