Sales & CRM

How to log sales calls to your crm automatically with AI

Sales & CRM3 AI tools7 steps6 friction points

Want an agent to handle this on your live data? See the Starch version →

After every sales call, someone has to write down what happened: what the prospect said, what objections came up, what was promised, what the next step is. For most small sales teams and operator-founders, that someone is you. The call ends, you have three more in the next two hours, and the CRM entry either gets done sloppily right after or not at all. The result is a contact record that's half-empty and a pipeline you can't actually trust.

This is exactly the kind of task that feels like AI should handle it. The raw material is text — a transcript, a few notes, maybe a recording. The output is structured data: a contact updated, a deal stage moved, a follow-up task created. There's no judgment required, no relationship nuance that needs a human. It's pattern-matching from unstructured input to a known schema, which is what large language models are genuinely good at.

ChatGPT, Claude, and Gemini can all do meaningful work here today. Paste in a call transcript and ask one of them to extract the key points, identify action items, and format the output as a CRM update — and you'll get something usable in under a minute. The limitation isn't the quality of the extraction. It's everything that has to happen around that extraction: getting the transcript in, getting the output into your actual CRM, and making sure it happens consistently on every call, not just the ones where you remembered to run the prompt.

Skip the manual work → See the Starch version

Sales & CRM3 AI tools7 steps6 friction points

AI walkthrough

How to do it with AI today

A practical walkthrough using ChatGPT, Claude, and other off-the-shelf LLMs — what they're good at, what you'll have to do by hand.

Tools that work for this

ChatGPTClaudeGemini

Step-by-step

1 Record your sales call using Zoom, Google Meet, or any tool that produces a transcript. Download the transcript as a text file or copy it directly from the platform's transcript viewer after the call ends.

2 Open ChatGPT, Claude, or Gemini and paste the full transcript into the chat window. If the transcript is long, Claude handles the largest context windows most reliably for this use case.

3 Paste your extraction prompt (see examples below) immediately after the transcript. Be explicit about which fields your CRM tracks — the AI will output whatever schema you describe, so the more specific you are, the less cleanup you do afterward.

4 Review the structured output. LLMs occasionally misattribute who said what on a call, or label an objection as an action item. Skim for those errors before you do anything with the data.

5 Copy the CRM fields the AI produced — contact name, company, deal stage, key notes, next steps, follow-up date — and manually paste them into the appropriate record in your CRM. There is no direct connection between the LLM and your CRM; this step is manual every time.

6 If the AI produced a follow-up email draft as part of the same prompt, copy that into your email client and send or schedule it. Again: manual copy-paste, no automation.

7 Repeat this sequence for every call. There is no memory between sessions — next week's call starts from a blank prompt.

Prompts you can copy

Here is a sales call transcript. Extract: prospect name, company, role, key pain points mentioned, objections raised, what I committed to, agreed next step, and ideal follow-up date. Format as a CRM update with labeled fields.

From this transcript, write a 3-sentence call summary suitable for a CRM note, then list action items as a numbered list with owner (me or prospect) and due date for each.

Read this call transcript and tell me: did the prospect express buying intent? What was their main hesitation? What did I promise to send them? Output as three short labeled answers.

Draft a follow-up email I can send to the prospect within 24 hours of this call. Reference two specific things they mentioned. Keep it under 150 words and end with a clear next step.

This transcript is from a discovery call. Score the lead 1-10 on fit based on the pain points described, and explain the score in two sentences. Then extract the top three things I should address in the next call.

Reality check

Where this gets hard

The walkthrough above works — until your numbers change, the LLM hallucinates, or you have to re-paste everything next month.

No connection to your actual CRM — every field the AI extracts has to be manually copied into HubSpot, Salesforce, or wherever your deals live. That's 3-5 minutes of admin per call.

Transcript ingestion is manual. You download the file, paste the text, and re-run the prompt fresh every time. There's no listening for a call to end and kicking off the workflow automatically.

Output format drifts. The structured JSON or labeled fields you carefully prompted last Tuesday look slightly different today because you reworded something. Your CRM entries end up inconsistent over time.

No memory across calls. Ask the AI 'what did this prospect say last month?' and it has no idea. Each session is stateless — the LLM has never seen your prior call with this person.

Context window limits bite on long calls. A 90-minute enterprise discovery call transcript can exceed what some models handle cleanly, leading to truncated summaries or missed details from the second half.

Nothing enforces the habit. When you're busy, the prompt doesn't run itself. Calls go unlogged, records stay stale, and the pipeline you're supposed to be managing reflects whatever you had energy to update.

Tired of the friction?

Starch runs the whole workflow on live data — no copy-paste, no hallucinated numbers, no re-prompting next month.

See the Starch version →

Starch alternative

The same workflow on Starch

Starch is an agentic operating system — it builds and runs the persistent software that handles this workflow continuously against your live data, so the logging happens whether or not you remembered to run a prompt.

Starch connects directly to HubSpot, Capsule CRM, or your custom CRM through its integration catalog — the agent writes extracted call data directly to the right contact or deal record, no copy-paste required.

Start with the Sales Agent CRM starter app — pre-built with deal tracking, contact history, and email context from Gmail — then describe any additional fields or pipeline stages you need and Starch adjusts the schema to match.

Starch syncs your Gmail and Google Calendar on a schedule, so it knows which calls happened, who was on them, and what threads exist for each contact before it even touches the transcript.

Describe the automation in plain English: 'After each call logged in Calendly, pull the transcript, extract action items and deal stage, update the contact record in my CRM, and draft a follow-up email.' Starch builds that as a persistent automation, not a one-time prompt.

The CRM starter app answers questions like 'who haven't I followed up with in 30 days?' against your actual call history — not a canned report, but a live query across every logged interaction in your pipeline.

No call goes unlogged because the operator forgot to run a prompt. The automation triggers on the calendar event or transcript arrival and runs without manual intervention — the record is updated before your next call starts.

Get closed-beta access →

Toolkit