An AI customer support assistant that answers from approved content
A travel and vacations company in North America issues travel certificates, and its support team kept answering the same questions about registration, bookings, deposits and travel information. We built a web chat assistant on ASP.NET Core and Azure OpenAI that answers certificate holders from the company's own approved content and its certificate data. It's live in production. When a question needs a person, the assistant creates an agent request for the support team.
| Before | With the assistant | |
|---|---|---|
| Repetitive questions | Certificate, registration, booking and deposit questions all reached the support team | The assistant answers them, so fewer repetitive questions reach agents |
| Certificate details | Agents checked certificate information by hand | The assistant checks certificate and business context through the client's API |
| Finding answers | Staff searched for content manually | Staff search and review conversations in an admin portal |
| Content changes | Updating answers leaned on developers | The support team manages FAQs and supporting content without developers |
| Hard cases | Agents were spread across simple and complex questions | Agents focus on cases that need a human |
The assistant understands what the customer means, checks their certificate, and answers only from content the client has approved. If it can't, a person takes over.
During the measured production period, the assistant handled 1,526 customer conversations from 1,225 distinct certificate holders.
That's 96.8% of questions answered using the client's FAQs, Terms & Conditions, inventory, certificate data and knowledge content.
53% of questions arrived outside the 9-5 Pacific, Monday-Friday support window, including 394 weekend questions.
The assistant answers from FAQs, Terms & Conditions, knowledge documents, inventory and certificate data the client approved.
When a question needs a person, the assistant creates an agent request, so the customer always has a next step.
The support team curates the FAQs and content, reviews conversations, and spots recurring issues from dashboards on activity, satisfaction and answer sources.
Yes. It's in production, and the results above are measured production counts.
From approved content only: FAQs, Terms & Conditions, knowledge documents, inventory information, certificate data and approved documents. During the measured period, 2,474 of 2,556 questions were answered this way.
It creates an agent request so the support team picks the question up and the customer isn't left without a next step. The support team also curates the content the assistant answers from.
We don't claim that.
Tell us which questions keep landing on your support team. We'll show you how an assistant grounded in your approved content, with people handling the hard cases, could work.

Founder and CEO

Chief Sales Officer
Thanks, we’ve got your details. A named member of our team will reply within one business day.
Talk to a QServices solutions architect about your ERP. You'll leave with two or three automation ideas you can act on, even if you never work with us.