From connected to activated: onboarding for PandaDoc MCP connector
Company
PandaDoc
Year
2026
Area of focus
Growth
AI & Conversational UX
Prototyping
Decision log
But in review we established that the drop-off happens after the connection succeeds: connected, then no first use.
2 / 2 · back
Context. The obvious target was the consent screen, because it was dense and had a lot of friction.
1 / 2 · read on
Options. Polish the consent screens, or design a bridge from “connected” to a first useful action.
Trade-off. A cleaner consent screen would help trust but could not move a metric that is lost later in the flow.
Decision. Keep the consent work small and spend the effort on the moment after connection.
The assistant can answer that from its own knowledge without calling a single PandaDoc tool, so the prompt could look helpful and still leave activation at zero.
2 / 2 · back
Context. Activation is defined as the first successful API request. The first idea was a starter prompt: “Show me what is possible with PandaDoc.”
1 / 2 · read on
Options. A capability tour, a general discovery prompt, or a task prompt that needs the user’s own data.
Trade-off. A task prompt shows one capability rather than all of them.
It forces a real read call, returns the user’s own documents, and points straight at the next action: chasing a signature.
2 / 2 · back
Decision. “@PandaDoc, show me my recent documents and which ones are still waiting on a signature.”
1 / 2 · read on
Context. The engineering spike showed that the access-request screen and the OAuth screen were technically the same step, shown twice.
Options. Keep two screens and give each one job, or merge them.
Trade-off. One screen has to carry identity, workspace, permissions and legal text without becoming a wall of text.
Decision. One screen, in a fixed order: who you are, where access applies, what the assistant can do, then the legal line and Connect.
Context. Users saw six bundled capabilities in system language and had to accept them all or nothing.
Read is required and says “Nothing will be created or changed”. The other two are labelled clearly, and payment actions sit under Send and publish. A user can see the risk of each group before deciding.
2 / 2 · back
Decision. Three groups based on what changes in the account: Read your content, Create and edit content, and Send and publish documents.
1 / 2 · read on
Context. Through regular syncs with engineers I tested every element. The workspace was shown as a selector, but in this flow there was nothing to choose.
A control that does nothing makes people doubt the rest of the screen.
2 / 2 · back
Decision. Show the workspace as fixed context next to the account, with one line of explanation: “Access is scoped to your role and workspace permissions.”
1 / 2 · read on
Legal. Legal required the consent text to sit below the Connect button. I kept that and gave it a clear type hierarchy so it reads as part of the decision, not small print.
ChatGPT. ChatGPT already shows a suggested prompt in its empty state after a connector is installed. Rather than add a screen of our own, we put our prompt there. Using the platform’s own pattern cost less than inventing one.
Reflection
A metric’s definition is a design input
The most useful change here wasn’t visual. It was noticing that a friendly prompt could avoid the event we were measuring.
Small engineering spikes settle design debates
Merging the two screens came from a technical finding, not from arguing over a mock-up.
Next question: graduated consent
Starting with read-only access and asking for write access when it’s first needed might lower the risk at the very first moment. I’d test it next.
