AI AGENT / GORGIAS • 2025
Piloting a Conversational Help Center
ROLE
Product Designer & Researcher
TIMELINE
January - March 2025
TEAM
Vina R., Product Manager
3 Engineers
Gorgias is a customer support platform that helps ecommerce merchants automate conversations at scale.
A single inbox houses inquiries across chat, email (and more), allowing human and AI agents to deliver high-quality customer service.
For shoppers who prefer to help themselves, they head to the Help Center before live support.
A curated hub of articles, FAQs, and self-service flows hosted in Gorgias empowers shoppers to find answers on demand.
THE ZERO-EFFORT PARADIGM
But the way people expect to find answers is fundamentally changing.
The same impatience that made two-day shipping feel slow now applies to finding help: shoppers no longer tolerate waiting or working for answers, especially when modern tools have proven that answers should require little to no effort.
THE PROBLEM
Help Centers ask shoppers to adapt to the system, not the other way around.
We validated this assumption through shopper interviews, dog-fooding merchant Help Centers, and supporting behavioral data.
Dead ends force them to find alternative solutions
When articles don’t solve their problem, there are no related suggestions or guided next steps.
Frustated, they escalate to another channel with all context lost.
OPPORTUNITY
So how can we turn fragmented self-serve experiences into seamless paths to resolution?
To solve this, we had to balance merchant needs (more automation) with shopper expectations (faster resolution), while aligning with Gorgias’s strategic vision to leverage AI as a revenue driver.
THE SOLUTION
The Vision: A mobile-first Help Center driven by AI-powered conversation
The Help Center shifts from a static document repository into a real-time support layer that adapts to intent and context.
Ask instead of search
One entry point. No clutter. Shoppers can ask questions in natural language and get immediate, personalized answers.
Proactive guidance
AI Agent proactively pulls order details, gathers relevant information, and suggests next steps — all before shoppers ask.
Full resolution in one place
Instead of directing shoppers elsewhere, AI Agent processes replacements, returns, and refunds directly within the conversation.
Seamless handoff with context
When AI can’t resolve an issue, a human agent can step in with full context, no repeating needed.
PILOT
Proof-of-concept: Testing the hypothesis
We conducted an A/B test against the current experience to see whether shoppers engage with conversation, find answers faster, and feel comfortable asking instead of browsing.
Optional article browsing
Though not part of the vision, we deliberately included browsing to test whether shoppers still default to traditional navigation.
Leveraging existing Chat UI
Reusing production-ready components helped us meet the 2-week build timeline.
Redirecting Flows to conversation
Testing whether conversation could eventually replace the branched, deterministic nature of Flows.
Qualitative feedback
A small-effort idea to capture open-ended insights, but it generated minimal engagement.
OUTCOME
Results & Impact
Automation increased and resolution time dropped, while shoppers began naturally prompting first and browsing as backup — a clear shift away from simple keyword searches toward longer, more natural language.
The project influenced the Q2+ roadmap (MVP, AI features from the vision, chat redesign), established a foundational vision for conversational patterns across the product, and spotlighted design as a strategic driver.
MARKET RESEARCH
I started by looking at products solving similar problems differently.
Competitors offered little differentiation from what we were doing, so I expanded research beyond support, exploring how other tools guide users from question to answer with less friction.
EXPLORATIONS
It influenced numerous directions I explored (here's a few):
The goal was to find a balance between familiarity, discoverability, and impact on the shopping experience.
AI mode in search
Blend article results with AI responses, adding dynamic follow-ups for shoppers who need more.
Embed Help Center in Chat
Surface answers through a dedicated tab in chat, keeping shoppers in their browsing flow.
Recovery for failed articles
Catch shoppers at frustration points when they react negatively to an article or linger too long.
…but they felt like reactive solutions on top of what already existed*
*and leadership agreed.
THE BOLD BET
What if we stripped the Help Center and offered a single conversational entry point?
AI Agent was automating up to 60% of conversations across other channels, drawing from Help Center articles, brand guidance, policies, and internal playbooks. So why not leverage it here?
But constraints and risks emerged quickly:
PIVOTING
Diverging to take a vision-first approach
Rather than designing within existing constraints, I defined the ideal shopper experience first: how it should feel, what shoppers needed, and which familiar patterns could scale while aligning with mental models.
Key principles of effective conversational AI that guided the vision:
Efficiency
Quick answers with little effort
Clarity
Clear, easy-to-underestand information
Trust
Accurate, verifiable, and consistent guidance
Relevance
Personalized interactions at all times
Built-in trust and control mechanisms
Smart prompts
Guide next steps without requiring effort
Source transparency
Show where answers come from
Validation before action
Ask permission before executing changes
PLATFORM THINKING
Laying the foundation for conversational commerce
In collaboration with a senior designer, we aligned on an opportunity to reframe the entire shopper journey around conversation — from discovery to support. The pattern became foundational, extending into areas like Voice of Customer and a modernized Chat.
Mockup: Ivana Tso (Design lead on VOC)
CONSTRAINING THE EXPERIENCE
The pilot: Balancing fidelity with constraints
With 2 weeks to build, we designed a scrappy proof-of-concept to validate core assumptions.
Engagement
Would shoppers embrace conversational input or retreat to the safety of traditional article browsing?
Efficiency
Would AI-powered conversation resolve issues faster than the existing search-and-read model?
TRADEOFFS
What to include, leverage, or save for later
I prioritized decisions based on effort (could we reuse existing components?) and learning value (was it essential for validation?).
Validating technical feasability
Can AI Agent handle common Help Center queries?
We dogfooded top search queries into merchants' chats. Result: AI followed up for clarification or gave helpful answers even for short queries and zero-result searches.
How do we reset AI Agent between sessions?
We rebuilt a re-trigger mechanism to fix an existing 72-hour session bug that left returning users with blank conversations.
LEARNINGS
Constraints sharpen decisions
Working within system constraints for the pilot forced me to make intentional choices about what to test. Shipping scrappy let us learn faster without sacrificing the core experience.
Zoom out to zoom in
Stepping back to question the initial solution helped me define a clearer direction and revealed opportunities for cross-collaboration, giving the project impact well beyond its original scope.