SGF Trip Assistant
Building the First AI-Powered Tourism Assistant in the U.S.
Client
Springfield CVB, Missouri's tourism hub.
Goal
Help people plan weekends and events—without digging through dozens of pages.
Build
8-week sprint, powered by GPT-4 Turbo and custom RAG.
Traffic
6,000+ messages, steady 550–700/month since launch.
Insights
Most users ask about food, family fun, and what’s on this weekend.
The Challenge
Springfield’s tourism website was full of great content—but it was too much. Visitors needed one place to do three things quickly:
Discover what’s happening this weekend
Build a custom itinerary with venues, food, and nightlife
And trust that the recommendations came from credible local sources.
The CVB team also had a vision: create the first AI-powered tourism tool in the U.S.—but only if it felt like genuine, local advice, not a generic chatbot reading from a brochure.
The Mandate
The tool had to be trustworthy, simple, and fast. Every answer needed to cite a live page or PDF—no hallucinations, no guessing. The interface had to feel intuitive and human—more like texting a local than using a help desk. We had just eight weeks to launch, with tourism season on the horizon. And once it was live, the content needed to stay fresh—refreshed quarterly, with live usage analytics that would let the CVB learn from every visitor interaction.
What We Built
We started with a full content audit, ingesting more than 400 pages of existing site content, partner blogs, event feeds, and downloadable guides. That gave the model a complete foundation—everything a human local might know, but sourced and verifiable.
From there, we designed a custom Retrieval-Augmented Generation (RAG) pipeline to keep answers accurate and on-brand, with citations under every response.
The experience itself lives on a mobile-first microsite with a clean chat interface. Users can explore with natural questions and share or copy their itineraries in just a few taps. On the backend, we included real-time analytics and an emoji-based satisfaction tracker to measure what visitors really want to know.
To keep it all relevant, we packaged a quarterly retraining kit so the content stays aligned with seasonal events, updates, and new listings.
Early Momentum
In its first 11 months, the assistant handled over 6,000 messages with no advertising. Nearly 1,650 unique visitors used the tool, and most stayed to ask 3–4 follow-up questions—clear evidence they were actively planning, not just browsing.
August 2024 marked a peak with 914 messages, but even after the season cooled, the tool held steady at 550–700 messages per month. The most popular queries? Exactly the long-tail questions traditional SEO never captured: family things to do, best local food, rainy-day activities.
One early user put it best:
“This feels like chatting with a local friend, not a robot.”
Why It Worked
Three things made this project stand out.
First, trust was built in. Every answer came with a source link—no mystery, no guessing. That transparency converted skeptics immediately.
Second, we shipped fast. From kickoff to launch in just eight weeks, the CVB proved that helpful, functional AI doesn’t need a 12-month RFP cycle.
And finally, the user experience came first. No logins, no loading screens, no barriers—just answers. That clarity drove adoption.
Takeaways for Other DMOs and Cities
Don’t start from scratch—start with the content you already have. With the right RAG pipeline, your blogs, PDFs, and event calendars become a 24/7 concierge.
Want trust? Show your sources. Citations aren’t just academic—they’re your credibility.
And if you’re still guessing what visitors want, stop. Let them ask, then learn from the data. That’s where your next campaign begins.