Auto insurance company triples self-service rates and cuts email volume in half

This national auto insurance company was experiencing incredible growth. But that growth brought a new challenge: the support team found itself overwhelmed. As Director of Member Services, put it: “We were a victim of our own success.” The existing chatbot, built on Zendesk’s knowledge base chat, wasn’t enough. Agents were fielding a growing queue of repetitive, easily-answerable questions, while more complex member needs went underserved.

When the company’s previous chatbot received a customer question, it worked by mapping the message to a single pre-defined intent and returning a fixed block of text. If the question didn’t match a known intent, the bot failed. Even when it did match, every customer got the same static paragraph regardless of their specific policy, situation, or the nuance of their question.

This created several compounding problems:

• Low self-service rates: Customers who couldn’t get a useful answer from the bot simply gave up and emailed or called, sending routine questions directly to the support team.
• Impersonal responses: Insurance questions are rarely generic. A member asking about their deductible, a recent rate change, or how to add a vehicle needs an answer tailored to their situation, not a catch-all FAQ response.
• 16-hour coverage gap: The Member Services team works daylight hours. For the remaining 16 hours of every day, customers had no way to get answers. Not even for simple questions like how to make a payment online.
• High-volume, low-value email: Customers would send emails with incomplete requests like “I have a 2017 Camry, can I add it?”. Human agents then had to follow up to collect what they needed, wasting time on both sides.
• High maintenance burden: Every response had to be manually created and updated in the bot, consuming team bandwidth and creating a lag whenever policies or procedures changed.

With the growth they were experiencing, the team knew the existing approach wouldn’t scale. They needed a solution that could handle more volume, answer more intelligently, and cover the hours when the team wasn’t available.

Quiq built the company’s next-generation AI agent using large language model (LLM) technology grounded entirely in the insurer’s own knowledge base and policy documentation. Rather than matching questions to pre-written answers, the AI agent reads what the customer is actually asking, understands the intent behind it, and generates a personalized response on the fly.

The system was designed to:

• Understand natural language, not just pre-defined intents: The AI agent can interpret questions phrased in dozens of different ways, follow multi-part questions, and handle topics that no single knowledge base article covers—by synthesizing answers from multiple sources.
• Generate personalized, contextual answers: When a customer asks about updating a vehicle or making a payment, the AI agent responds to exactly what they asked—not a generic version of the question.
• Take the right action, not just the right answer: Rather than stopping at a text response, the AI agent collects the information and takes action. A member asks to update their coverage and the AI agent collects the information and submits it into the system.
• Reflect the insurance company’s brand voice: This insurer has a deliberately edgy, unconventional brand that sets it apart in a conservative industry. Quiq configured the AI Agent to match that tone, delivering responses that feel on-brand, while staying within the compliance and underwriting guardrails required of a regulated insurance carrier.
• Prevent hallucinations and ensure accuracy: Before any response reaches a customer, it passes through an attribution model fine-tuned for fact-checking. The system verifies that the generated answer is grounded in the insurance company’s knowledge base—not invented or hallucinated. If an answer fails the check, it loops back with additional instructions until it meets accuracy criteria. On sensitive topics, responses can be locked to verbatim policy language rather than generated text.
• Operate 24/7 without agent involvement: Because the AI agent runs continuously, it covers the 16 hours each day when the Member Services team isn’t available resolving roughly half of all inbound inquiries without human involvement.

From a member’s perspective, the experience is immediate and specific. A few examples from real use:

• Payment questions: One of the most common inbound inquiries was “How do I make my payment online?” Previously, a customer asking this question might wait 10 minutes on hold to get the answer from an agent. Now, the AI Agent instantly collects payment by texting or providing a payment link.
• Vehicle updates: When a member types “I financed a 2023 Camry—do I need to update my policy?” the AI agent responds to the specific vehicle they mentioned, explains what’s needed, collects the information, and updates the policy. The human support team receives a completed form with everything they need.
• Multi-source synthesis: When a member asked whether their aftermarket stereo would be covered after their car was stolen, the AI agent synthesized an answer that drew on multiple separate knowledge base articles explaining that comprehensive coverage was required and that the deductible applied. The team member who had written the knowledge base was surprised: “Where did you find that?” The answer hadn’t existed in any single article; the AI agent had assembled it from pieces.

For questions outside what the knowledge base can reliably cover, the AI agent is transparent about uncertainty rather than guessing. It provides the best available context while clearly indicating it isn’t giving a definitive answer, and offers the option to escalate to a licensed agent for complex questions that require human judgment.

For the Member Services team, the AI agent fundamentally changed the nature of the work. Routine inquiries such as payment questions, coverage summaries, how to add a vehicle no longer go to human agents first. The AI agent handles them directly, freeing the team to focus on complex member interactions, sensitive situations, and the kind of nuanced conversations where a human is required.

The reduction in email volume was particularly significant. Rather than receiving incomplete requests that required follow-up, the team now receives submissions with all the information necessary. This eliminated the back-and-forth cycle that was adding time and friction to routine requests.

The 24/7 availability also changed how the team thinks about capacity. Previously, any customer with a question outside of business hours had to wait or send an email that would sit in a queue overnight. Now, those customers get answers immediately, and the team arrives each morning to a smaller queue of genuinely complex issues rather than a backlog of simple ones.

By replacing a rigid, manually maintained chatbot with an LLM-powered AI agent, this auto insurer turned its member support operation from a bottleneck into a scalable advantage. Customers get accurate, personalized answers instantly at any hour of the day. The support team is no longer overwhelmed by routine questions. And the business now has the CX infrastructure to expand without degrading the experience.

Key outcomes:

• 3x increase in self-service resolution rate compared to the previous Zendesk-based chatbot
• 55% reduction in inbound email support tickets
• 75% CSAT score for AI-assisted interactions—matching the benchmark set by human agents
• 16-hour coverage gap eliminated: members now get answers around the clock
• Reduced team maintenance burden: knowledge base updates flow automatically to AI responses without manual chatbot content rewrites