From IT project to cultural shift
When Univé decided to partner with OpenAI, it wasn’t just another technology deployment. As one of the largest cooperative insurers in the Netherlands—serving millions of members across insurance, mortgages and financial services—the company viewed AI adoption as an organisational transformation. The goal was not to push a new tool onto desktops, but to turn employees into “builders” who could safely redesign their own work.
The starting point was leadership. Instead of product demos, Univé ran dedicated AI sessions for its entire management community. Managers moved from approving isolated AI initiatives to creating the conditions for responsible experimentation across every team. Yous van Halder, Director Data & AI at Univé, put it bluntly: “Most organisations try to scale AI by building more solutions. We chose to scale AI by creating more builders.”
Governance built in from day one—not bolted on later
Many European enterprises fear that AI governance will throttle adoption. Univé proved the opposite. Enterprise authentication, connector permission inheritance, privacy assessments, security reviews and continuous monitoring were embedded into the ChatGPT Enterprise rollout from the very beginning. Permissions always follow the underlying enterprise systems, so the AI never sees more than an employee is already authorised to access.
This approach resonates with the EU AI Act, whose full enforcement starts on 2 August 2026. The Act mandates that high-risk AI systems—including those used in insurance underwriting and claims decisions—must have human oversight, robust documentation and fundamental rights impact assessments. By engineering accountability into its GenAI platform from day one, Univé is already living up to the law’s spirit, well before many competitors.
Governance became an accelerator, not a barrier. Employees had clear guardrails for security, privacy and human accountability, which gave them the confidence to experiment. That trust is now visible in the numbers: the insurer’s workforce generates around 40 prompts per active user each week, and staff have created roughly 1,500 custom GPTs tailored to internal workflows.
Concrete results: from hours to minutes in claims and underwriting
The most striking example sits in pet insurance claims. A ChatGPT Workspace Agent now assembles the claim file, reviews veterinary invoices, checks policy conditions, identifies missing information and highlights anomalies before the claims handler even opens the dossier. What used to take hours can be ready for decision in minutes. The trained claims professional remains fully accountable for the final call—AI prepares the work, humans make the decision.
Underwriting follows a similar pattern. When an underwriter logs in, a Workspace Agent has already reviewed the incoming queue, combined information from approved enterprise sources, flagged missing documents and prioritised high-risk cases. Instead of hunting for data, underwriters focus on applying their expertise to better, faster decisions. AI adoption now spans virtually every knowledge-based function, from claims and underwriting to finance, HR, legal and IT.
Why this matters for Europe right now
The Univé case study lands at a critical moment. Tomorrow, Article 50 of the EU AI Act requires that users be informed when they interact with an AI system and that synthetic content be clearly labelled. For insurers, this means that any AI-generated customer communication or automated underwriting output must be traceable and transparent. Univé’s architecture—where every AI-prepared recommendation leaves a clear audit trail and a human always signs off—is a blueprint for compliance.
Moreover, the Act’s rules for general-purpose AI models (GPAI) now apply directly, enforced by the European Commission’s AI Office. OpenAI’s ChatGPT Enterprise, which runs on models like the recently released GPT-5.6 family, falls squarely under these provisions. Univé’s partnership with OpenAI, complete with enterprise-grade authentication and continuous monitoring, demonstrates how European organisations can use frontier models without sacrificing regulatory alignment.
From a practical standpoint, ChatGPT Enterprise is available across the EU with data residency in European data centres. Companies pay on a per-seat basis—typically in the range of $60 per user per month (roughly €55), though volume agreements often bring the cost lower. For a cooperative serving millions of Dutch members, that investment quickly paid for itself through reduced manual preparation time and faster claims resolution.
A wider workforce readiness push
Univé is not an isolated experiment. OpenAI has recently accelerated its workforce readiness initiatives, expanding ChatGPT Edu deployments at institutions like the University of Phoenix to help working adults upskill without leaving their jobs. The Univé model—where employees spend hundreds of collective hours each week rethinking their work, not just prompting—fits directly into that vision. As Van Halder says: “Our competitive advantage is not that we use AI. It is that thousands of employees are learning how to reinvent their own work every single week.”
The numbers back him up. Over 1 million businesses worldwide now use OpenAI tools for organisational transformation, and the company’s enterprise and workforce platforms handle more than 800 million weekly active users producing over 2.5 billion daily messages. What sets Univé apart is not scale, but depth: 97% licence activation and 85% weekly usage are figures most corporate IT rollouts can only envy.
What European companies can learn
Univé’s leadership distilled their experience into a few practical lessons that ring especially loud on the eve of the AI Act’s full enforcement:
- Treat AI as an organisational capability, not just a technology project.
- Embed governance early so it becomes an accelerator for innovation, not a gatekeeper after the fact.
- Invest in leaders as much as in the technology—they create the conditions for change.
- Give employees not just access, but permission, time and structure to redesign their own work.
As the European regulatory landscape shifts from voluntary codes of practice to hard statutory enforcement, the companies that will thrive are those that, like Univé, already built the muscle of responsible experimentation. The conversation inside Univé has already moved from “Should we use AI?” to “What should we build next?”—a question that every boardroom in Europe now needs to answer.
Does the EU AI Act require every internal enterprise AI tool to be registered?
Not all internal AI tools must be registered, but high-risk use cases—such as AI used in insurance underwriting, claims assessment or creditworthiness decisions—fall under strict requirements for documentation, human oversight and fundamental rights impact assessments. Univé’s approach of keeping a human in the loop and maintaining full audit trails already mirrors the Act’s expectations.
How does Univé ensure data privacy under GDPR while using OpenAI’s cloud models?
Univé deployed ChatGPT Enterprise with enterprise authentication, permission inheritance from existing systems, and continuous monitoring. OpenAI’s enterprise offering provides data residency in European data centres and contractual commitments that align with GDPR, including no training on customer data. Combined, this allows the cooperative to use frontier AI without compromising member privacy.
Can other European insurers replicate Univé’s results without a huge budget?
Yes. The core of Univé’s success was not a massive financial investment but a deliberate cultural shift: leadership alignment, embedded governance and giving employees time to experiment. Even smaller insurers can adopt the same principles and start with a controlled pilot, perhaps using the more affordable GPT-5.6 Luna token pricing ($0.20/1M input tokens, $1.20/1M output tokens) if building custom applications via the API.
