The price of AI tokens has fallen 97% from GPT-4 to GPT-5.4, and GPT-5.6 delivers better coding agent performance with 54% fewer output tokens and 57% less time per task. But token cost alone does not show whether AI creates value. As teams move from chat to agentic workflows, leaders need to track useful work per dollar: tasks completed, time saved, and workflows ready to scale. OpenAI recommends five steps to manage AI investments with confidence.
1. Sharpen visibility into usage and spend
Enterprise admins need a plain view of who is using AI, which products or models they are using, how much capacity they consume, and what kind of work that usage supports. Without that visibility, a growing bill is hard to interpret - it could reflect waste, productive experimentation, or a workflow that is becoming business-critical. ChatGPT Work supports longer, multi-step tasks, so usage can vary widely by workflow. Admins need to see the work behind the usage, not just the credits consumed.
Updated usage analytics and spend controls in the Admin Console let admins see adoption, credit usage, and spend by user, product, and model. They can track trends over time, identify emerging patterns, and understand when usage reflects broad adoption, a power-user workflow, or a recurring business process that may deserve more investment. Insights at the workspace, team, and product level help guide investment and enablement decisions.
2. Evaluate model efficiency by outcome ROI
The lowest token price does not always produce the lowest total cost. A cheaper model may fail, retry, or create work that needs correction. A more capable model may cost more per token but reach an acceptable result faster, with fewer attempts and less review. Evaluate models on the work they need to perform. Use evals that reflect real tasks, including edge cases, and define "good enough" before testing. Then measure the full cost of reaching that standard: model and tool usage, attempts, completion rate, latency, and human review.
For priority workflows, track cost per accepted outcome. In customer support, that might be a resolved case. In engineering, it might be a tested change that passes review. Pair that cost with business value such as time saved, cycle time reduced, or capacity created. Model choice is only part of the equation. Clear instructions, focused tools, reusable context, and explicit stopping conditions can reduce loops and wasted spend. Use smaller or faster models when they meet the quality bar, and reserve frontier intelligence for complex, ambiguous, or high-stakes work.
3. Govern advanced workflows before they scale
Governance should be the operating layer that determines which AI work can scale. Define what context ChatGPT can use, which tools it can access, what actions it can take, who approves higher-risk steps, and how additional capacity is granted when teams find valuable workflows. This becomes more important as teams adopt plugins, connectors, Computer Use, and other frontier capabilities that can operate across enterprise systems.
ChatGPT Work gives admins centralized controls for access, approved context, connected tools, permitted actions, usage, and spend. Spend controls such as workspace defaults, group limits, individual overrides, and review requests with project context help leaders support high-value work without raising limits broadly. For priority deployments, OpenAI's AI Deployment Engineers can work directly with customers on evals, architecture, latency, reliability, and workflow design. Privacy and governance should be part of that work from the start: sensitive workflows need the right access controls, retention posture, compliance visibility, and approval paths before they scale. Where applicable, OpenAI's enterprise privacy controls, including Zero Data Retention options, can help deploy AI in high-trust environments.
4. Fund workflows that can compound
Manage AI investments as a portfolio: broad access for everyday productivity, function-specific workflows that improve repeatable work, and a smaller number of strategic bets built around proprietary company context. The strongest candidates are workflows that repeat at meaningful scale, have clear ownership, and can be measured for quality, risk, and business value. Funding should follow maturity. Exploration should test whether the model can handle the task; validation should test representative cases against a clear quality bar; production funding should support the integrations, controls, reliability, and change management required to scale.
Shared capabilities such as identity, trusted connectors, curated knowledge, evaluations, observability, model routing, and reusable agent patterns should be funded centrally. That makes each new workflow easier and safer to launch.
5. Match capacity to proven demand
Once a workflow proves its value, match the product, capacity, and support model to its demand. ChatGPT Work provides ready-made capabilities for chat, coding, agentic workflows, connectors, plugins, Computer Use, and administration. Companies can extend that foundation with proprietary data, permissions, evaluations, and workflow logic where those elements create differentiated value. For production workloads, the commercial structure should match usage patterns: Guaranteed Capacity for production systems that need access certainty, Scale Tier for predictable high-volume API workloads, and Batch API, Flex processing, or Prompt Caching for asynchronous work or repeated context.
For larger strategic deployments, OpenAI Frontier and Deployment Company can help enterprises build, deploy, and manage AI coworkers across enterprise systems. This approach lets leaders scale proven work with the right product, capacity, and support model instead of making each workflow rebuild its own infrastructure.
Why this matters for management
The shift from chat to agentic workflows means AI spend will increasingly reflect business process automation, not just user prompts. Managers who track only token consumption will miss the real picture of value creation. The five steps above give a framework for investing in AI that compounds: build visibility into usage, measure cost per accepted outcome, govern workflows before they scale, fund repeatable workflows as a portfolio, and match capacity to proven demand. Managers can build these capabilities through dedicated training programs like AI for Management.
Your membership also unlocks:
