Robin AI
AI Workflow Automation · Event Ops
Overview
I designed Robin, an operational AI agent embedded in Sequel.io that turns a rough marketing brief into a fully configured webinar campaign, and after the event, surfaces enriched analytics that tell sales who to call first.
Role
Product Designer, AI experience and interaction architecture
Domain
B2B SaaS, AI workflow automation, enterprise event operations
Primary users
Demand-gen marketers, event operations leads, and revenue teams
Hero image: the Robin workspace in context, the conversational panel turning a plain-language prompt into a structured event setup. One frame that captures messy input to structured output.
The problem
Webinar operations are spread across too many tools. Setting up a single recurring series means jumping between separate products for content, landing pages, email layouts, tracking links, embed codes, and CRM segmentation. And once the event ends, marketers have to dig through raw analytics by hand to figure out which attendees are actually sales-ready.
That sprawl broke down on two fronts:
For users
Every setup carried the risk of a small misconfiguration becoming a live-day failure: a broken calendar link, a registration form that does not fire. After the event, finding the best leads meant days of exporting and cross-referencing CSVs.
For the business
That breadth made onboarding hard and leaned heavily on Customer Success to validate routine setups, which capped how fast the business could scale.
The design pivot
Avoid the generic floating chatbot. Robin had to be a system-aware agent that reads, builds, and modifies native Sequel objects, from pre-event setup to post-event lead enrichment, always under the user’s control.
Image: the old workflow, mapped. The tool-sprawl reality, the separate products a marketer juggled for one event, with the handoff points where things broke.
Process
Unstructured input → Context extraction → Structured card prototyping → Launch QA guardrails → Post-event revenue activation
Image: process diagram. The five stages as a clean flow, ideally with a real screen under each.
The system ecosystem audit
I audited the product objects Robin would have to touch, events, series, registration forms, embed widgets, emails, analytics logs, and CRM mappings, and mapped how the agent reads and writes each one transparently.
Image: object map. The Sequel objects Robin operates on and the relationships between them. Evidence that the agent was designed against a real system model, not bolted on.
Interaction strategy: a hybrid UI
Pure chat windows make poor configuration spaces: they have no layout hierarchy, and complex B2B settings get lost in a scroll of messages. So I built a hybrid model: users express intent in conversation, and Robin responds with interactive configuration cards and post-event insight panels, editable tables, toggles, and clearly laid-out options the user can adjust inline.
Image: the hybrid model. Conversation on one side and an interactive configuration card on the other, ideally mid-edit so the talk-then-adjust loop is visible.
My role
As the product designer for Robin, I owned the interface strategy, component model, and workflow architecture that embedded the agent into Sequel. Working with a lead PM, an AI and ML engineering team, and a CS lead, my contributions were:
The hybrid interaction framework
A conversational side panel that acts directly on live configuration blocks in the main canvas, rather than a chat that just returns text.
The modular configuration-card system
Reusable components for settings fields, toggles, and data tables that let the agent present complex setups cleanly and let users edit them in place.
The multi-object generation sequence
Defining how Robin turns one unstructured intent into coordinated updates across events, registration links, and email flows.
The post-event intelligence dashboard
Translating raw engagement logs into ranked, enriched lead profiles and top-account cards a sales team can act on immediately.
Cross-functional alignment
I ran the technical mapping sessions with AI and ML engineering to establish prompt and safety rules before frontend work began.
Key decisions
Hybrid interaction model
Pairs free-text intent with interactive data cards users can edit inline.
Outcome: avoids conversational dead-ends, and complex event properties stay fast to change.
Persistent quick-start presets
Action shortcuts on Robin’s empty state for high-frequency jobs like Check Setup and Pull Lead Scores.
Outcome: lowers the discovery barrier and points users at high-value automations on day one.
Review before execution state
A mandatory preview and explicit approval before Robin touches anything live.
Outcome: builds trust and contains the risk of an AI error reaching a live campaign.
Built-in launch QA
A diagnostic check inside Robin that audits email triggers, registration flows, and CRM setup before launch.
Outcome: catches the errors that would otherwise become launch-day support tickets.
Enriched lead-priority cards
Turns raw CSV exports into cards with engagement scores, titles, and firmographics.
