The complete guide to implementing AI Agents: lessons from Wakam, leading European insurer

About Wakam

Wakam is a B2B2C insurance company that designs custom insurance solutions distributed as white-label products through its Play & Plug® technology platform. Their solutions reach customers through 100 distributor partners across 30+ countries. With a turnover of nearly one billion euros in 2023, Wakam has established itself as the European leader in digital and embedded insurance.

As a mission-driven company employing 250 people across 5 European countries, Wakam operates with the purpose of "making insurance transparent and impactful." The company has built its corporate culture around 11 cultural markers that promote collaboration, curiosity, learning, open-mindedness, and innovation.

The data silo problem every enterprise faces

Enterprise AI adoption often gets stuck in endless proof-of-concept cycles. Companies build impressive demos but struggle to achieve meaningful adoption at scale. Wakam, one of the world's top fintech companies, shows a different path: 70% employee adoption and 136 deployed AI agents within months of launch.

Before considering AI agents, Wakam faced a challenge familiar to most growing enterprises: critical business knowledge was trapped in silos across the organization. With 250 employees spanning 5 countries and multiple business units, finding the right information at the right time had become a significant productivity issue:

• Insurance operations needed constant access to regulatory documentation, partner contracts, and operational procedures.
• Business development teams required comprehensive partner profiles before client meetings.
• Customer service representatives needed quick access to policy details and claims procedures.
• Legal teams had to navigate complex regulatory requirements across different jurisdictions.

This information existed across Notion documentation, SharePoint contract repositories, Slack conversations, Excel spreadsheets, and other databases. Accessing it meant knowing where to look, who to ask, or spending hours searching across multiple platforms. 

In 2023, when GPT-3.5 and Retrieval Augmented Generation (RAG) emerged, Wakam's leadership saw an opportunity: AI could unlock these information silos and give employees instant access to the organization's collective knowledge.

Build or buy: why building AI agents in-house fails for most enterprises

Enterprise AI implementations typically fail for three critical reasons: 

• resource constraints that prevent keeping pace with market evolution, 
• the complexity trap of building everything from scratch, 
• and the misconception that technical success equals business adoption. 

Wakam's initial journey illustrates each of these failure modes.

Resource constraints create an impossible race against AI market velocity

When Wakam first encountered GPT in late 2023, their five-person data science team began building a custom AI chatbot with RAG capabilities. While they had the expertise to create a working prototype, maintaining and extending it required constant engineering effort.

Every new feature meant weeks of development time. They soon realized the AI market was advancing faster than their small team could match. To offer employees functionality comparable to commercial platforms, they would need to triple their team size, an unrealistic proposition for most enterprises.

Moreover Wakam's team found themselves rediscovering fundamental challenges that AI platform companies had already solved: implementing effective RAG, managing vector databases, handling model orchestration, and creating user interfaces. 

Technical success doesn't guarantee adoption

Even when Wakam's custom chatbot worked well technically—answering questions using company data and browsing the internet—it remained a tool used primarily by technical team members. Without change management, training, and organizational support, sophisticated AI implementations struggle to achieve meaningful adoption.

Wakam recognized these limitations early and pivoted to specialized AI agent platforms. Purpose-built solutions let companies bypass years of trial and error and focus on driving business impact, rather than matching the development velocity of dedicated AI companies.

Choosing an AI agent platform for your enterprise: Wakam's evaluation framework

Operating in a regulated industry with complex data needs, Wakam's platform selection went beyond basic functionality to address real enterprise requirements: security, compliance, and seamless integration.

Model agnosticism

In 2024, the AI industry was shifting rapidly between OpenAI, Anthropic, Mistral, and emerging models. Wakam recognized that tying their infrastructure to a single provider could create long-term risks. They needed a platform that could adapt as the model market evolved, giving them flexibility to switch providers based on performance, cost, or compliance requirements while protecting them from dependency on a single provider.

