Opinion8 min read
Why 90% of AI Projects in Mid-Market Companies Fail — And What the Other 10% Do Differently
Most AI projects don't fail because of technology. They fail because nobody understands how the process actually works.
Read more
Use Case
Claims Processing with AI: From 48 Hours to 4 Hours Turnaround Time
January 27, 20267 min read
48 hours. That's how long the average initial processing of a claim took at a mid-sized insurer. Not because the adjusters were slow, but because the process was. Documents in one system, policy data in another, communication via email, approvals via physical signature folders. What follows is the story of a transformation — no buzzword bingo, just concrete numbers.
The Starting Point
The insurer processes over 30,000 claims annually. The process: claim comes in (email, mail, portal), gets manually recorded, policy data is checked in the legacy system, coverage is assessed, the case is categorized and either settled directly or escalated to a specialist. Average processing time: 48 hours. Error rate in initial recording: 12%. Customer satisfaction with processing speed: mediocre.
The Problem in Detail
The core problem wasn't lack of competence, but fragmentation. An adjuster had to use three different systems for a single claim, transfer data manually, and make decisions for which they first had to gather information from various sources. 60% of processing time wasn't claims work — it was system navigation.
The Solution: AI Agent with Human-in-the-Loop
We developed a specialized AI agent that handles initial processing. The agent receives the claim, extracts relevant information (claim type, amount, date, policy number), accesses the legacy system via MCP, checks coverage, and creates a processing recommendation. For clear cases — such as glass damage under 1,000 euros with unambiguous coverage — the case is settled automatically. For more complex cases, the agent creates a complete file with all relevant data and escalates to an adjuster. The adjuster doesn't just get the case, but the complete preliminary work: policy data, coverage analysis, historical claims data for the customer, settlement recommendation.
The Results
After three months in production: Average initial processing time: 4 hours (down from 48). Automatically settled cases: 34% (without human intervention). Error rate in initial recording: under 2% (down from 12%). Adjusters spend 70% less time on data entry and system navigation. Customer satisfaction with processing speed increased measurably.
What We Learned
Three insights from this project. First: the agent's confidence score was critical. We adjusted the threshold for automatic settlement several times in the first weeks — not because the AI was bad, but because the right balance between speed and safety only reveals itself in production. Second: the adjusters weren't losers of this transformation. They went from data entry clerks to decision makers. Third: integration via MCP was the key. Without standardized access to three core systems, the project would have failed due to integration complexity.