Outcome: closes the marketing-to-sales gap and cuts the time to route high-intent leads.
Image: one decision, shown. The Review before execution state is a strong candidate, with the actual UI for it.
Experience & craft
The Robin workspace hub
A single workspace where conversation and configuration sit side by side: the user types naturally on one side and watches the changes land in an interactive panel on the other.
Image: the hub. The full workspace, annotated so a reviewer who has never seen Sequel understands what each region does.
The automated event-series builder
Asked to plan an educational campaign, Robin drafts the full structure, naming cadence, schedule, and brand guidance, and presents every component in an editable form table the user can adjust before anything is generated.
Image: series builder. The editable form tables mid-setup, showing that the user reviews and adjusts before generation.
Post-event lead and account prioritization
After the event, Robin works on the data, not just the content. It reads attendance time, poll responses, and Q&A activity to surface the top accounts, displays them as enriched profiles with CRM context, and lets marketers route qualified leads to sales with their full history attached.
Image: the prioritization dashboard. The ranked top-accounts view with an enriched profile card expanded. This is the half of the product that touches revenue, give it a strong, real screen.
Edge cases and guardrail design
The hardest design lived in the moments most demos skip:
Latency that does not feel like dead time
Complex multi-step configs take the model 5 to 12 seconds to generate, and a blank spinner makes users abandon. I designed a streaming progress UI that names each step as it happens, Drafting copy, Setting calendar hooks, Building tracking URLs, so the wait reads as work, not a freeze.
No accidental destructive edits
Because Robin can change live event spaces, an unverified generation could overwrite a real tracking link. The review-before-action pattern stages every proposed change, highlights it in amber, and requires an explicit approval click before anything touches the live workspace.
Large post-event datasets
Big analytics sets can blow past the model’s context limits. I designed a paginated summary card that breaks large data into smaller segments, so users can work through the analytics without the system hitting a wall.
[Optional state detail: one line each on the empty state, an error mid-generation, and how the amber-staging diff is visually distinguished would add UI-craft depth here.]
Image or short GIF: the streaming progress UI in motion and the amber review-before-action state. Both are far more convincing as motion or a clear before and after than a single static frame.
Validation
I validated Robin’s workflow with scenario-based testing alongside enterprise event managers.
The task
Take an abstract email brief, turn it into a three-part webinar series with unique tracking parameters, verify the registration layout, then read the enriched account cards to judge sales-readiness.
The key insight
Users got anxious when Robin generated several email steps automatically. They wanted to read the exact copy before saving. That drove an Expand preview accordion inside the configuration cards, reinforcing user oversight at the moment they needed it.
Image: before and after of the Expand-preview change. Left, the collapsed card that caused the anxiety. Right, the expanded preview.
[Optional second testing beat: one insight reads as a lucky catch, two read as a process. If another moment surfaced, something users fumbled or a flow you reordered, add it in the same task, observation, change, result shape.]
Impact
Honest framing: these are the things Robin was designed to move, written as intentions rather than measured results. The real numbers go in the brackets.
Setup speed
Compress the path from brief to live event.
[Measured change?]
Support load
Deflect routine configuration questions away from Customer Success.
[Ticket data?]
Pipeline routing
Give revenue teams a clear, ranked list so hot accounts reach sales within SLA.
[Adoption or time-to-route?]
Sales velocity
Surface enriched top accounts the moment the webinar ends, rather than days later.
[Internal metrics placeholder: measured reductions in setup time, CS ticket volume, and downstream feature adoption. Drop your real numbers in here; even one directional figure beats none.]
Image: closing shot. A clean final hero or a simple metrics card once you have real numbers to put on it.
Reflections
Orchestration beats generation
In high-complexity B2B work, useful AI design is not about producing blocks of text. It is about orchestration: giving users the confidence to run multi-step technical actions safely, and surfacing insight they can act on.
Next, for V4
A Live Producer Copilot inside the live webinar room, helping organizers monitor incoming questions, schedule polls, and read audience trends in real time.
[Optional: a line on something you would do differently, or a constraint you would revisit, rounds this out and reads as senior maturity.]
Let's talk
Pick what works best for you: a quick video call or a simple email.
