Meeting summaries with AI: from conversation to CRM
Logging the meeting is the task salespeople put off the most and the one managers need the most. How to turn the transcript into CRM fields with AI, with the prompt, the two-minute review and the care the recording requires.
A meeting summary with AI is the process in which a tool records and transcribes the conversation with the customer, the AI turns the transcript into a summary organized by CRM fields and the salesperson reviews and approves it before the record goes into the deal. Done this way, the record is ready minutes after the meeting, in the customer’s own words, and the salesperson trades twenty minutes of typing for two minutes of reading. If you lead sales, your decision has four parts: which fields the summary fills, who approves before the data enters the CRM, how the customer learns about the recording and which tools the team may use. Without those four definitions, each salesperson builds their own setup and the CRM keeps collecting notes nobody can compare.
When I open the CRM of a sales operation, the meeting notes tell the same story. “Good meeting, customer interested, send proposal.” The record gets written on Friday night, from memory, when it gets written at all. In the pipeline meeting, the manager asks again who attended, what the problem was and what was agreed, and the salesperson answers with what they remember. The data is missing because logging competes with selling, and selling wins. In its 2024 State of Sales report, Salesforce says sales reps report spending 70% of their time on non-selling tasks, in a survey of 5,500 professionals across 27 countries. Meeting notes are a large part of that and one of the easiest to hand to AI.
How does an AI meeting summary work?
The flow has five steps, and only the fourth needs the salesperson. Video conferencing tools such as Zoom, Teams and Meet offer recording and transcription on many plans, and specialized tools such as Fathom record, summarize and in some cases send the result to the CRM. The choice depends on what the company already pays for and what IT approves.
- Record. The tool joins the meeting with a notice to attendees, and the salesperson confirms out loud, at the opening, that the conversation is being recorded.
- Transcribe. The tool produces the text of the conversation, identifying who spoke.
- Summarize. The AI reads the transcript and fills a fixed template with the fields the CRM needs.
- Review. The salesperson reads the summary, corrects names, numbers and commitments and approves it. This step takes two to five minutes.
- Log. The approved summary goes into the deal, as a note and in the structured fields, and the next step becomes a task with a date.
Step 3 is where most companies go wrong. When each salesperson asks for “a summary of the meeting”, each AI returns a different format, and the manager receives long texts that fit no field. A fixed template, the same for the whole team, makes it possible to compare deals and feed reports.
Which CRM fields should the summary fill?
The summary serves the CRM when each part of it has a destination. I separate what goes into structured fields, which feed filters and reports, from what stays in the note, which serves whoever reads the deal’s history. The list below is a starting point for consultative B2B sales, and each company adjusts it to the fields it already uses.
| Field | What the AI extracts | Illustrative example | CRM destination |
|---|---|---|---|
| Main problem | The pain in the customer’s words | “We lose track of the order once it leaves sales” | Deal text field |
| Impact | Cost, time or risk the customer mentioned | “About 30 late orders a month” | Text field and note |
| Attendees and roles | Who spoke and their apparent role in the decision | Operations director, influencer | Associated contacts with role |
| Decision criteria | What the customer said will weigh in the choice | ERP integration, implementation time | Deal text field |
| Timeline and event | Date or reason that sets when to decide | Budget approved in March | Expected close date, with justification |
| Objections | Doubts and resistance, with the answer given | Price above the current vendor | Note and objection field |
| Next step | What was agreed, with date and owner | Demo for the operations team on the 14th | Task with date |
The decision-role and next-step fields are the ones that change management the most. They show whether the deal has access to the people who decide, the subject of B2B buying committee, and whether there is a booked commitment or only a promise to get back. If the company uses a qualification methodology such as MEDDIC or MEDDPICC, the summary fields follow its letters.
Item 8 becomes the summary email the salesperson sends the customer the same day. A customer who confirms the problem, the criteria and the next step in writing signs off on their part of the process, and the salesperson gets a document to take to the rest of the committee.
Can AI write straight into the CRM?
It can, and some tools already do. The question for leadership is what may go in without review. I recommend starting with everything going through the salesperson and releasing automatic entry field by field, as reviews confirm the quality.
| Type of data | Automatic entry | Entry with salesperson approval |
|---|---|---|
| Meeting date, duration and attendees | Yes | |
| Transcript and recording link | Yes, with restricted access | |
| Summary note | After a few weeks of reviews without relevant corrections | At the start |
| Problem, impact, criteria | Yes | |
| Expected close date and amount | Yes, always | |
| Deal stage | Yes, always, based on the stage exit criteria |
Amount, close date and stage stay with the salesperson because they feed the sales forecast. A deal that moves stage because the AI read “let’s move forward” in a polite sentence contaminates the number that goes to the executive team. The architecture of agents that write to the CRM, with what each one does on its own and who approves the rest, is in AI agents.
Do you need to tell the customer the meeting is being recorded?
Yes, and the notice protects the relationship as much as the company. Recording and transcription process personal data: voice, name, title and what each person said. Brazil’s General Data Protection Law (LGPD) sets out, in article 6, the principles of transparency and necessity, which call for clear information to the data subject and processing limited to what the purpose requires. The legal basis and the details of each case belong with the company’s legal team, and the full topic is in AI and LGPD.
