Source: Identifying and Scaling AI Use Cases (OpenAI)

OpenAI's 2025 business guide: where to look for AI use cases, the six use-case primitives, and the impact/effort prioritisation matrix used by its customer teams.

Contents

Summary

A three-step method for enterprise use-case discovery: find where AI adds value, teach employees six recurring use-case types, then collect and prioritise ideas with an impact/effort quadrant. Drawn from OpenAI’s customer deployments; written for ChatGPT Enterprise adoption but applicable to API projects.

Key claims

  • Based on 300 successful implementations, 4,000+ adoption surveys and 2M+ business users; the primitives come from 600+ customer use cases.
  • Three principles: leadership should lead AI adoption; complex use cases often slow you down, so let employees find what works; hackathons, workshops and peer learning accelerate adoption.
  • Look for opportunities in three areas: repetitive low-value tasks (ask people for their “anti-to-do list”), skill bottlenecks (waiting on another team’s expertise), and navigating ambiguity (getting unstuck).
  • Six primitives: content creation, research, coding, data analysis, ideation/strategy, automation.
  • Impact/effort quadrant: High ROI focus (high value, low effort: start here), Self-service (low effort, individual use), Scope and prioritize (high value, high effort: often transformational, e.g. Indeed’s job-match explanations, which took months of testing and iteration and lifted applications started by 20%), Deprioritize (high effort, low value, e.g. building something an existing tool already does).
  • Re-score the matrix each quarter, since rising model capability can turn high-effort ideas into low-effort ones.
  • Estée Lauder’s repeatable process: cross-functional trio (business user, subject-matter expert, tech lead); design (two-page brief), prepare data, build and test, launch with user guide, pivot and scale from feedback.
  • Next step is department workflow mapping: break multi-step workflows into tasks and map each to a primitive.

Feeds

AI use-case discovery · POC to production

Suggest changes →