# Stop re-typing call notes: log every sales call to your CRM automatically Source: https://getstartedwithai.net/tutorials/call-transcript-to-crm Site: Get Started With AI Tagline: Build AI automations that actually run your business. Audience: solo professionals, freelancers, consultants and AI implementers in training Contact: hello@getstartedwithai.co.uk This Markdown companion is generated from the prerendered page for AI crawlers. # Stop re-typing call notes: log every sales call to your CRM automatically Sales teams lose 20 to 40 minutes per call to manual note-taking, re-listening and CRM entry. This build removes that entirely. You record the call, an AI transcription tool converts it to text, an extraction prompt pulls out summary, key points and next actions, and a Zapier or Make automation pushes a structured note into your CRM contact record. From the moment the call ends to the CRM being updated: about 4 minutes, with no one lifting a finger. The build works for any sales or service team. The tool choice depends on what meeting platform you use. ## Prep before you build (5 minutes) - Consent check. In the UK you must inform all parties that the call is being recorded. For outbound sales calls, state it verbally at the start. For inbound support calls, the consent notice usually lives in your onboarding policy. - Pick a transcription tool. Otter, Fireflies and tl;dv each integrate with Zoom, Google Meet and Teams. Fathom integrates with Zoom and HubSpot natively. If you are already on HubSpot Sales Hub Professional, Fathom is the fastest path because it writes summaries back to contact records automatically. - Confirm your CRM has a Notes or Activity field that accepts structured text. HubSpot, Salesforce, Pipedrive and Close all do. If yours does not, a custom field for Last Call Summary works fine. - Accounts needed: meeting platform, transcription tool on a paid plan (Otter Business from 20 dollars per month, Fireflies Pro from 10 dollars per month, Fathom free tier for individual reps), CRM with API or Zapier integration, and a Zapier or Make account. ## Step 1: Connect the transcription tool to your meeting platform (30 minutes) 1. Install the transcription tool as a meeting bot. In Otter, Fireflies or tl;dv, add your calendar and grant calendar read access. The bot joins every meeting automatically, or you can set it to join on invite only. 2. Run a 5-minute internal test call. Confirm the transcript is readable and speaker labels are correct. If speaker labels are wrong, re-train the speaker profiles before going live on client calls. 3. For Fathom, connect your HubSpot account under Settings then Integrations. Fathom matches the contact by the email address used to book the meeting. If a contact is missing, it creates a draft note you approve before writing to the CRM. ## Step 2: Build the extraction prompt (20 minutes) Transcripts are long. The CRM note should be short. You need a prompt that reliably extracts the right structure from any call. 1. Open Claude or ChatGPT. Use this as your starting prompt: You are a sales admin assistant. Read the call transcript below and return a structured note with exactly four sections. One: Call summary in two to three sentences. Two: Key points raised, as a bulleted list of five items maximum. Three: Agreed next actions, each with a name and a due date if mentioned. Four: Deal or account flag, one sentence on stage or sentiment. Return plain text, no markdown. 2. Paste in a real transcript and review the output. Adjust until the four sections are consistently clean. 3. Save the final prompt to a ChatGPT Saved Prompt or a Claude Project instruction so it is reusable across the team. 4. If your transcription tool has a built-in AI summary, compare it to the prompt output. Built-in summaries are often shallow; the custom prompt usually extracts better next actions because it is told explicitly what to look for. ## Step 3: Build the Zapier automation (45 minutes) This is the step that removes the manual copy-paste. The trigger is a new transcript in your transcription tool. The action is a note created in your CRM. 1. In Zapier, create a new Zap. Set the trigger app to your transcription tool. Otter has a native trigger called New Recording. Fireflies has New Transcript. Connect your account and test the trigger with a real past transcript. 2. Add an optional Formatter step using the Text action to clean up any special characters the transcript tool adds. Prevents garbled text in the CRM. 3. Add an action step using the ChatGPT or OpenAI action: Send Chat Message. Paste your extraction prompt into the System Message field. Map the transcript body from step one into the User Message field. 4. Add a final action step using your CRM. HubSpot uses Create Note or Log Activity; Pipedrive uses Create Note. Map the ChatGPT output to the note body, and map the contact name or email from the transcript trigger to the contact association field. 