RAG capabilities and data integration 

Wakam needed an AI platform that could securely tap into their proprietary knowledge: insurance regulations, partner contracts, operational procedures, and market intelligence. The platform had to integrate seamlessly with existing data sources (Notion, Slack, Snowflake, HubSpot and SharePoint) and be simple enough for business users to manage without technical support.

Security and compliance 

As an insurance company operating across 32 countries, Wakam faced strict data protection regulations. They needed enterprise-grade security, audit trails, and the ability to control data access at granular levels. The platform of choice had to support their existing identity management through SSO (Entra ID) and provide clear data governance capabilities.

Extensibility through APIs 

Wakam also needed the ability to build custom integrations and specialized agents. They required robust API access that would allow their developers to create sophisticated workflows while still leveraging the platform's core capabilities.

Solving the “sensitive” data problem through advanced permissions

One of the biggest concerns enterprises have about AI agents is data security: how do you give agents access to information they need while ensuring sensitive data doesn't end up in the wrong hands? 

Dust's approach solves this through a dual-layer permission system designed for AI-first environments. Dust organizes data into spaces—data containers that can be company-wide accessible or restricted for confidential information accessible only to designated agents.

Agents only retrieve information from their assigned spaces, and users can only interact with agents if they have access to all the spaces those agents require.

Compliance teams could create agents accessing sensitive regulatory documents through restricted spaces, while finance teams could build agents with financial data accessible only to executives and finance members. This dual-layer approach, controlling both agent-to-data and human-to-agent access, enabled Wakam to democratize AI while maintaining strict data governance.

Wakam's evaluation led them to select Dust. 

The implementation playbook that drove 70% adoption at Wakam

Wakam achieved 70% monthly active usage within two months of launching Dust by treating AI agents as both an organizational change and a new way of working that required comprehensive support:

1- Executive sponsorship and purpose clarity

Wakam's implementation began with unambiguous executive committee sponsorship. Leadership communicated that AI agent usage was a strategic priority and a fundamental shift in how employees would access information and complete tasks—not an optional experiment.

This messaging helped employees understand they were adopting a capability that would make their jobs more effective. Leadership tied AI agent usage to business outcomes rather than technological innovation, making adoption feel essential.

During weekly company meetings, leadership regularly featured AI agent success stories, new capabilities, and usage metrics. These updates maintained momentum and reinforced the importance of adoption while celebrating employees who found innovative applications. They also positioned AI agents as the preferred method for information retrieval and task automation, not as an additional option competing with existing workflows.

2- Comprehensive enablement and employee empowerment

Wakam established support systems even before the company-wide launch, with multiple touchpoints designed to accelerate adoption across different learning styles. But the most critical aspect of their approach was empowering every employee to build their own agents. Of their 136 deployed agents, approximately 40 were built by the AI Engineering team while 96 were created by employees across different business units.

The AI Engineering team couldn't possibly understand every business challenge well enough to build optimal solutions. They gave employees the support needed to create their own agents and directly address their own business needs:

• Comprehensive training programs: Every employee received training on AI agent capabilities, covering practical use cases relevant to each department, hands-on practice with real business scenarios, and guidance on when to use AI agents versus traditional tools. Advanced sessions taught employees to identify opportunities for new agents, build them independently, structure agent instructions effectively, and troubleshoot common problems.
• Weekly open office hours where any employee could bring questions, challenges, or requests for help. The AI Engineering team used these sessions to identify common stumbling blocks and refine their support approach.
• Dedicated Slack support channel for real-time AI agent support, the channel became both a support mechanism and a knowledge-sharing platform where employees could see solutions to common challenges and learn from each other's creative applications.
• Hackathons development sessions where cross-functional teams spent full days building specific AI agents collaboratively. These sessions paired business experts who understood the problems with technical team members who could guide the implementation, resulting in sophisticated agents that neither group could have built alone.
• Documentation and curation systems: Wakam developed systematic documentation for each agent and created a "meta-agent" that helped employees build their own Dust agents.
• Progressive complexity support: The AI Engineering team structured support to match complexity levels. Simple agent creation is fully self-service. Moderately complex use cases receive guidance on tool selection and data integration. Highly complex agents receive full collaborative development support from the AI Engineering team.

