Pipedrive voice AI integration: calls and follow-ups

Start with one callback workflow

For a small sales team, a useful pilot is following up with leads who requested a call. The agent confirms their requirement and preferred callback time. A salesperson receives a follow-up activity with the original CRM record attached. Keep deal-stage changes and sales commitments under human review until your results justify automation.

Eligible Pipedrive record → ConversAI call → result webhook → saved CRM activity → assigned salesperson

What you need before building

• A Pipedrive account with permission to access the relevant records and create activities. Choose the appropriate Pipedrive API authentication method and keep credentials on your server.
• A ConversAI Labs agent and API key. Follow the quickstart and API reference for supported call operations and webhook configuration.
• A receiver and durable store that map each call ID to its CRM record, track processing state and recover failed CRM writes.
• A defined contact-permission policy, calling window, opt-out rule and owner for unresolved outcomes.

Build the call-to-activity flow

1. Select the record. Begin with an explicit callback request. Store the Pipedrive person and lead or deal identifiers, owner and approved contact details. Send the agent only the context needed for that call.
2. Start and correlate the call. Use the documented ConversAI call operation and save its call ID alongside your CRM identifiers. Do not identify a deal from a phone number alone: one person may have several deals.
3. Process the outcome. The documented active events are call.analysed and call.failed. Use event type plus call_id for duplicate detection, and separate a completed conversation from a verified business outcome.
4. Create the CRM activity. Pipedrive supports POST /api/v2/activities. Set the subject, activity type, owner and the relevant lead or deal reference. Set the primary person through participants. Check the current Activities API field definitions before implementing.
5. Confirm the write. Save the returned activity ID. Keep an agreed future callback open; only mark an activity complete when that activity has actually happened. Route unclear outcomes or failed writes to a person.

Decide what each call result should do

These are proposed business rules for your integration. Confirm available result fields and configure any required CRM fields before rollout.

Test recovery as well as the happy path

Test duplicate webhooks, missing CRM records, invalid credentials, rate limits, opt-outs and ambiguous callback times. Persist receipt before acknowledging a webhook. Retry CRM failures independently of making a new call. If a write times out, reconcile whether an activity was created before retrying, so one call does not create two tasks.

Track calls attempted, calls connected, outcomes reviewed and activities successfully saved as separate counts. An answered call alone does not prove that the CRM workflow succeeded.