Case Study
If you have ever woken up with a sudden neck twinge, you know what happens next. You book a same day physio appointment and your provider submits your claim. But behind the scenes, your insurer must confirm that your provider is real, licensed, and in good standing before processing the claim. Historically, that meant employees jumping between internal registries, association and clinic websites, hoping they match. Multiply that by hundreds of claims a day and you get a slow, tedious bottleneck that drains hours of time, slows claims turnaround, and frustrates customers and staff alike.
What if an AI Agent could take on the verification work, make it explainable, and keep humans focused on exceptions? That is the question our 2025 Innovation Challenge winners set out to answer.
The Provider Verification AI Agent scans the internet like an investigator. It checks college registries, association and clinic sites, and trusted professional listings, reasons about matches, and returns a cited decision for reviewers. The four core pillars that power this AI agent include:
The System Prompt
This is the agent’s instruction manual. It defines the role, objectives, behavior, and constraints. In our case, the system prompt states that the agent’s purpose is to validate health providers by searching authoritative sources, then explain and cite evidence for every decision.
The User Prompt
This carries the dynamic inputs that drive each run. It includes provider name, business type, location, and the matching rules the agent applies to determine a true match. The prompt gives the agent clear parameters so verification is targeted and consistent.
The Context
This is where the agent gets powerful. Context gives access to structured, relevant data before going online, for example a CSV export of all registered physiotherapists in Ontario. The agent uses this dataset to narrow its search, confirm a name in an official registry, and only then cross-check on the web if needed.
The Tools
These are the agent’s hands and eyes on the web. Web Search looks up information across public sources. Web Reader extracts and interprets text directly from visited pages. Together they give the agent reliable reach and precise reading comprehension.
The System Prompt, User Prompt, Context, and Tools combine to create an intelligent, reliable, and safe automation that can research and validate providers with minimal manual effort.
What started as an Innovation Challenge concept is now running in the real world. Today the Provider Verification AI Agent is already in production for one of our insurance clients and early results show promise.
The agent has already boosted claims adjudication speed, turning many five-minute manual checks into seconds. Backlogs that used to stretch for weeks are reduced as more claims clear on the first pass. Context-first matching lifts hit rates on known providers, while agentic web search and reading fill the gaps when needed. Each decision includes sources and a short explanation so reviewers can trust the outcome and analysts spend less time hunting across registries and clinic sites and more time on true exceptions. These metrics roll up to the outcomes that matter most: speed, accuracy, lower risk, and a better experience for both members and employees.
Technology Used
