This specialty lender sees meaningful lift in loan conversions with agentic AI agents supporting customers through the application process.

The company’s loan application process had a persistent drop-off problem. Customers would start an application and abandon it — often at two predictable friction points: getting locked out of their account during a password reset, and running into trouble during the instant bank verification step. These weren’t complex problems, but they were common enough to meaningfully suppress conversion rates.

Making things worse, the chat experience the company had in place wasn’t built to help. It was menu-driven, which meant it could only handle questions it had explicitly been programmed to anticipate. A customer with a slightly different question — or the same question phrased differently — would hit a dead end and either give up or route directly to a human support agent with a question that didn’t require one.

The compounding effects showed up clearly in the support queue:

• Application drop-off at predictable friction points: Password resets and bank account verification were sending customers into a loop they couldn’t exit on their own, directly costing the business loan originations.
• A chat experience that couldn’t handle nuance: The menu-driven system could match a handful of pre-scripted intents. Anything outside those options produced no useful answer, sending customers to a human or out the door.
• Human support agents fielding answerable questions: The support team was handling a steady stream of questions that had nothing to do with the complexity those roles were designed for — coverage the business was paying for, applied to work that didn’t need it.
• No way to re-engage dropped-off applicants: When someone abandoned the application mid-process, there was no reliable mechanism to reach them and bring them back. Those leads were effectively lost.

The company was thoughtful about where AI belonged in an experience serving financially vulnerable customers. They weren’t looking for a chatbot to deflect volume. They needed AI that would genuinely help people complete a process, answer questions accurately, and know when a human was the right next step.

Quiq built two distinct capabilities for the company’s consumer loan product, each targeting a different failure mode in the existing experience.

• LLM-powered knowledge agent grounded in internal documentation: Quiq integrated directly with the company’s Confluence knowledge base — the same content their human support agents used internally — and transformed it into a customer-facing AI knowledge agent. Rather than mapping questions to pre-written scripts, the AI agent reads the question, understands what the customer is actually asking, and generates an accurate, specific answer from the source material. Customers get answers to general questions before they ever reach the support queue.
  
• Two agentic process guides for the highest-friction moments: For the two application steps where drop-off was most concentrated, Quiq built step-by-step process guides. When a customer hits a wall during a password reset or instant bank verification, the AI agent doesn’t just answer a question — it walks them through the process, step by step, until the issue is resolved. This is different from a knowledge response: the AI agent is actively guiding the customer through a task rather than informing them about it.
  
• SMS re-engagement for mid-application drop-off: After establishing the web chat experience, the company expanded to SMS to reach customers who had started an application but didn’t finish. Rather than letting those leads go cold, the AI agent reaches out over SMS with contextually relevant prompts to bring the customer back to where they left off.
  
• Outbound notifications for time-sensitive steps: The company also launched outbound notifications to keep customers moving through steps in the loan process that have a time component — appointment confirmations, document acknowledgements, and similar moments where a prompt response directly affects whether the process moves forward. Response rates from these notifications have outperformed traditional channels.
  
• Responsible AI guardrails built for a regulated, mission-driven context: Because their customers are often in financial hardship, the company approached the deployment deliberately. The AI agent is configured to be transparent about the limits of what it can answer and to offer a path to a human support agent when a question requires one. Quiq’s configuration reflects both the company’s compliance requirements and their underlying philosophy that AI in this context has to genuinely help people — not create new friction.
  

Password reset during an active application: A customer midway through applying for a loan gets locked out of their account. Previously, this was a dead end — the menu-driven chat couldn’t help, which meant either abandoning the application or waiting for a human agent during business hours. Now, the AI agent walks the customer through the reset process step by step, resolving the issue and returning them to where they left off in the application. A loan that would have been lost is completed.

Bank verification friction: Instant bank verification is one of the more technically unfamiliar steps for borrowers who haven’t connected a bank account through an app before. When a customer gets stuck, the AI agent guides them through what the system needs, why, and how to provide it — in plain language, without requiring them to reach a support agent.

General questions before application: A prospective borrower on the OppLoans website has questions about how the loans work, what rates look like, or what happens after approval. The AI agent draws on the same knowledge the company’s support team uses, generating a specific, accurate answer — not a generic FAQ response. The customer gets what they need to make a decision, without waiting for an agent.

SMS re-engagement after drop-off: A customer who started an application but didn’t finish receives an SMS outreach. Because the message is relevant to where they stopped in the process — not a generic marketing message — it prompts them to return and complete what they started. This channel has delivered measurable conversion lift on its own: a 1% increase in completed loan originations attributable to SMS re-engagement alone.

For the support team, the AI agent handles the front-line questions that were previously routing to humans by default. General product questions, application status questions, and the two highest-volume friction points in the application flow are now resolved before they reach the queue.

The team’s time is now directed toward the customers and situations that actually require a human — complex circumstances, sensitive conversations, and decisions that need judgment rather than information retrieval. That’s a meaningful shift in how the role functions day to day.

The outbound SMS program also changed the team’s relationship with conversion. Rather than passively waiting for customers to return after dropping off, the company now has an active mechanism to bring them back — run entirely through the AI agent, without requiring human outreach at scale.

By replacing a menu-driven chat experience with an LLM-powered AI agent grounded in their own knowledge base, and expanding to SMS re-engagement and outbound notifications, this specialty lender turned a customer experience gap into a direct revenue driver.

• Meaningful lift in loan conversion rates from web chat alone — measurable top-line revenue growth directly attributable to the AI agent
• Additional conversion lift through SMS re-engagement of customers who had abandoned the application mid-process
• More inquiries now resolved by the AI agent before reaching the human support team
• Reduced application drop-off at the two highest-friction points — password resets and bank verification — reduced through guided, step-by-step AI assistance
• Outperforming outbound notification response rates vs traditional channels that successful keep customers moving through time-sensitive process steps