Customer voice notes: turn them into notes and tasks

It's eight in the evening and you're in the van. There are four voice notes waiting on WhatsApp: one a minute and a half long, one fifty seconds, two shorter ones. You play them while you drive and get the gist. Then, at home, you play them again to write down the address, the day they asked for and the boiler model someone mentioned in passing. Next morning a colleague asks "what did the woman on Olive Street want?" and you're hunting for the same voice note a third time. This article is for owners who get a lot of voice messages from customers. By the end you'll know how to set things up so every voice note turns into a written summary and, where needed, a task, and you only listen when it actually adds something.

Why a voice note gets played two or three times

The trouble with voice notes isn't their length. It's that what's inside them can't be searched, copied or forwarded. The first listen tells you what the customer wants. The second is for pulling out the details: a time, a flat number, a reference. The third comes when someone else needs those details and the only place they live is your phone.

For the customer, a voice note is the easiest thing in the world: talk while walking, then forget about it. For you, each one is a small bit of admin. Five a week, who cares. Thirty a day, which is normal for plenty of repair shops, builders, clinics and made-to-order businesses, and it's a real slice of your working day that never shows up anywhere.

Step 1: transcription is good, but it's not enough

The first thing many people find out is that WhatsApp can already transcribe voice notes on your phone. It added voice message transcripts in late 2024, in several languages including English and Spanish (TechCrunch covered the launch). Turn it on, because it saves you plays. But be clear about what it doesn't do:

• The text stays in that chat, on that phone. Your team can't see it, it doesn't reach your calendar and it doesn't create a task.
• It's word for word. "Erm, so, like I was saying the other day, actually no, maybe better if…" Reading that is faster than listening, but you still have to dig out what matters.
• It can't tell urgent from routine. A complaint and a question about opening hours look exactly the same.

Transcription is step one. The time savings come from what happens next: someone, a person or an agent, turning that text into something you can act on.

Step 2: decide what summary you want from every voice note

Before you think about tools, write down what you need to know from each voice message. A summary card that works for almost any business has six fields:

1. Who: name and, for existing customers, which record or order it relates to.
2. What they want, in one line: "Move Thursday's appointment to the afternoon", not three paragraphs.
3. Hard details: dates, times, addresses, references, amounts, models.
4. What needs doing and by when: the action, who owns it, the deadline.
5. What isn't clear: the missing detail or the one that may have been misheard.
6. Tone: if they're in a hurry or annoyed, you want to know before you reply.

Here's the kind of voice note a small building and repairs firm gets every day:

"Hi, it's Lucy, the flat on Olive Street. The kitchen tap's still dripping even though you replaced it, and now the water's coming out really weak… Could you pop round Thursday afternoon? I'm home from five. Oh, and can you let me know if that's covered by the guarantee or what?"

The card that should come out of it:

• Who: Lucy, Olive Street (kitchen job last month).
• What she wants: the new tap checked; it drips and has low pressure.
• Details: available Thursday from 5 pm.
• Task: book a Thursday afternoon visit and confirm whether the guarantee covers it.
• To check: building and flat number, if not already on file.
• Tone: a bit annoyed; this is a complaint about finished work.

Reading that takes ten seconds. And anyone on your team can act on it without having heard the recording.

Step 3: put the task where you already work

The most common mistake is creating a new place for "voice note tasks". Don't. The card has to land wherever your team already looks every morning: the shared calendar, the order sheet, whatever board you use, or an internal note on the customer's own conversation.

The sensible setup is for the agent to draft the card and suggest the task, and for a person to confirm it with a tap. At first you approve everything. Once you trust the cards, you can let simple ones (an opening-hours question, a confirmation) close on their own and keep for yourself only the ones that move the diary or money. You approve, the agent does the legwork.

One detail makes a big difference: every card should link back to the original voice note. If something doesn't add up, you play that one recording right then, instead of scrolling through the chat.

Step 4: what the agent must never take for granted

Automatic transcription is very good with ordinary sentences and noticeably worse with the things you care about most: numbers, names and the words of your trade. "Fifteen" and "fifty" sound alike in a voice note recorded on a busy street. Your street name, a boiler brand or a part number can come out as something else entirely.

So set three rules from day one:

• If a detail isn't clear, it's flagged "to check" and the agent asks the customer in writing: "Just so we get it right, could you confirm the flat number?" Nobody minds a message like that, and it saves you a wasted trip.
• One voice note with several requests means several tasks. Long voice notes often bundle two or three things ("the tap, and while you're here, can you look at the blind"). Squash them into one task and the second one gets lost.
• Complaints and sensitive topics go to a person. The agent can prepare the card, but the reply to Lucy, who's annoyed about work she's already paid for, should come from you or someone on your team.

When it isn't worth setting up

Not every business needs this. Think twice if:

• You get few voice notes. Under ten a week, WhatsApp's own transcription and a bit of discipline will do.
• The voice notes carry sensitive data. In a clinic or a law firm, a voice message can include health details or case information. Before running it through any tool, check where it's processed and what agreement you have with the provider (we covered this in our article on GDPR and AI).
• Your customers record in noisy places or switch languages. It works, but test it on your own real recordings, not on a demo.
• The real problem is something else. If customers send voice notes to book because there's no other way to do it, what you might be missing is a booking link, not a voice note system.

How to try it in a week

Take the last twenty voice notes you've had from customers. Write the card for five of them yourself, by hand, as a benchmark. Then run all twenty through the system and compare: did it understand what each customer wanted? How many details did it get wrong? Did it flag as uncertain the things that really were uncertain? If eighteen out of twenty cards are usable as they are, you're on track. If it's tripping up on numbers, tighten the "to check" rule before going further.

And note how much time you used to spend on voice notes. It's usually a lot more than you think.

If you'd like to see how an agent picks up your WhatsApp messages, voice notes included, and only passes you what needs your judgement, take a look at our receptionist.