Sales call transcription tools: how to choose

Zoom, Meet and Teams already transcribe your calls, and dozens of meeting assistants offer it too. For a sales team, what separates them comes down to a few things.

The short answer: for a sales call, a transcript is worth four things, speaker separation, correct proper names, accuracy in your language and timestamps. The transcript built into your video tool is enough to keep a record. As soon as you want to find an objection or a competitor across every call, you need a tool that files and analyses what it transcribes.

What a transcript needs to do for sales

An internal meeting transcript and a sales call transcript have different jobs. The first keeps a record. The second has to tell you, weeks later, who said what, on which topic, and at what point in the call.

Separate the speakers

A continuous block of text with no speakers is useless in sales. You need to know whether the price objection came from the customer or from your rep anticipating it. Check that every line is attributed to a person, and that the tool tells your side from the customer's, even when three people speak for the buyer.

Get proper names and product terms right

This is where general transcription fails most. The customer's name, their company, your product, the competitor they mention: these are the most important words in the call, and the ones a general model has never seen. A misspelled competitor is a competitor you will never find with a search. Ask whether the tool can take a list of terms to recognise.

Be accurate in your customers' language

Many tools are trained on English first. If you sell in French, German or Spanish, Zoom, Meet and Teams support those languages, but quality varies with accent, pace and background noise. The only reliable test is yours: transcribe three of your own calls and read them end to end.

Timestamp every line to replay a passage

Reading a passage does not tell you everything. The tone of a hesitation, a silence after the price is announced, those you hear. Every line should carry its timestamp, and a click on the text should start the recording at the right moment.

The four families of transcription tools

Every call transcription tool falls into one of these four families. They differ less in the quality of the text than in what they do with it afterwards.

Built into Zoom, Meet and Teams

This is the option you already pay for. In Zoom, audio transcription for cloud recordings requires a Pro, Business, Education or Enterprise account, cloud recording and the setting turned on. The output is a VTT file, timestamped and attributed to speakers, with an "Unknown Speaker" label for anyone not identified.

Google Meet offers transcripts from the Business Standard edition of Google Workspace upwards, in English, French and several other languages; the file lands in the organiser's Google Drive. Teams transcribes meetings with the speaker's name and a timestamp, stores the text in the organiser's OneDrive, and only the organiser and people from their organisation can start it.

What is missing: each transcript lives in its own folder, on the platform of whoever organised the call. When the customer invites you to their meeting, you often get nothing. And nothing links Tuesday's call to the one three weeks ago with the same customer.

Note takers with a bot

An assistant joins the meeting as a participant, records, transcribes and sends a summary. It works on every platform, including when the customer sends the invite. In return, the bot shows up in the participant list, and its summary stays centred on the meeting, not on the customer.

Transcription APIs

Services like Deepgram turn an audio file into text, with speaker separation and timestamps, and accept a list of terms to recognise. It is the best raw quality, at the cost of development work: you have to fetch the recordings, call the API, store the text and build something to read it. For a sales team with no developer, it is not an option.

Sales tools that analyse after transcribing

Here transcription is only a step. The tool then cuts the call into passages, spots objections, competitors and commitments, and attaches everything to the customer or the deal. It is often the most expensive family, but the only one where the transcript does something other than sit in an archive.

Bot or no bot, and the customer's consent

There are two ways to capture a call. A bot joins the meeting as a participant. An extension or an app records from the rep's own machine, adding nothing to the list. The second spares the rep from opening the meeting with "don't mind this participant".

No bot does not mean no notice. The French data protection regulator, the CNIL, says so in its guidance on listening to and recording calls, flagged as being updated. The other party is told at the start of the call and pointed to written information. They can object to the recording before the conversation ends. The same page sets a retention of six months at most for recordings, and up to one year for analysis documents. Your own country's rules may differ, so check them.

• Announce the recording in one sentence at the start of every call, and note the answer.
• Point to written information: a line in the invite or in your privacy policy.
• Plan the stop: the rep must know how to turn recording off if the customer refuses.
• Tell your team: your reps are being recorded too.

Why a transcript alone is not enough

Nobody rereads a transcript. A 45-minute call runs to several pages, and a team of five reps produces dozens a week. The text exists, but it sleeps in a folder.

The questions a sales director asks are never about one call. Which objections keep coming up this quarter? Which competitor is named most often? Why did we lose the last three deals? To answer, you have to search every call at once, by meaning and not only by exact word, and then count.

A transcript answers "what was said in this call?". Your question is "what are our customers saying?".

How we transcribe at Meidly

We record Google Meet, Zoom and Teams calls in the browser, through a Chrome extension, with no bot in the meeting. The Zoom and Teams desktop apps are not supported. You can also import audio, video or text files, and the emails exchanged with your customers from a connected mailbox.

• The engine: we transcribe with Deepgram Nova 3, which accepts a list of terms to recognise.
• The names we already know: we give it the customer's name, their company, the known participants, your team and the competitors already named. Those are the words a general model gets wrong most often.
• The exact text stays the record: the transcript is never rephrased. A readable version only removes hesitations and repetitions, and you switch between the two in one click.
• After transcription: the conversation is cut into passages tagged by meaning (pain, objection, competitor named, budget, buying signal...) and attached to the customer's file, with its commitments and its sales. An assistant answers your questions by citing those passages, with real tallies.

Your data is hosted in the EU and trains no AI. If all you want is a written record of your meetings, your video tool's transcript is enough. If you want to know what makes your customers sign, what blocks them and what to bring back on the next deal, that is what we built it for.

Which tool to pick, by need

• You want a written record of your meetings, with no budget: turn on transcription in Zoom, Meet or Teams, after checking your plan includes it.
• Your customers often invite you to their own meetings: a tool that records from your side, with or without a bot.
• You have a developer and custom needs: a transcription API, with a list of terms.
• You want to find objections and competitors across every call: a tool that analyses and files by customer.

Either way, test on your own calls, in your customers' language, with real customers. Read the proper names first: that is where you will see the difference.