Turn Otter.ai customer calls into a Notion product feedback board

Build an agent workflow that turns finished Otter.ai calls into product feedback entries in Notion, deduplicated by theme.

Trigger: an Otter.ai workspace webhook on the conversation.completed event. The agent should run once per finished conversation and receive the conversation id, title, host, participants, and summary from the webhook payload. If the agent needs more context (full transcript or metadata), it can call Otter's List Workspace Conversations to pull recent items or read details from the completed conversation.

Step 1: classify the call. The agent decides whether this is customer-facing product signal. Customer-facing means discovery, onboarding, support, or CS check-in calls with an external customer or prospect who is actively discussing our product. Explicitly skip: internal team syncs, hiring interviews, vendor pitches, and calls that never mention product feedback. If the call is not customer-facing, stop and do nothing.

Step 2: extract product signal. From the transcript and summary, pull out feature requests, pain points, workarounds, and competitor mentions. For each item, capture a short verbatim quote, a one-line paraphrase, and a theme (e.g. pricing, onboarding, reporting, mobile, integrations, reliability). Also infer sentiment (positive, neutral, negative, blocking) and note the customer name and call date.

Step 3: dedupe by theme against Notion. The user will point the agent at a Notion feedback database (data source id). For every extracted item, call Notion's Query a Data Source to look for an existing page on the same theme (match on the Theme property and, when helpful, on the paraphrased description). Pick the best match; if nothing matches, treat it as new.

Step 4a: if a matching theme page exists, use Append Block Children on that page to add a new bulleted list item containing: the verbatim quote, the customer name, the call date, and a link back to the Otter conversation. Also add a short note about sentiment. Do not create a new page.

Step 4b: if no matching theme page exists, use Create a Page as a child of the feedback data source. Set properties for Theme, Customer, Sentiment, Call date, and Source link (link back to the Otter conversation). In the page body, add the verbatim quote, the paraphrased summary, and a bullet list started with the first evidence entry so future calls can append to it in the same shape as step 4a.

Behavior rules the agent should follow: never file the same call's evidence twice on the same page. If the customer or call date is missing, use best-effort values from the Otter conversation metadata. Prefer appending to an existing theme page over creating a new one when the match is reasonable. Keep verbatim quotes short (one or two sentences). Always include the Otter conversation link so anyone reading the board can jump back to context.

Setup inputs the workflow should ask the user for on first run: the Notion feedback database (data source id), the exact property names for Theme, Customer, Sentiment, Call date, and Source link, and any custom rules for what counts as customer-facing (for example, include partner calls, exclude a specific channel).