Cold Call Microsoft’s Path to Adopting and Scaling AI Across its Sales Organization
7 snips
May 12, 2026 Shunyuan Zhang, Harvard Business School professor studying machine learning for marketing, and Iav Bojinov, HBS professor researching AI product integration, discuss Microsoft’s rollout of Copilot and autonomous sales agents. They explore why Copilot initially stalled, how experimentation and learning curves drove later adoption, the levers that changed sales work, and the limits and governance questions raised by agentic AI.
AI Snips
Chapters
Transcript
Episode notes
Adoption Fails Because People Not Process
- AI adoption failures often stem from organizational and behavioral layers rather than model capability.
- Microsoft saw adoption fall from 22% to 5% because people lacked incentives, role clarity, and time to experiment.
General Purpose AI Needs Experimentation Space
- General purpose generative AI breaks traditional deployment playbooks because it can do many things but doesn't tell users what to do.
- Deployments must create space and support for experimentation to surface valuable use cases.
Productivity J Curve Explains Early Dropoff
- Adoption follows a productivity J‑curve: initial slowdown, then recovery and gain if users persist.
- Outcome depends on product capability, learning processes (training/playbooks), and people’s willingness to invest effort.