While Wakam has achieved 70% monthly adoption, their ambition is to reach 90% monthly usage and 70% weekly usage. More importantly, they want every team to build at least one agent tailored to their specific roles. 

4- Measuring and maintaining momentum

Wakam built internal dashboards to track adoption metrics, they monitored user activity rates by team, identified the most valuable agents, and tracked productivity impact across different use cases. 

Early wins emerged quickly: the legal team cut contract analysis time by 50%, while data teams enabled self-service intelligence that dramatically reduced processing time. This data informed ongoing support priorities, helped identify opportunities for additional training or tool development, and provided compelling evidence for continued expansion.

The future is scaling from “assistants” to “agents”

Wakam's AI journey illustrates the natural evolution from simple assistants to sophisticated action agents. Understanding this progression helps enterprises plan their own implementation roadmap.

Phase 1: Knowledge assistants (Wakam’s foundation)

Wakam started with basic assistants that help employees work more effectively, e.g.: HR policy assistant, or contract review assistants. These knowledge-focused assistants represent the natural starting point for enterprise AI, where agents help humans access information and improve output quality while humans retain responsibility for all actions and decisions.

Phase 2: Action agents (Current state)

Wakam is now building agents capable of taking actions autonomously vs just providing information. Two agents illustrate this:

• Harvey (Legal Agent) operates across the corporate legal team's entire digital workspace. With access to Notion, Outlook, web search, SharePoint, and calendar tools, Harvey can read, write, and remember context. While human-activated, Harvey handles complex corporate legal workflows that previously required manual coordination across multiple agents.
• MoneyPenny (Personal Productivity Agent) acts on your behalf across Wakam's digital workplace—Outlook, Slack, Notion, and HubSpot. MoneyPenny retrieves emails, prepares meetings, synthesizes weekly activity, writes to Notion pages, and summarizes Slack mentions. Rather than employees choosing which agent to use for each task, MoneyPenny orchestrates multiple actions based on intent.

Phase 3: Autonomous agents 

Wakam’s ultimate goal is agents that operate as domain experts and are capable of addressing entire job functions or major portions of role responsibilities. These agents will increasingly operate proactively, reacting to system events, generating scheduled analyses, monitoring business metrics, and alerting humans only when exceptions require attention or decisions exceed predetermined parameters. In highly regulated industries such as insurance, these agents will operate strictly within predefined, validated, and auditable boundaries aligned with applicable regulatory frameworks.

This evolution from human-initiated to agent-initiated workflows requires careful change management and risk assessment. Not every process should be automated, and the order in which you automate matters for both effectiveness and organizational acceptance. Wakam’s approach is grounded in a compliance-by-design philosophy: every autonomous capability is subjected to legal, risk, and security assessments before deployment, with periodic reviews.

Your enterprise is ready for AI agents

The key insight from Wakam's story is that successful AI implementation is fundamentally a change management challenge, not only a technology challenge. Their 70% adoption rate came from executive sponsorship, comprehensive enablement systems, and empowering employees to build their own agents.

This approach only becomes feasible when enterprises partner with AI platforms that handle complex technical infrastructure while providing flexibility to address unique business challenges. 

Start with knowledge assistants, focusing primarily on knowledge retrieval, data synthesis, or document restructuring, to build organizational comfort and momentum. Even simple use cases create value and prepare your culture for more sophisticated applications. Always empower your employees to create agents for their own business challenges. As adoption grows, identify high-value use cases where action agents can deliver measurable impact.

The question isn't whether AI agents will transform your enterprise operations, but whether you'll implement them strategically to maximize adoption and business value. The enterprises building this capability today will have a decisive advantage as AI agents become essential to internal operations.

Special thanks to Etienne Debost (CTO @Wakam) and Sofia Calcagno (Head of Machine Learning eXpertise @OCTO) for sharing their insights and experiences. Most of the information and lessons highlighted in this story are drawn from their presentation at the Paris Crafting Data Science Meetup. For a deeper dive into Wakam’s AI journey, we encourage you to watch the full session here.