Reduced support calls by 10% with a guided chatbot
SnappTrip
Overview
SnappTrip is a full-service online travel booking platform that processes millions of tickets each month. By 2026, rising support costs had become a critical operational concern.
Main Problem: SnappTrip had a high volume of repetitive customer support requests, but users struggled to find answers within the app due to poor discoverability and multi-step navigation, leading them to rely on call center agents. This caused high operational costs, long wait times, low customer satisfaction, and overloaded support teams, prompting the need for an chatbot to automate common queries.
This resulted in:
- High support costs
- Long waiting times
- Low customer satisfaction
- Heavy workload for customer support agents
Role
User research
User experience design
User interface design
Tools
Figma, FigJam, Claude, ChatGPT
Team
- 1 Product Designer (me)
- 1 Product Manager
- Engineering (4×), QA (2×)
Business requirements
- Reduce support calls per ticket by 25%
- Achieve 60% self-service resolution
- Achieve CSAT above 3/5
- Keep human escalation below 25%
Research insights
Users struggled to:
- Find cancellation option
- Understand refund rules
- Track reservation status
- Know whether their ticket could still be cancelled
- Understand cancellation penalties
- Wait for human agents for simple questions
1. Users wanted the fastest path, not more information.
Most participants already knew they wanted to cancel or check their reservation—they simply couldn’t find where to do it.
“I know what I need. I just don’t know where to start.”
Design Opportunity
-
Surface the most common intents immediately.
-
Reduce navigation before users can take action
2. Users Needed Clarity Before Cancelling.
Before confirming a cancellation, users wanted to understand:
- How much money they would receive.
- When they would receive it.
- Whether any penalties applied.
Many contacted support simply for reassurance.
Design Opportunity
-
Display refund estimates before confirmation.
-
Explain cancellation policies in plain language.
-
Set clear expectations for refund timing.
3. Reservation status created unnecessary support contacts.
Many users contacted support even though their reservation was still processing normally.
The biggest uncertainty was:
- Is my reservation confirmed?
- Should I wait?
- Is something wrong?
Design Opportunity
Provide proactive status updates with contextual explanations and recommended next steps.
4. Users preferred guided conversations over searching.
Participants found it easier to answer a few simple questions than browse multiple Help Center articles.
Rather than searching, users expected the product to guide them.
Design Opportunity
Design a conversational flow that asks one question at a time and progressively narrows down the user’s intent.
Competitors approach
I used AI-powered deep research to examine leading travel companies’ AI and chatbot experiences, distilling the best design patterns and identifying key gaps and opportunities.
I collected screenshots from my benchmark research for design inspiration and added them to Figma as a reference board.
- The strongest experiences combine chat with structured UI — conversational input works for simple refinements, while cards, filters, calendars and selectors handle complex decisions.
- Progressive disclosure reduces cognitive load — assistants collect missing information step-by-step instead of asking users to complete a long form upfront.
- Quick replies and suggested prompts improve discoverability — chips and starter prompts help users understand what the assistant can do and reduce blank-page friction.
- The market is moving toward agentic booking — while many products still hand users off to traditional booking flows, MakeMyTrip demonstrates a fully conversational search-to-payment experience.
- Most chatbots aim to delay human support handoff — they try to resolve the user’s issue within the chatbot experience before escalating to a customer support agent.
User flow
The chatbot centralizes the most common railway support requests into a single conversational experience. Users can easily complete tasks such as Fine Inquiry & Cancellation Registration, Change Date & Route, Cancellation Request Tracking, Order Status Tracking, Get Ticket, and Frequently Asked Questions. By understanding user intent and guiding them through the appropriate flow, the chatbot reduces friction, minimizes unnecessary support calls, and enables faster self-service. I included only the cancellation flow in this case study.
Train cancellation flow
First designs
I used AI to create an initial prototype without any design work in Figma and ran an AI-assisted usability test. It revealed two issues: search was needed for users with more than three orders, and the placement of options in the flow needed improvement. I addressed both, then presented the refined flow to the Product and Tech teams before moving to high-fidelity designs.
Train cancellation flow
Usability issues
To validate the guided chatbot experience, I conducted usability testing on an interactive prototype.
The evaluation focused on whether users could complete their tasks without confusion or human assistance.
Chatbot Train Ticket Cancellation Flow
1. Objectives
- Determine whether users can locate and complete the ticket cancellation flow in the chatbot without assistance.
- Assess whether the chatbot’s language, menu structure, and button labels are clear and understandable.
- Identify points where users hesitate, get confused, or make incorrect selections.
