Artificial intelligence does not fail to scale for technological reasons. It fails because organizations do not change fast enough.
Artificial intelligence does not fail to scale for technological reasons. It fails because organizations do not change fast enough.
Most large insurers and brokers no longer need to be convinced of the strategic importance of AI. In 2026, the real question is elsewhere: why does scaling up remain so difficult? And what does successful adoption look like when AI moves from pilots to fully integrated deployment into the day-to-day operation of a regulated business?
Nearly $1.5 trillion would have been invested in AI in 2025. Yet nearly two-thirds of companies have still failed to deploy it across the organization.
The main obstacle is not the capacity of the models, but the organization itself. Treating AI as a simple technological subject almost systematically leads to deployment difficulties.
In fact, scaling slows because companies accumulate a form of organizational “capacity debt”: misaligned roles, unclear decision-making responsibilities, insufficient governance or even incentive systems that no longer correspond to the way work is actually done.
Successfully deploying AI on a large scale requires that three dimensions evolve in parallel:
- a voluntary transformation of working methods,
- the integration of safeguards adapted to regulated environments,
- and the gradual reduction of this organizational debt to enable predictable and repeatable deployment.
Rethink work rather than adding a technological layer
Scaling up AI is difficult in large part because it profoundly transforms human organization.
As Roy Jakobs, CEO of Philips, summarized at the World Economic Forum in Davos 2026:
“When you bring new people into your organization, you need to rethink how the team will work together to accomplish the same tasks.”
This logic is particularly relevant in the insurance sector.
Introducing AI into claims management, underwriting assistance, contract management, fraud detection or distribution operations amounts to integrating a new player into workflows. AI suggests actions, generates content, identifies trends and accelerates decision-making.
This development is positive. But it involves rethinking in depth:
- the way in which activities flow from start to finish,
- the decisions that can be entrusted to AI and those that must remain human,
- the location of validations and controls,
- how teams collaborate with AI systems,
- or the way in which work is distributed, evaluated and valued.
In practice, AI cannot be deployed at scale without parallel evolution in HR, governance, security, risk management, finance and many other business functions.
AI transforms tasks more than jobs
In most cases, AI replaces tasks rather than entire jobs.
It supports tasks that are structured, repetitive, and heavily rule-based. Human roles are then recomposed around what remains essential: judgment, creativity, exception management, human interactions and even responsibility.
In insurance, this development is already visible: contract updates, certain claims management steps, data extraction, file preparation, classification or sorting can be automated or semi-automated.
It is in these areas that AI can generate rapid productivity gains. But this raises a structuring question: what to do with the capacity thus freed up?
It is difficult to deploy AI responsibly without explicitly deciding how that capacity will be reallocated.
The “human in the lead” model is gradually becoming established
The “human in the lead” model is gradually becoming established in regulated environments. It is based on principles which structure the way in which the use of AI is governed.
Far from slowing down innovation, these mechanisms constitute the safeguards that make its deployment on scale possible.
Define what AI can do
Organizations must establish levels of autonomy according to usage:
- automatic execution for simple and low-risk tasks,
- recommendations for intermediate decisions,
- generation of content subject to human validation,
- advisory role only for the most critical decisions.
Integrate validations into workflows
A human must retain responsibility for validating the relevance and conformity of the results produced by the AI.
When these validation points are integrated directly into processes, teams can work faster because responsibilities, escalations and exceptions are already defined.
AI is a tool, not a compass.
Make traceability a default requirement
In Europe, regulatory requirements in terms of explainability, fairness and auditability are particularly strong.
Decisions impacting customers, pricing or claims must be traceable and explainable. If humans remain responsible, then governance is the operational condition.
HR at the heart of scaling up
In the context of large-scale AI, human resources become a key function.
They will have to play an increasing role in the transformation of professions, career paths, incentive mechanisms, training and internal mobility. Historical definitions of roles themselves become a source of friction when they no longer reflect the reality of work.
This will involve creating new trajectories to allow employees to progress towards missions with higher added value, potentially several times during their career.
Adaptability becomes a central skill. Mid-career requalification schemes could thus become structural, with increased emphasis on digital and AI skills.
Rethink roles, responsibilities and incentive models
Remuneration models could also evolve.
If AI takes over some complex or risky tasks, companies will need to review how they assess the contribution of different roles. At the same time, certain functions – notably legal, compliance or risk management – will see their importance reinforced.
AI indeed introduces new regulatory obligations, particularly in insurance, where the requirements relating to data protection remain high.
As responsibility cannot be delegated to AI, organizations must strengthen explainability, bias management and governance mechanisms.
Conclusion
Scaling up AI is above all a work transformation challenge, much more than a technological challenge.
AI introduces a new “teammate” into the company, which requires new structures, new responsibilities and adapted governance.
Successful organizations will be those that:
- consider AI as a true partner,
- rethink their workflows around its capabilities,
- put in place solid governance frameworks,
- develop the AI culture of their teams,
- support employees in their skills development,
- while maintaining clear human responsibility.
Ultimately, the competitive advantage will not only come from the performance of the algorithms, but from the ability to build the right relationship between humans and AI: an AI that augments human capabilities, and employees capable of managing it effectively.
