The Vibe Father
Dutch cooperative insurer Univé reports that 97 percent of its ChatGPT Enterprise licenses are activated and 85 percent of licensed users are active weekly. OpenAI’s July 31 case study also says employees created about 1,500 custom GPTs and average 40 prompts per active user each week.
These figures come from a customer story published by the vendor, so they should be read as reported results rather than an independent productivity study. The operating pattern is still useful.
Why adoption appears to have stuck
Univé treated AI as an organizational capability rather than a tool installed by IT. Leaders received dedicated sessions about changing work. Employees were given time and permission to redesign tasks. Governance was built into the rollout through enterprise authentication, inherited connector permissions, privacy assessments, security reviews, monitoring, and named human accountability.
| Layer | Reported practice |
|---|---|
| Leadership | Managers set direction and create space for experimentation |
| Access | AI permissions follow the underlying enterprise systems |
| Building | Employees create task-specific GPTs and share patterns |
| Decisions | AI prepares evidence while trained professionals decide |
| Measurement | Activation, weekly use, prompts, and workflow outcomes are tracked |
The claims workflow is the clearest example
A Workspace Agent can assemble a pet-insurance file, review invoices and policy terms, identify missing information, flag anomalies, and prepare a traceable recommendation. OpenAI says preparation that took hours can take minutes. The claims professional remains accountable for the final decision.
That division is more credible than “fully automated claims.” It moves retrieval and organization to the agent while keeping judgment and responsibility with a trained person.
What other teams can copy
- Measure sustained weekly use, not licenses purchased.
- Inherit permissions from source systems instead of creating a shadow access layer.
- Give employees protected experimentation time and a place to share working patterns.
- Require evidence and human ownership for consequential decisions.
- Review the growing catalog of custom agents for duplication, stale instructions, and data risk.
Bottom line
The Univé story suggests that broad adoption follows clear permission boundaries and employee agency. The next question is durability, whether thousands of custom GPTs can be governed, maintained, and measured as models and business rules change.