In sales practice, the notice has three moments. The invite says the meeting will be recorded for internal records. At the opening, the salesperson asks whether the customer agrees and says what the recording is for. If the customer declines, the salesperson turns the tool off, takes notes by hand and uses AI afterward on those notes. Many customers ask where the recording is stored and for how long, and the salesperson needs to be able to answer, which requires a written retention rule in the AI use policy.
Two more precautions complete the picture. Access to recordings stays restricted to the people working the account and to leadership, with no folder open to the whole company. And the summary leaves out what does not serve the deal: personal comments, health information, opinions about the customer’s colleagues.
What mistakes does AI make in meeting summaries?
AI summarizes well what was said clearly and gets wrong what was left incomplete or ambiguous. The most common errors I see in reviews are predictable, and the two-minute review exists to catch them:
- Names and titles. The transcript gets proper names wrong and the AI repeats the error with confidence.
- Numbers. “Thirty” becomes “thirteen”, “thousand” disappears, and a customer estimate shows up as fact.
- Commitments nobody made. “It would be interesting to see a demo” becomes “customer requested a demo next week”.
- Words attributed to the wrong person. When two people from the customer speak one after the other, the tool may merge what they said.
- Optimistic tone. The summary tends to record interest and soften objections, and the salesperson has to put back into the text the doubt the customer made clear.
The last instruction in the prompt, which lists numbers, names and dates for checking, cuts review time because it points to where to look. The mechanics behind AI’s confident errors are in AI hallucination.
How does the manager use meeting summaries?
With summaries in the same template, the pipeline meeting changes character. The manager arrives knowing the problem, the roles and the next step of each deal, and uses the time to discuss strategy in place of asking for an account of events. Deals with no dated next step show up in a filter before the meeting.
In coaching, the summary points to which part of the recording to listen to. If the impact field came back “not mentioned” in four of a salesperson’s five meetings, the manager knows which question to train and can listen to the minute where it should have appeared. The method is in sales coaching.
For marketing, the problem and objection fields, with the customer’s own phrases, become the most reliable source of messaging the company has. For post-sale teams, the summary of the last meeting before signature says what the customer expects and how they will measure success, and the handoff to B2B post-sale no longer depends on the salesperson’s memory.
The effect on CRM adoption shows up fast. A salesperson who does not have to type opens the CRM to read what the AI prepared, and the record starts serving them. That is the central point of CRM adoption: the team uses the system when the system hands back finished work. Today’s summary also feeds preparation for the next conversation, as I show in pre-meeting briefing with AI.
A four-week rollout plan
- Week 1, template. Define the summary fields based on the fields the CRM already has, write the prompt and choose the recording tool approved by IT and legal.
- Week 2, pilot. Two or three salespeople use the flow in every meeting. The manager reads the summaries and notes the most frequent corrections.
- Week 3, adjustment. Adjust the prompt based on the corrections, write the recording notice and the retention rule, and open it to the team.
- Week 4, routine. Take the filter of deals without a next step to the pipeline meeting and use one summary per salesperson in the 1:1.
If your leadership wants AI meeting notes to become a team standard, with fields, review and recording notice defined, that is the work of the in-company applied AI training. I run the program on your company’s CRM and real meetings, with prompts by meeting type, a summary template tied to the system’s fields and data rules written with your legal team, and I prepare managers to use the summaries in the pipeline meeting and in coaching.
Frequently asked questions
Which tool should I use to summarize meetings with AI?
The one the company already pays for and IT approves. Zoom, Teams and Meet offer transcription on many plans, and tools such as Fathom record and summarize. The field template and the salesperson’s review matter more than the brand.
Can AI update the deal stage on its own?
I do not recommend it. Stage, amount and close date feed the forecast and stay with the salesperson, based on each stage’s exit criteria.
Do I need the customer’s permission to record the meeting?
State it in the invite and confirm at the opening, saying what the recording is for. If the customer declines, do not record. The legal basis for each case is defined with the company’s legal team.
How much time does a salesperson save with AI summaries?
It depends on how many meetings they run and how they logged them before. Measure it in your own operation: time the manual logging of a few meetings and compare it with reviewing the summary.
Is the AI summary reliable?
It is reliable once reviewed. AI gets names, numbers and commitments wrong, and the two-to-five-minute review exists to fix those points before the data enters the CRM.
What should we do with the full transcript?
Keep it with restricted access and a defined retention period. It serves coaching and occasional lookups, and what goes into reports is the reviewed summary.
Want to bring this to your team?
The topics in this article also become in-company training: designed from a diagnosis of your operation, with real cases from your team and a deliverable that stays with the company.
More writing on the same subject.

Weekly sales report with AI: the Monday summary
A manager who builds the weekly report on Sunday night arrives at the meeting tired and with numbers that do not match other teams'. How to set up a weekly summary that the CRM calculates, AI writes and the manager reviews in minutes.
11 min readRead more
B2B price increases: how to announce one and keep the client
An increase that arrives as a generic email from finance pushes the client toward a competitor or toward a request for an exception. How to segment the customer base before announcing, when to give notice, what to say in the conversation, which trade-offs to offer and what to measure in the following 90 days.
14 min readRead more
Sales proposals with AI: fast draft, human review
AI writes the first draft of a proposal in minutes when it receives the discovery summary, the company template and the pricing rules. The salesperson and the manager review scope, prices, timelines and promises before anything goes out. The workflow, the prompt and the review checklist.
12 min readRead more