Transformed a static pest control website into a conversion-driven platform by integrating AI-powered chat and online scheduling, reducing time-to-first-contact by 45% and moving 35% of new bookings to self-service channels — while improving mobile experience and site performance by 30%.
Impact
- 45% reduction in time-to-first-contact via AI chat (from hours to seconds)
- 35% of new bookings migrated to self-service online scheduling (from 0% baseline)
- 25% estimated drop in bounce rate via improved visual hierarchy and conversion CTAs
- 30% improvement in perceived brand trust score based on pre/post usability testing
The Problem
GecoPlax was growing in service capacity but constrained by a manual, phone-dependent contact process. The website drove traffic but converted almost none of it independently.
• 67% bounce rate — visitors arrived, found no immediate path to action, and left
• 4-hour average time-to-first-contact — all inquiries required a phone call during business hours, with an average 90-minute callback queue
• 0% self-service booking — no mechanism existed; every booking required human intervention
• Page load time of 4.3 seconds (LCP) — mobile visitors on slower connections abandoned
before the page finished loading
• Mobile traffic was 63% of total, but mobile bounce rate was 78% vs. 54% on desktop
The business cost was significant: an internal analysis showed that roughly 40% of web visitors who left
without contacting would have converted if they'd received an immediate answer to one of three questions — service area availability, pest type coverage, or pricing range.
Why now: A regional competitor had launched a basic chat widget and seen visible traction.
Leadership committed to a full conversion redesign rather than a patch.
What the research showed
We went in with three questions:
- Why do visitors leave without contacting GecoPlax?
- What questions do prospects need answered before deciding to book?
- What are the most common first-contact scenarios, and can they be automated?
Four tracks answered them: interviews with recent customers, six months of call centre logs, heatmap and session analysis, and an analytics baseline.
They pointed at the same thing from four directions. Visitors were not leaving because they distrusted GecoPlax or because they were only browsing. They were leaving because the site never answered the two questions standing between having a pest problem and booking a visit: do you cover my area, and do you handle my pest?
| Method | Participants / data | Key finding |
|---|---|---|
| User interviews | 8 recent customers — 4 who booked, 4 who left without booking | 5 of 8 said “I wasn’t sure if they covered my area or my pest type.” The primary barrier was uncertainty, not intent. |
| Call centre log analysis6 months | ~3,200 inbound calls | 38% of all first calls asked one of the same 12 questions — service area, pest type, pricing, availability. Every one of them was answerable without human judgment. |
| Heatmaps & session recordingsHotjar | 8,400 mobile sessions over a 30-day window | 71% of users scrolled past the contact form without interacting with it. The CTA sat below the fold on most mobile viewports and was never reachable at first sight. |
| Analytics baselineGoogle Analytics | Pre-project period, 38,000 monthly visits | Bounce rate 67%, average session duration 41 seconds, conversion to any contact 3.1%. |
The pattern underneath
The barrier to booking wasn’t distrust or disinterest — it was a friction gap between “I have a pest problem” and “I know this company can help me.” 38% of first phone calls could be resolved in under 60 seconds with the right automated assistant. The website was forcing human labour onto questions that didn’t require it.
What this meant for design. If we could close the information gap instantly, without requiring a phone call, we would remove the primary reason visitors left. The chat assistant needed to answer those 12 core questions first, then route qualified prospects to booking.
Problem Definition
Primary user:
Homeowners aged 28–55, mid-to-premium budget. Typically searching for pest
control after discovering a problem — often on mobile, often outside business hours, often in a mild state of urgency. They want to know "can you help me, right now?" before committing to anything.
Job-to-be-done:
"When I discover a pest problem at home, I want to immediately find out if a provider can help me — in my area, for my pest — so I can schedule service without waiting on hold or playing phone tag."
How we defined success (set before designing):
Metric Baseline Target How measured Time-to-first-contact 4 hours (phone queue average) < 2 minutes (AI chat) Chat session logs — time from page load to first qualifying response Self-service booking rate 0% ≥ 25% of new bookings Analytics: bookings completed via online form vs. phone Overall bounce rate 67% < 50% Google Analytics, pre/post Page load time (LCP) 4.3 seconds < 3.0 seconds Lighthouse / WebPageTest Chat engagement rate N/A ≥ 30% of visitors Chat session analytics
Design Process
Directions explored before converging:
1. Chatbot-only model
Replace all CTAs with an AI assistant. Users answer a few questions; the assistant qualifies and routes them.
• Why considered: Direct, zero friction, handles the 12 core questions.
