Garantie Avantage Plus

From inbox to structured warranty claims with Agentic AI

Accelerating warranty claims processing with agentic AI

The context

Garantie Avantage Plus is a Canadian provider of extended mechanical protection plans for new and used passenger vehicles, leisure vehicles, and recreational vehicles. Their tiered warranty programs help cover mechanical failures of parts (think engines, air conditioning, motorhome appliances, and more) after the manufacturer’s coverage ends.

Their claims process used to run entirely through a central inbox, where agents handled everything manually. Each submission meant sorting attachments, retyping data from PDF work orders or images sent from garages, reviewing contracts line-by-line, and chasing down missing details. It worked, but it was slow and repetitive.

Here’s how we integrated agentic AI into Garantie Avantage Plus’ existing warranty, contract, and ticketing software to remove bottlenecks, speed up their claims process, and give their team more time for meaningful work.

Project details

Industry

Other

Technologies

GPT-5 mini (via OpenAI API)

Gmail API

Custom orchestration system

Open source mail client

Services

The challenge

The core challenge in modernizing a manual, inbox-based claims process was building an agent smart enough to handle messy, unpredictable submissions, while still being reliable enough to earn the trust of the claims team. Claims don’t arrive in cleanly formatted templates. They typically show up as grainy scans with no selectable text, handwritten notes, jumbled attachments, and emails from garages that all communicate differently. We needed a system that could process all of it quickly, consistently, and accurately. 

Another hurdle was accounting for the edge cases that claims agents deal with every day. We couldn’t just build for the ideal scenario—we had to create a system that could spot and adapt to the unusual. That meant digging into the realities on the ground, like VIN number typos, missing contract details, inconsistent labour and parts breakdowns, and preferred language handling. These complexities shaped how we divided tasks into smaller, manageable steps, then used prompt engineering, structured data extraction, and format control to guide the system’s behaviour. We also built logic to flag incomplete or unsupported claims before they became blockers.

Our third challenge was designing fallback systems that could keep everything moving when things didn’t go as planned. Not every file could be parsed cleanly, and not every submission followed the rules. So we taught the system to recognize its limits. When it couldn’t move forward with confidence, it handled the case off for human review instead of stalling out. These built-in safety nets made the process more resilient and helped build trust in the tool.

Mechanical woman writing notes on a pad, beside a car hood.

Person working on a computer with manual notes and a calculator.

The task at hand

Garantie Avantage Plus came to us with a clear vision, so we jumped straight into build mode. In the first week, we worked closely to align on goals and logic flows. They brought detailed diagrams, and our AI software developers brought the tech know-how to bring it all to life.

At the core of the solution is the GPT-5 mini model, accessed through OpenAI’s API. Because it’s multimodal, we could process PDFs, images, and text within a single system. Everything runs securely within Garantie Avantage Plus’ own OpenAI environment. Around the model, we built a custom orchestration layer to control inputs and outputs, giving us both precision and flexibility.

The project initially used an open source mail library to connect to Gmail’s IMAP server. This approach gave us the flexibility to swap out email providers in the future, like switching to Outlook, without rewriting major parts of the system. Early tests ran smoothly, and the IMAP rate limits appeared to meet Garantie Avantage Plus’ needs.

But there was a catch. IMAP access is still controlled by Google, and they strongly prefer users to go through their official Gmail API. That means IMAP rate limits can change without notice, and any updates to the limits on their public documentation may not reflect reality. Sure enough, during the week of a widespread AWS outage in the US East region, we started noticing unexpected slowdowns. We can’t say for certain that Google throttled IMAP access, but the timing lined up, and the system had been stable for weeks beforehand.

To stay ahead of any future issues, we migrated to Gmail’s official API. It offered much higher throughput and significantly faster performance. The tradeoff was reduced flexibility if Garantie Avantage Plus ever wanted to switch providers. Still, it made sense. The result was smoother processing, fewer delays, and a more reliable setup overall.

The project took 3.5 months from start to finish, with close collaboration every step of the way. Garantie Avantage Plus had a full month to test, explore, and share feedback before go-live, which helped us fine-tune the experience.

The end result

Garantie Avantage Plus’ updated warranty, contract, and ticketing portal went live in November 2025. The team has been especially pleased with how fast and responsive the new claims system feels in day-to-day use.

The new flow kicks in the moment a customer or garage submits a claim by email. The AI agent scans the message, checks whether it qualifies as a claim, and filters out anything unrelated, such as transfers or general inquiries. Instead of sending automatic replies when information is missing, the system now replies directly within the client’s mailbox, listing what details were found and what’s still missing. This gives claims agents full visibility and control. They can double-check what the model picked up, spot anything it might have missed, and decide how to respond. We’re also adding a new mailbox tag so agents can manually move any email thread when they’re ready for the system to create a ticket. This updated flow keeps humans in the loop while still cutting down on the grunt work.

Once all required information is in place, the system creates a clean, consistent ticket. It organizes the submission, cross-references it against the customer’s warranty, and highlights the contract sections that apply to the claim. Then it performs a high-level online search to compare part numbers against market prices, surfacing suitable alternatives if needed. This gives the claims agent a clear view of whether the repair aligns with the customer’s coverage and whether the quoted prices are fair. With everything centralized and formatted, agents can focus on assessing the claim and making the final call (approve or deny) based on facts.

What’s next

The new tool is already cutting inbox clutter, reducing repetitive follow ups, and giving lean teams more breathing room when staff are away. But there’s still more to come! Later this year, we’re rolling out a new feature to handle a common real-world scenario: garages sending an estimate first, then following up later with the final bill. Instead of creating duplicate tickets, the system will detect and merge related documents into a single claim, even when the information has been updated during the repair process. This will keep submissions organized, claims flowing, and the process even more efficient.

This project is proof that AI isn’t here to replace people. It’s here to support them. By automating the repetitive parts of claims processing, we’ve freed up time and mental energy for human agents to focus on what they do best: thoughtful analysis, good judgment, and customer care.

With the right tools in place, AI becomes a trusted teammate, not a threat. Interested in exploring what Agentic AI can do for your organization? Get in touch.