Support Specialist
Conversational Design
Service Design
Information Architecture
We integrated a chat-based support platform and built an AI chatbot that, today, solves 90% of all customer queries at Synthesia. Reducing support costs while improving customer satisfaction and operational efficiency.
As Synthesia scaled, its CRM-based support system struggled, leading to long response times, slow resolutions, and misrouted sales queries. Customers couldn’t easily self-serve, and the support team was overwhelmed with repetitive tickets and sales queries. In April 2022, after months of careful consideration to our request to change platforms, the company approved a migration to Intercom, moving away from CRM-based support into a dedicated, chat-based, support platform, creating an opportunity to design a chat-driven support experience that improved operational efficiency and scalability.
Research
To understand the key challenges, the project leveraged quantitative data from thousands of support tickets, CSAT surveys, and stakeholder interviews across Support, Sales, and RevOps. The findings revealed:
• Users struggled to find basic answers leading to 70% of support queries being repetitive questions, overwhelming agents.
• Agents spent a lot of time categorising tickets and doing data entry, which led to inefficient support and data entry errors.
• Sales inquiries were lost in support queues, slowing down revenue opportunities.
• Maintaining data integrity and transparency was a concern from senior leadership as we scaled.
Design solution
Taking an iterative, user-centred approach, the team integrated, designed and deployed:
• An AI-powered chatbot trained on the knowledge base and our top 50 customer queries, automating common questions.
• Integrated automatic data captures in chat flows, improving data accuracy and freeing agents to spend time with customers.
• Automated query routing, ensuring sales leads were prioritized and escalated correctly.
• A fully in-sync support system, connecting Intercom, CRM and our data warehouse via API for everyone's benefit.
Results
After a progressive rollout, the new support system led to:
• Response times improved from 24 hours → ~1 hour.
• Resolution times dropping from 48 hours → ~4 hours.
• Customer satisfaction increased from 68% → 90%
• Immediately, 60% of support queries started being resolved instantly via chatbot, reducing agent workload.
With a fully scalable, AI-powered support infrastructure, the team rapidly validated the ROI of a dedicated support platform. It reduced workloads for support agents, enabling agents to focus on issues requiring human input, and improving both sales pipelines and overall service quality and efficiency. Not to mention the long term savings in support costs.
+22%
CSAT Improvement
24h -> 1h
Avg. Response time
48h -> 4h
Avg. Resolution time
Intro
As Synthesia scaled rapidly, its existing CRM-based support platform and linear chatbot struggled to keep up. The support team had been advocating for a more robust support and chat solution for months—one that could improve efficiency, speed up response times, and scale with the company’s growth. In April 2022, after months of internal discussions, senior leadership approved the migration to Intercom, a dedicated, chat-based support platform that would also enable us to build and train an AI chatbot.
A linear support chatbot without access to human agents had been driving user crazy for months
Research
This project had a unique advantage—I wasn’t an outsider conducting research from scratch. As the founding support specialist, I had spent months gathering both quantitative and qualitative insights, experiencing pain points first-hand, and working directly with users and internal teams. This allowed me to move quickly while ensuring every decision was rooted in real data.
Before designing the chatbot and workflows, I analysed support tickets, CSAT surveys, and internal feedback to pinpoint the most pressing issues.
Quantitative Analysis
The team and I had been categorizing thousands of support tickets through a ticketing system I had implemented on our previous CRM-based system to find patterns. So it didn't take us long to uncover:
Our top 50 most common customer questions and most visited knowledge base articles.
Ticket resolution times (where delays constantly happened)
Ticket volumes and response times by category, including sales tickets which were often routed to support
Response vs. resolution time gaps (where efficiency could improve)
Key findings:
70%+ of support tickets were repetitive—product questions, account issues, and common troubleshooting.
Average first response time was 24 hours—too slow for a SaaS company.
Few users self-served via the knowledge base beyond top articles—indicating low discoverability.
This data helped us support our hypothesis that a well implemented chat-based platform, paired with a ML-trained chatbot could significantly reduce ticket volume and improve response times.
Qualitative Insights
To complement the hard data, I gathered direct user feedback through multiple sources:
CSAT surveys (collected after support interactions)
Conversations with key customers & enterprise users
Interviews with customer success, moderation and sales teams
Conversations with key stakeholders to migrate support to a different platform
Recurring themes emerged:
Users couldn’t find answers easily, leading to frustration with the original linear chatbot and unnecessary tickets.
Delayed responses made customers frustrated, especially enterprise customers with time-sensitive issues.
The support team felt overwhelmed, handling repetitive questions instead of complex, high-value cases.
Additionally, key stakeholders had raised concerns on data integrity if we were to migrate support to a different platform.
Challenges
This combination of quantitative and qualitative research shaped the support design strategy—the platform and chatbot needed to automate common queries, improve self-service, correctly route queries and allow agents to focus on high-priority cases - all while maintaining data integrity.
From our research, we distilled three key challenges we needed to resolve:
Reduce response & resolution times
First response times averaged 24 hours, and resolutions took around 48 hours—unacceptable for a fast-growing SaaS business.Ensure seamless data integration
Support, CRM, and the data warehouse needed to stay in sync and maintain data integrity as the company scaled.Improve triaging & escalation
Support, sales, and content moderation requests often landed in the wrong queues, slowing down responses and leading to inefficient support and missed revenue opportunities.