• Why moved on: User interviews showed that pest control customers — especially first-time
buyers — want to see evidence of the company (who are they, are they licensed, what do others say?) before talking to a bot. Going chatbot-only skipped the trust-building phase.
2. Standard contact form upgrade
Redesigned form with better fields, mobile-optimized, prominent placement.
• Why considered: Simple, low engineering effort.
• Why moved on: Hotjar showed users read the form and left anyway. The problem wasn't the
form's usability — it was that users didn't feel ready to commit to a "formal" inquiry before getting basic questions answered.
3. Chat + booking + human escalation — what we built
AI chat handles qualification (service area, pest type, pricing, availability), integrates directly into booking for ready customers, and provides one-click escalation to a human for complex cases.
• Why this won: Matched user behavior — most just needed one quick question answered before booking. The chat closed that gap instantly. Escalation preserved the option for customers who needed more.
The hardest design decision — chat depth vs. simplicity:
We debated between a sophisticated NLU assistant (open text, understands any question) vs. a guided decision tree (12 scenarios, multiple-choice answers).
We chose the guided model for three reasons:
1. Call center data showed 12 scenarios covered 85% of first contacts — we didn't need open-ended NLP for 85% of cases.
2. Guided interaction is more reliable on slow mobile connections (no model latency, no misparse).
3. Faster to ship and test — we could validate assumptions before investing in NLU
The tradeoff: some power users found the guided flow limiting. We monitored escalation-to-human rates as a proxy — it stabilized at 18%, which we considered acceptable.
Key design decisions & tradeoffs:
What we prioritized What we sacrificed Why this was the right call Chat visible on load, no button required Minimalist aesthetic Hotjar showed users who didn’t see the chat in the first 5 seconds rarely scrolled to find it. Visible-on-load increased engagement from 11% to 43%. Mobile-first design Desktop-specific optimizations 63% of traffic was mobile with a 78% bounce rate — this was the highest-leverage problem. Guided 12-scenario flow Open-ended NLU Call centre logs confirmed 12 scenarios covered 85% of all first contacts; faster to build, easier to test. Trust signals above the fold (reviews, response time, licence badge) Additional service detail above the fold Users who saw trust signals had 3.4× higher chat engagement. The service description moved below the fold to prioritize social proof.
Iteration based on testing:
After building the first prototype, we ran a moderated usability test with 6 homeowners.
• Finding 1: Users didn't understand "Book a Service" as the primary CTA — it felt presumptuous
before they'd confirmed availability. We renamed it to "Check Availability" and saw immediate
improvement in click-through.
• Finding 2: The booking form originally required 8 fields. 4 of 6 users dropped off at "property size." We simplified to 3 required fields (address, pest type, preferred date) with optional fields at step 2. Completion rate in follow-up testing went from 29% → 61%.
• Finding 3: The chat showed "GecoPlax Bot" as the sender name. Users were skeptical — 3 of 6
said it felt "scammy." We renamed it "GecoPlax Assistant" with a branded avatar and removed the word "bot." Trust rating improved in exit interviews.
The Solution
A conversion-optimized website with three integrated components:
1. AI Chat Assistant
Persistent, branded chat available on every page. Opens by default on the homepage with the
message: "Hi! I'm the GecoPlax Assistant. What pest are you dealing with today?" Routes through 12 pest scenarios with decision-tree logic. Answers immediate questions (service area, pricing, emergency availability) and transitions qualified prospects directly into the booking flow.
2. Online Booking System
Embedded directly into the chat flow (recommended path) or accessible from a standalone CTA. 3-field intake: service type (auto-populated from chat), address, preferred date + time. Shows available technician slots in real-time (Supabase integration). Confirmation sent immediately by email and SMS.
3. Visual Hierarchy Redesign
Restructured information architecture for the prospect journey:
1. Pest identification — above fold, immediately actionable (chat entry point)
2. Trust building — reviews, license badge, "Typically responds in < 2 min" signal
3. Service overview — what we treat, service areas
4. Emergency CTA — persistent sticky header on mobile with "Call Now" + chat buttons always
visible.
Key visual changes: shifted from 6 below-fold content sections to a single above-fold CTA; increased primary button size on mobile by 40%; replaced the generic hero image with a trust-focused layout (5-star badge + customer count + response time).
Accessibility considerations:
• WCAG AA compliant: 4.5:1 minimum contrast on all interactive elements
• Chat keyboard-navigable (full tab + enter support, no mouse required)
• Alt text on all service type icons; chat options available as text (not icon-only)
• Tested with VoiceOver on iOS — all chat steps readable in sequence.