- Verify that critical information (refund amount, refund timeline, cancellation fee) is presented clearly and at the right moment.
- Measure user confidence and satisfaction after completing the process.
2. Key Research Questions
- What is the unassisted task completion rate for cancellation?
- What is the average time and number of messages/taps required to complete the flow?
- At which step do users experience the most errors or hesitation?
3. Participants
Recommended sample size: 5–8 participants. Per Nielsen’s research, testing with 5 users typically surfaces around 85% of usability issues. Suggested participant mix:- Users experienced with purchasing train tickets online
- Users with limited experience using apps/chatbots (to test the lower bound of usability)
- Where possible, include existing real users of the product
4. Task Scenarios
Scenarios are framed as short stories rather than direct instructions (e.g., “cancel your ticket”), so participants behave naturally.- Base scenario: “Imagine you bought a Tehran–Mashhad train ticket for tomorrow, but your plans changed. Use the chatbot to cancel your ticket and get your money back.”
- More complex scenario: “You purchased two round-trip tickets, but you only want to cancel the return ticket.”
- Edge-case / error scenario: “The ticket you want to cancel departs in less than 2 hours — see what the chatbot tells you and how you respond.”
- Follow-up scenario: “After cancelling, find out when and how your refund will be issued.”
5. Metrics
| Type | Metric | Description |
| Quantitative | Success Rate | % of participants who complete each cancellation task without assistance |
| Quantitative | Time on Task | Time taken to complete each scenario |
| Quantitative | Error Rate | Number of missteps, wrong taps, or repeated messages per task |
| Quantitative | SUS / SEQ Score | System Usability Scale or Single Ease Question after each task |
| Qualitative | Verbatim Feedback | Direct quotes, confusion moments, tone/reactions |
| Qualitative | Clarity Perception | User’s understanding of refund amount, cancellation fee, and refund timeline |
6. Methodology
- Test type: Moderated (in-person or remote with screen sharing) — chosen because financial flows are sensitive and hearing the user’s think-aloud reasoning is valuable.
- Session length: 20–30 minutes
- Recording: Screen and audio recording, with participant consent
7. Session Structure
| Step | Duration | Notes |
| Welcome & introduction | 2 min | Explain purpose broadly, without revealing exact goals |
| Background questions | 3 min | Prior experience with train ticket purchases / chatbots |
| Task scenarios (think-aloud) | 15 min | Participant narrates thoughts while completing tasks |
| Post-task questionnaire | 3 min | SUS or SEQ after each scenario |
| Closing interview | 5 min | Open-ended follow-up questions |
8. Sample Closing Interview Questions
- Which part felt most confusing?
- How confident did you feel that the cancellation actually went through?
- If you could change one thing about this process, what would it be?
9. Analysis & Deliverables
- Prioritized list of usability issues (Critical / Major / Minor) with a severity matrix
- Clipped video highlights of key moments for the product team
- Final report with concrete, actionable design recommendations
Report
Improved designs
Designs after usability testing
After improving the design, I created an AI-generated prototype and conducted a second usability test with AI using the same personas.
Prototype & animation with AI
In this chatbot flow, users can only cancel one-way or round-trip train tickets. I disabled the other flows to keep the case study focused on this specific journey. I also created the robot icon animation and prototype using AI.
AI Throughout My Design Process
- Research: Benchmarked 6+ chatbots and analyzed some of support conversations.
- Analysis: Identified pain points, UX patterns, and opportunities.
- Ideation: Explored 5+ conversation flows and interaction models.
- Prototyping: Built and iterated interactive prototypes in hours instead of days.
- Motion: Generated chatbot animations and micro-interactions.
AI reduced research and production time by ~50%, freeing up more time for strategic UX decisions, flow validation, and experience refinement.
Result
We launched the chatbot gradually, starting with 20% of users and then expanding to 50%.
-10%
Support calls per ticket
55%
Self-Service resolution
3
CSAT
22%
Human escalation rate
What I Learned
Initially, I designed full flows screen by screen in Figma, but as maintaining detailed flows across so many verticals — flights, bus, train, hotel, tour, and villa — became unsustainable, we shifted to a component-based system, where each flow step simply references the general component that should appear there, making the design easier to maintain and scale.
Leveraging AI throughout this project helped me accelerate the process — from communicating animations and interactions more clearly to the tech team, to quickly categorizing user insights I gathered, to generating a range of design ideas to explore.
Looking ahead, many of the responses to user questions should be dynamically generated using AI rather than fully pre-designed, allowing the system to handle a wider range of user needs without requiring a manually maintained flow for every scenario.