The Problem With Contact Forms
Every agency knows the feeling. A potential client lands on your site at 11 PM, browses your services, has a question about pricing, and then leaves. No form filled out. No email sent. Just another lost opportunity sitting in your analytics as a bounce.
We were losing leads the same way. Our contact form sat at the bottom of the page collecting dust while visitors with real buying intent slipped through the cracks. The math was simple: if someone has a question and nobody is there to answer it, they move on to the next agency.
Live chat was the obvious alternative, but it came with a catch. Someone had to be on the other end. For a lean agency, that meant either hiring for coverage or accepting that "live" chat was only live during business hours. Neither option worked.
Why We Built Maple
Before offering our AI Concierge service to clients, we wanted to prove it on ourselves. If we could not make it work for our own sales process, we had no business selling it to anyone else.
So we built Maple - a custom AI sales assistant designed to do three things well:
- Answer visitor questions instantly about services, pricing, and timelines
- Detect buying intent and guide conversations toward booking or lead capture
- Capture qualified leads through a frictionless form, then notify us immediately
The goal was not to replace human interaction. It was to make sure no visitor left empty-handed, regardless of the time zone or hour.
How Maple Works
Here is what a typical visitor experience looks like:
The greeting. Five to ten seconds after landing on the site, Maple introduces itself with a friendly message. No pop-up. No modal blocking the page. Just a subtle chat bubble in the corner.
The conversation. Visitors ask about web design pricing, project timelines, what services we offer, or whether we can handle their specific industry. Maple answers naturally, pulling from a deep knowledge base of MooseBase's offerings. It knows the $0 down + $180/month pricing model, the 2-4 week timeline, and the difference between a standard website and a custom software build.
Intent detection. This is where it gets interesting. Maple monitors conversations for signals across four categories:
- Booking intent - phrases like "schedule a call" or "free consultation" trigger a booking button linked to our Cal.com calendar
- Lead intent - mentions of "quote," "pricing," or "get started" prompt a lead capture form
- AI interest - questions about AI chatbots or automation surface a link to our AI services page
- Urgency - words like "ASAP" or "tight deadline" trigger a rush booking option
These are not hardcoded keyword matches. The system uses word boundary detection across 50+ patterns to avoid false positives while catching real intent.
Lead capture. When a visitor is ready, Maple presents a progressive 5-step form: email, company name, project type, budget range, and timeline. The form is smart - if someone selects "New website" (a fixed-price service), it skips the budget question entirely. Every field after email is optional.
Instant notification. The moment a lead submits, our team gets an email with the visitor's details, their conversation history, and the page they came from. No lead sits in a database waiting to be discovered.
Under the Hood
For the technically curious, here is what powers Maple:
Google Gemini 2.5 Flash handles the AI. We chose it for the speed-to-quality ratio. Responses stream in real time via server-sent events, so visitors see text appear naturally rather than waiting for a full response to generate.
Cloudflare Workers runs the entire backend at the edge. That means sub-second response times regardless of whether the visitor is in New York, London, or Dubai. No cold starts. No origin server round trips.
Cloudflare D1 stores conversations with a 24-hour TTL. If a visitor closes their browser and comes back the next morning, Maple picks up exactly where they left off. Session IDs persist in localStorage across browser closes.
Rate limiting and validation protect against abuse. Each session is capped at 15 requests per minute, 50 messages per conversation, and 4,000 characters per message.
Analytics tracking logs 12 distinct event types - from chat opens and quick action clicks to lead form completions and booking link clicks. Every interaction feeds back into optimization.
What Makes Maple Different
Most chatbots feel like chatbots. They follow rigid scripts, give generic answers, and frustrate visitors more than they help. Maple was designed to avoid all of that.
It remembers. Conversations persist across browser sessions. A visitor can start chatting, close the tab, and pick up the same conversation hours later when they come back.
It adapts to the device. On desktop, Maple appears as a compact chat window in the bottom-right corner. On mobile, it opens as a drawer that takes up the right amount of screen space without blocking the page.
It only acts when relevant. Intent buttons (Book a Call, Get a Quote, Explore AI Services) only appear when the conversation warrants them. There is no generic "Would you like to schedule a demo?" after every message.
It knows when to step back. Maple does not try to handle everything. Off-topic questions get a polite redirect. Complex technical requirements get routed to a human call. The system is designed to qualify and connect, not to close deals on its own.
The Results
Since deploying Maple, three things changed:
After-hours leads are no longer lost. Roughly 40% of our chat interactions happen outside of traditional business hours. Every one of those would have been a missed opportunity with a static contact form.
Visitors engage faster. The average time from page load to first chat interaction dropped significantly compared to form submissions. When the barrier to asking a question is typing a message instead of filling out fields, people ask.
Lead quality improved. Because Maple has a real conversation before capturing details, the leads that come through are pre-qualified. They already know our pricing, our timeline, and what we offer. Discovery calls became shorter and more productive.
What This Means for Your Business
Maple started as an internal experiment. It worked well enough that we productized the approach as our AI Concierge service.
The same system that runs on moosebase.com - intent detection, lead capture, session persistence, edge deployment - can be customized for any business. Hotels use it for guest concierge services. Restaurants use it for reservations and menu questions. Professional services firms use it to qualify inbound leads.
If your website gets traffic but your contact form is not converting, the problem is not the traffic. It is the form. An AI assistant that actually talks to visitors, answers their questions, and captures their intent will outperform a static form every time.
Want to see what Maple could look like for your business? You are already looking at it. Try the chat bubble in the corner.