Frederic Stallaert

The through-line

Across Frederic Stallaert's posts and talks, one preoccupation keeps resurfacing: insurance operations fail not at the point of AI deployment but at the point where data enters the organisation. Whether he is talking about claims, bordereaux, or GenAI pilots, the argument is the same: "AI doesn't fail at the idea stage. It fails at the ingestion stage" , because "over 80% of enterprise data is not AI-ready" . Over time this has sharpened from a general efficiency message, sneller werken, meer volume, minder kosten , into a more pointed critique of the market's habit of waiting for someone else to fix intake, whether that is Lloyd's Blueprint Two or a vendor's AI theatre . The later posts also tie this data argument explicitly to regulation, framing auditability and governance as inseparable from the data problem rather than a separate compliance layer .

On efficiency as a means, not an end

Stallaert repeatedly pushes back on treating speed and automation as inherently good for the customer. "Efficiëntie is geen doel op zich. Het is een middel" , he argues, and the real test is what happens with the time that is freed up: "de echte vraag is wat je doet met de tijd die je wint" . If that time is not reinvested "in beter advies, persoonlijk contact en de momenten waarop klanten je echt nodig hebben" , the organisation has simply delivered the same service faster, which raises the question "hoe ga je dan je marge in de toekomst nog verdedigen?" .

On intake as the real bottleneck

For Stallaert, the "inbox is still the front door" of insurance , and most of the friction people attribute to claims handling or delegated authority actually starts before a human ever opens the file. "A claim does not start when a handler opens it. It starts when it arrives" , and the same logic applies to bordereaux: "the bordereaux did not get easier while we waited" for a market-wide fix, because "the market spent seven years treating intake as someone else's problem to solve" . He extends this into a broader diagnosis of AI failure: the issue is rarely the model, it is that "the data underneath wasn't ready" , so the operative question becomes "what is our AI actually reading?" .

On governance and auditability

Stallaert treats regulatory defensibility as a design requirement, not an add-on. "In insurance, compliance isn't a checkbox. It's a condition of operating" , and he insists accuracy and auditability "need to be baked in. Not bolted on" . He extends this to enterprise AI generally, arguing the biggest risk "isn't the model. It's whether you can explain, validate, and defend the outcome when an auditor, regulator, or customer asks how it was produced" .

On operational discipline in a soft market

Reporting from the MGAA Annual Conference 2026, Stallaert frames a market shift: "the hard market hid operational debt. The soft market will expose it" . He lists concrete symptoms, "the claims handler still sorting PDFs," "the MTA that never gets keyed," "the finance team reconciling spreadsheets by hand" , and reclassifies them as "margin, service and capacity risks" rather than back-office annoyances . His prescription is explicit: "the market does not need more AI theatre. It needs proof that technology can remove friction without removing control" .

From the stage

In interviews, Stallaert fills in the career and product reasoning that the LinkedIn posts skip. He describes joining ML6 in 2016/2017 as one of the first European companies applying AI in business, which is where he built his expertise "bridging the gap between technical AI capabilities and business implementation across large organizations" ¹⁴. He frames his later move as a search for roles with greater organisational impact, which led him toward "productizing AI to increase accessibility for companies of all sizes" ¹⁵. On the mailroom specifically, he argues that large language models let document interpretation and task generation happen "without requiring complex configuration or AI expertise," which is what allows mailrooms to become more efficient ¹³. He also describes Paperbox's mission in broader terms than the written posts do, combining human talent with AI capability while scaling "across multiple countries" and staying "lean, disciplined" in execution, particularly within insurance ¹⁵.

Takeaways

• Measure intake time, not just handler speed: "the clock is already running" before a claim is even opened .
• Do not treat freed-up time from automation as a win by default; reinvest it in advice and customer contact or margin erodes .
• Audit your own data pipeline before scaling AI: "over 80% of enterprise data is not AI-ready" and the inbox remains the real front door .
• Build auditability and accuracy into AI systems from the start rather than adding governance after deployment .
• In a softening market, fix the operational debt (unkeyed MTAs, manual reconciliation, PDF sorting) before it becomes a margin and capacity risk .
• Do not wait for market-wide infrastructure fixes like Blueprint Two; automate your own processes now .