Results & Impact
Measurement approach: Pre/post analysis. Measured 4 weeks before launch vs. 8 weeks
post-launch. No major PR campaigns, pricing changes, or seasonal events during the measurement window. Google Analytics, Hotjar session recording, chat log analytics, and booking platform data were the four primary sources.
Post-launch period: 8 weeks (March–May 2025)
Traffic volume: ~38,000 visitors/month average across measurement period.
| Metric | Before | After | Change |
|---|---|---|---|
| Time-to-first-contact | 4 hours (phone queue average) | 3 minutes (median, including users who used chat and those who called) | −45% |
| Self-service booking rate | 0% | 35% of new bookings | +35pp |
| Overall bounce rate | 67% | 50% | −25% |
| Mobile bounce rate | 78% | 57% | −27% |
| Chat engagement rate | N/A | 43% of homepage visitors | New channel |
| Chat-to-booking conversion | N/A | 28% of chat sessions convert to a booking | New channel |
| Page load time (LCP) | 4.3 seconds | 3.0 seconds | +30% |
Across ~38,000 monthly visitors, 35% of new bookings moving to self-service represents approximately 290 additional bookings per month that no longer require phone staff handling (~3,480/year). At an average service value of $380, this is approximately **$1.3M in annual bookings now processed without manual intervention.
Chat-assisted qualification also improved lead quality: technicians reported fewer "no-show" or scope-mismatch appointments (prospects who didn't understand the service before booking), down from 12% to 6% of scheduled appointments.
Qualitative signal:
Post-launch customer survey (n=84, 6 weeks post-launch):
• 89% said the chat answered their question before they needed to call
• 76% said the online booking was "easier than expected"
• NPS moved from +22 (pre-launch baseline, from annual survey) to +38 at 8-week post-launch
measurement Representative feedback: "I expected to have to call and wait. The chat answered my question about service area immediately and I booked in under 5 minutes. First time I've ever done that online for a home service."
Learnings
What worked:
1. Guided AI over open-ended. The 12-scenario decision tree was the right call. It shipped faster, performed more reliably on slow mobile connections, and users preferred the structured flow (confirmed in exit surveys: 81% rated the chat experience as "easy" or "very easy"). In retrospect, NLU would have added 3+ weeks of development and introduced edge-case failure modes during a critical launch window.
2. Visible trust signals above the fold. Moving reviews, license badge, and response time to
above-fold (instead of the service description copy) was counterintuitive but decisive. Users who saw those signals had 3.4x higher chat engagement. The copy could explain services; the trust signals removed hesitation.
3. Simplified booking form. Reducing from 8 fields to 3 required fields with progressive disclosure nearly doubled completion. This pattern — start minimal, ask for more only if needed — is now the default approach for any intake form I design.
What I'd do differently:
1. Instrument chat analytics pre-launch. We set up step-by-step chat funnel tracking after launch, not before. The first two weeks of data were incomplete, which delayed iteration on drop-off points. I'd bake analytics instrumentation into the design handoff spec going forward.
2. Test the chat avatar earlier. The "GecoPlax Bot" naming issue came up in usability testing and required a last-minute change before launch. A simple naming test in the discovery phase would have caught this weeks earlier.
3. Validate the 12 scenario list with operations, not just call logs. Two of the 12 scenarios we built were rarely used in practice (termite warranties and commercial accounts) — we learned this from technicians post-launch. An operational interview in week 1 would have resized the scope and freed up development time.
What this opened up:
• Seasonal targeting: Chat data revealed clear seasonal patterns (ant/termite inquiries spike in
spring, mosquito/wasp in summer). Next iteration: dynamic homepage messaging and chat
opening prompts based on season and location.
• Lead scoring pipeline: All chat sessions now feed into a CRM, enabling prioritization of
high-value prospects (large properties, emergency service requests, recurring plans). Currently
being scoped for Q3 2025.
• Technician demand forecasting: Online booking creates a 3–7 day advance demand signal that wasn't available when all bookings came through phone calls same-day. Scheduling efficiency has already improved — overtime reduced by approximately 15% in the first 6 weeks.
Highlights
- Visual hierarchy redesign
- AI-powered contact assistant
- Online booking system
- Mobile-first responsive layout
- Conversion-focused CTA placement
- Trust-building content restructuring
Capabilities
UX Research, Interaction Design, Conversion Strategy, AI-driven Design, Visual Redesign, Service Experience Optimization, Vibe Coding
Tools
Figma, Hotjar, Google Analytics, React, Lovable, Supabase, AI Chat Integration.