5. Test end to end with a real call. Check the CRM record and confirm the note appears under the correct contact. ## Step 4: Add next-action routing (30 minutes) The basic build logs the note. This step gets the next actions into your task queue so nothing drops. 1. In Zapier, add a Filter step after the ChatGPT extraction. Check whether the text contains an action keyword such as follow-up, send, book or confirm. 2. If it does, add a Create Task action in your CRM or task manager. In HubSpot, associate the task to the deal record and assign it to the rep. In Salesforce, use Create Task. In Pipedrive, use Create Activity. 3. The filter keeps the queue clean. Not every call generates an explicit next action; the filter stops empty tasks being created. 4. In Make, the same logic uses a Router module. Route A handles calls with next actions, Route B handles informational calls with no task required. ## What you ship A fully automated call-to-CRM pipeline. Every call is transcribed, summarised and logged to the correct contact record in under 5 minutes with no manual input. Next actions land in the task queue automatically. A sales rep running 8 calls a day saves 2.5 to 3 hours of admin. Across a team of 5, that is 12 to 15 hours a week returned to selling. For implementers: this is a 2,500 to 8,000 pound build for a sales or service team depending on CRM complexity and whether call routing or sentiment scoring is included. The retainer for ongoing prompt tuning and maintenance sits at 500 to 1,500 pounds per month. ## What to build next Pair this with the AI sales outreach and CRM follow-up tutorial to close the full loop from outbound to call to pipeline. If the team is on HubSpot, the AI meeting notes comparison covers whether Fathom, Otter or Fireflies is the right fit before you commit to the build. ## Related builds More tutorials in Automation. Each one runs in 60 to 90 minutes and ships a finished asset. ### Capture every lead automatically: a 30-minute Zapier build 30 to 60 minutes. Solo Pro. [Open tutorial](/tutorials/first-zapier-automation-30-min) ### Decide between native AI Steps and a Custom GPT in 10 minutes 60 to 90 minutes. Solo Pro. [Open tutorial](/tutorials/zapier-ai-features-vs-custom-gpt) ### Run automations Zapier cannot: the first Make scenario 60 to 90 minutes. Solo Pro. [Open tutorial](/tutorials/first-make-scenario-visual) ### Self-host the automation stack: when n8n is worth the effort 60 to 90 minutes. Solo Pro / Implementer. [Open tutorial](/tutorials/n8n-self-hosted-when-worth-it) ## Questions people ask before they start ### Do I need consent to record calls? In the UK, yes. You must inform all parties that the call is being recorded. For outbound sales calls, say so at the start of the call. For inbound service calls, the consent notice usually sits in your terms of service or onboarding communication. Recording without notice is a breach of the Investigatory Powers Act and GDPR. Most transcription tools display a disclosure banner when the bot joins; confirm your legal team approves the wording. ### Which transcription tool is best for this build? Fathom is the fastest route if you are on HubSpot and Zoom. It writes summaries to contact records natively and the free individual tier covers the basics. Fireflies is a better fit if you need multi-CRM support or you are on Teams or Google Meet. Otter is strong for voice quality and speaker separation. tl;dv has a generous free tier and good Zapier support. The right choice depends on your meeting platform and CRM. ### Will the AI extraction get things wrong? Sometimes, yes. Transcription accuracy drops on poor audio, strong accents or heavy jargon. Review the first 20 logs before trusting the pipeline unsupervised. Add a Slack alert step in Zapier that pings the rep when a note is created, so they can correct it within the hour if needed. The extraction prompt improves with iteration; save the corrected version as your working prompt. ### How much should I charge to build this for a client? A standalone call-to-CRM build with one CRM integration and no task routing runs 2,500 to 4,000 pounds. Add task routing, sentiment scoring or multi-CRM support and the range moves to 5,000 to 8,000 pounds. The retainer for ongoing prompt refinement and monitoring sits at 500 to 1,500 pounds per month. Run your own pipeline on this build first; the demo is the close. ### What if my CRM does not have a Zapier integration? Most CRMs above entry-level have either a native Zapier action or a Webhooks integration. If yours does not, use Make with the HTTP module to post directly to the CRM API. Check your CRM documentation for the Notes or Activities endpoint. If the CRM has no API at all, consider whether it is the right CRM for a team that wants to automate.