Workspace Onboarding Assistant
Five scenarios that turn an open-ended AI chat into a guided setup for new photographers, from first message to booking link.
The interaction model, before and after
Overview
VSCO Workspace is where photographers run the business side of their work — clients, jobs, quotes, contracts, invoices, and booking links in one place.
In June 2026, we built an AI assistant that sets up a new photographer’s business in conversation, from business details to a booking link they can send a real client. I designed the conversation design system it runs on — the scenarios and components — starting with onboarding.
The problem
Workspace is powerful but hard to navigate. New photographers took 33 days on average to book their first client — the point where it clicks — and only 0.5% finished the onboarding checklist. The checklist told photographers what to set up, but they still had to find each page and fill in every field themselves.
A simpler Workspace is a long rebuild away. In the meantime, we turned to an AI assistant that does the setup with them.
But a checklist and a chat contradict each other. One is fixed steps in order; the other lets photographers ask anything. We needed both.
Decision 1 — Five scenarios the assistant has to handle
A checklist and a chat pull against each other. To hold both, I worked out every situation the assistant has to handle during onboarding. There were five.
For each one I asked what the photographer needs and what they see when it works or fails. The answers gave me the components, and the same small set carries all five.
| Scenario | What the photographer needs | Components |
|---|---|---|
| Running the six setup steps | Set up my business without learning where everything lives | loading state, confirm before create, checkpoint card, link to the real page |
| Asking for something it can’t do | Tell me what’s possible so I stop guessing | text reply, quick replies for what it can do |
| A person needs to take over | Get me to someone without losing my place | support button, setup pauses, progress saved |
| Setup is finished | Tell me I’m done and what’s worth doing next | completion card, links to the next areas |
| Left partway and came back | Pick up where I stopped | nudge in Workspace, resume, chat history |
Every component traces back to a scenario, and no scenario is left unhandled. That’s what makes it a system instead of a set of screens, and it’s why the same set still works outside onboarding.
The handoff needed its own rule. Support is a real person, so the assistant stops rather than guesses: setup pauses and progress stays saved until the photographer comes back.
I helped write the rules the assistant follows: confirm before creating a record, and never mark a step done until the system confirms it.
Decision 2 — Moving progress into the conversation
Chat is open-ended, but onboarding has an end. Photographers need to see how far along they are, and they need a way back in if they stop halfway.
The team had scoped a progress bar pinned above the chat. In internal testing people missed it — they were reading the thread, and the bar sat outside it. So I built both and showed PM and Eng the choice: a bar you have to look up at, or a card that arrives in the conversation as each step closes. We built the card.
The two options I showed the team
Progress now sits where the work happens: a card at each step, a nudge in Workspace if someone leaves partway, and a completion card at the end.
Decision 3 — Show the real page while the assistant works
The assistant does the setup for photographers, which is the point. But a new photographer can finish setup with a job ready and still not know where that job lives in Workspace, or how to reach the features the assistant doesn’t cover yet.
So when a step matched a real Workspace setting, I opened that page with the chat beside it. Photographers see what was saved and where to change it later.
This costs a beat of attention in a flow I had just promised would be effortless. I took the trade, because what a photographer learns about Workspace outlasts the setup itself.
When setup is done, the completion card points to what they can set up next.
What we shipped
A setup flow that creates real Workspace records — business profile, product, pricing, quote, client, job, booking link — one step at a time in conversation.
- Five scenarios running on one shared set of components
- Confirmation before every record it creates, in plain language, and no step marked done until the system confirms it
- Progress in the thread: a checkpoint card at each step, a resume nudge, and a completion card
- A handoff to a support person that pauses setup until the photographer comes back
- In-chat feedback, so the team can see session by session whether it worked
Outcomes
Flagged for internal testing in mid-June 2026, then launched to paying photographers on June 23. The first AI assistant in Workspace.
The component library is the part that lasted. It carried into Workspace Atlas, the simplified system we’re rebuilding toward, and the capabilities in progress now — creating invoices and contracts — are being built from the same components.
Workspace-wide, signup → activation sits at 4.5–5.7%. Most of those signups came through the free tier that launched the same day, where the assistant didn’t run, so I treat that number as context rather than the assistant’s result.
Reflection
Designing a chat is harder than designing a screen. A screen only has to handle the paths I drew. A chat has to handle whatever a photographer types, so most of the work sat in the edge cases: a request the assistant can’t do, a model that gets it wrong, someone leaving mid-step.
What made it manageable was the five scenarios. They turned an open-ended chat into a finite set of cases, and the component library came out of that set. Everything we add to the assistant now reuses it.
The numbers we have can’t tell whether it worked. What I’d measure: photographers who used the assistant against those who didn’t, both reaching a first completed flow, plus drop-off at each setup step and how many come back and resume.