💡Thumbot: Virtual Assistant for Thumbtack
Team: Jinbo Shen, Sisira Mondreti, Uraiba Zafar
Tools: Figma, Voiceflow, Notion
Role: Conversational UX Design, Intent Modeling, Flow Prototyping
Timeline: Fall 2024
🧭 Project Overview
Thumbtack connects users with local service providers. However, users often face decision fatigue while browsing listings. We proposed Thumbot, a virtual assistant embedded into the Thumbtack platform to simplify key actions: booking appointments, asking questions, and leaving reviews. The goal was to increase efficiency and satisfaction across common customer journeys.
🧠 Assistant Personality
- Clear: Maintains clarity and consistency to reduce miscommunication
- Kind: Patient and emotionally aware to prevent user frustration
- Simple: Minimizes cognitive load with accessible language
- Informative: Reduces need for human support by anticipating user needs
🎯 Main Intents & Sub-Tasks
Thumbot was designed with three primary intents, each broken down into meaningful, user-centered sub-flows:
1. Book an Appointment
- Sub-tasks: Handyman, Plumbing, Electrical, HVAC, Roofing
- Logic: Service ➝ Provider ➝ Date/Time ➝ Confirmation
2. FAQs / Information Assistant
- Topics: Services, Remodeling, Events, Weddings, Home Maintenance
- Logic: Topic selection ➝ Response ➝ Follow-up options
3. Leave a Review
- Sub-tasks: Star Rating, Written Feedback
- Dynamic response logic based on sentiment (e.g., 1-star triggers apology + comment option)
🧪 Conversation Design & CUI Logic
- Mobile UI: Non-intrusive chat bubble at bottom-right corner
- Desktop UI: Larger message pane while preserving content visibility
- Multi-turn flows with fallback and clarifying questions
- Includes error handling for out-of-scope queries (e.g., haircut services)
📌 Sample Flows
🛠️ Intent 1: Book an Appointment
User: "I need someone to fix my faucet"
Agent: "Sure! Let me find you a top-rated plumber near you. Here are the top 3..."
➝ User selects one ➝ Enters preferred time ➝ Thumbot confirms booking
ℹ️ Intent 2: FAQs
User: "Do you help with wedding planning?"
Agent: "Yes! Thumbtack connects you with planners, caterers, and decorators. Would you like to see a list?"
⭐ Intent 3: Leave a Review
User: "I'd like to leave a 1-star review"
Agent: "We're sorry to hear that. Can you tell us what happened?"
➝ Option to escalate to customer service if needed
🧪 User Testing & Iterations
| User | Challenge | Change Implemented |
|---|---|---|
| User A | Agent misinterpreted "plumber" in a review as a booking intent | Disabled “listen for intent” during review feedback stage |
| User B | Overwhelmed by excessive text in info responses | Condensed responses and added hyperlinks to Thumbtack’s pages |
| User C | 1-star reviews lacked emotional recognition | Integrated empathetic prompts and escalations for negative feedback |
| User D | “Information Assistant” label felt unclear | Renamed to “FAQs” for clarity |
🔮 Reflection
This project strengthened my expertise in conversation design, sub-intent mapping, and flow prioritization. More importantly, it highlighted the value of emotional nuance and multi-modality in user assistance. Thumbot now stands as a scalable, smart, and kind virtual assistant that makes service discovery effortless.