Design
Training the chatbot and designing conversation flows
The core of the new system was our AI-powered chatbot—but this wasn’t an out-of-the-box solution. Since this was pre-ChatGPT, Intercom didn’t offer a native LLM chatbot, so we had to train a Machine Learning model manually.
Key steps in chatbot design:
Structuring conversation flows
Used card and tree sorting with the support team to categorize queries into clear categories and paths (e.g., product questions, troubleshooting, account issues, sales inquiries).
AI training using supervised learning
Manually trained the bot on the top 50 most common questions using ticket history data and knowledge base articles.
Seamless knowledge base integration
Ensured relevant help articles were accessible through the knowledge base and integrated into operational conversation flows (where our AI-chatbot wasn't available) before escalating to an agent.
Escalation logic for complex cases
Built fall-back mechanisms so customers could easily reach a human when needed.
Separation between support and sale
Different version of the chatbot were used pre-login and post-login, ensuring the chat catered for different audiences and teams and helping split sales and support queries.
An image of the customer-facing chat interface
💡 Result: Users self-solved over 60% of issues without needing an agent, reducing support volume and freeing up time for complex cases.
Structuring the Agent Experience & Workflows
While automation was key, the agent support experience also needed improvement. The old CRM-based system made it hard for agents to prioritize urgent tickets, leading to delays.
Key improvements for agents:
Conversation workspaces
Built customized inboxes based on query type, urgency, and user profile (e.g., billing issues, technical support, enterprise accounts).
SLA notifications & reminders
Designed real-time SLA alerts to help agents meet response & resolution targets.
Macro templates & quick responses
Created pre-written response templates for repetitive questions, cutting response time drastically.
Automated tagging & prioritization
Ensured tickets were auto-tagged based on the paths a conversations users took, so the right agents could handle the right issues.
💡 Result: Agents became significantly more efficient, and first response time dropped from 24 hours to ~1 hour.
Intelligent Query Routing & Sales Handoff
One of the main business challenges we had at the time was the lack of proper triaging between support and sales.
Previously, users would reach out through support, even for sales inquiries, leading to:
Missed revenue opportunities → Sales leads were stuck in support queues.
Support overload → Agents handled non-support queries, slowing down response times.
Key routing improvements:
Automated lead qualification
Used Intercom conversations and workflows in conjunction with a CRM integration to detect, triage and route escalate queries to the right teams.
CRM-integrated workflows
Ensured lead data was synced with the Sales CRM, so no opportunity was lost.
💡 Result: Sales-qualified leads were automatically routed, increasing efficiency for both teams and improving customer experience.
Data Integration & System Scalability
A key concern from senior leadership was data integrity—they wanted to ensure that Intercom, CRM, and the data warehouse were fully synced.
Key integrations & scalability improvements:
CRM + Intercom data mapping
Worked with RevOps and engineering teams to ensure customer profiles, queries and data remained up to date and in-sync.
Data syncs
Together with the engineering and RevOps teams, we set up a REST API that automatically pushed and pulled new customer data, subscription details, and past interactions between Intercom, the CRM and our Data Warehouse.
Payment platform integration
Enabled agents to see subscription & billing details directly in Intercom, removing unnecessary back-and-forth.
💡 Result: The data concerns were mitigated. And agents now had full context on customer history, product usage, and past support interactions, improving trust, service quality and response speed.
Design Validation: Usability Testing & Iteration
Synthesia’s start-up mindset was to launch fast and iterate later. And while that made sense, I knew even basic usability testing upfront could prevent major rollout issues. So I adhered to my deadlines and without any budget or platform, tested anyway:
What I did to validate the design:
Internal usability testing
Ran quick tests with agents to optimize inbox layout and workflows before launch.
Scenario testing
Simulated real customer conversations to refine chatbot logic.
Global team testing
Discovered that the initial inbox setup didn’t work well for US-EU handovers, so I redesigned workflows to improve transitions.
💡 Result:
Smoother agent experience → Reduced confusion and improved efficiency.
Fewer chatbot misclassifications → Ensured automation worked without frustrating users.
No major post-launch failures → The system was solid from day one.
Our live chatbot with under 1 hour response times
Outcomes
The results were game-changing for the support team and the business:
🚀 Response time: 24h → ~1h
🚀 Resolution time: 48h → 4h
🚀 CSAT score: 68% → 90%
Business Impact:
Proved the ROI of a dedicated support platform, securing leadership buy-in.
Scalable support infrastructure that could grow with the company.
Happier, more efficient agents, reducing burnout and improving retention.
Improved user experience, with faster, more accurate support across channels.
Key Learnings
💡 User insight drives the best solutions – Being deeply embedded in support operations gave me a major advantage in identifying the right problems to solve.
💡 Testing, even on a small scale, makes a difference – Quick usability testing prevented rollout issues and improved agent workflows.
💡 Automation is powerful, but only when designed around real needs – The chatbot and workflows weren’t just about reducing tickets; they enabled customers and agents to work smarter.
