How many client requests arrived as voice notes today? Before you answer, open the practice WhatsApp Web and count the unplayed ones.
The scene below is illustrative, built from what repeats in a small practice. Thursday, 5.40pm. Eleven voice notes waiting. The first runs 4 min 12 s: a client explaining that the branch has moved and asking whether that changes this month's filing. The third runs 48 seconds and asks for a duplicate invoice. The sixth runs 7 minutes, opens with football and ends with a certificate due on Tuesday. The tenth is the same client's partner repeating the sixth request, because nobody answered.
Four real requests in that inbox. Zero lines in the task tracker.
The request arrived. It just did not arrive as a request
An accounting practice runs on traceable requests: it comes in, becomes a task, gets an owner, gets a deadline, goes out. Audio breaks the first step. It arrives and sits there, inside a green bubble that exists only for whoever opened that chat, on that device.
The scale is not a feeling at the front desk. WhatsApp users send an average of 7 billion voice messages a day, a figure WhatsApp itself published in March 2022. And the app has become the counter businesses stand behind. Brazil has the clearest public series on this: the Cetic.br TIC Empresas 2024 survey found 74% of companies using WhatsApp or Telegram as their online presence, against 42% in 2017. In the same survey, 49% used WhatsApp, Skype or Facebook chat messages as an internet sales channel.
On the receiving side sits a large, fragmented profession. The Brazilian Federal Accounting Council lists 549,680 active accounting professionals and 106,938 active accounting firms. Most of them are not a global firm with a service centre. Most are three to ten people, one WhatsApp Business number, and nothing in between the voice note and the task.
Four things a badly written email does and a voice note does not
A badly written email still works as a request. A well spoken voice note does not. The difference is not how polite the client is. It is the format.
- Search. In November, when the client asks when exactly they requested the certificate, nobody finds it. Sound does not appear in a keyword search.
- Forwarding. Passing the file to the colleague in tax moves the audio, not the request. That colleague will sit through the same 7 minutes, football included.
- Copying. The amount, the date and the document number are in there, but they cannot be copied. Someone will type them by ear, then check them twice.
- Splitting. One voice note usually carries two or three requests at once. Text splits into three lines with three owners. Sound does not split.
The branch voice note is the clean example, still illustrative: changing a branch address and issuing a duplicate invoice are two tasks for two teams, records and billing. As sound, they are one task, owned by whoever happened to listen.
What exists today, and where each route stops
| Route | What it solves | Where it stops |
|---|---|---|
| Playing it back at 1.5x or 2x | Shortens the listen on short notes, with nothing to install | You still listen to all of it, and no text is left behind to search later |
| WhatsApp built-in transcripts | Reads one voice note on the phone, with the transcript generated on the device itself, a feature WhatsApp announced in November 2024 | Triggered message by message with a long press, and without the full language list at launch. It does not put the text on the computer screen, which is where the practice works |
| Downloading the file and uploading it to a standalone transcription service | Returns good text for one specific file | Download, upload, wait, then retype the text where it belongs. Multiply by eleven notes a day |
| Asking the client to put it in writing | Works with the organised client | People send voice notes because they are driving or standing at a counter. The written version arrives later, or never |
| Having someone listen through the inbox and take notes | Turns sound into real text | It costs one person's hour every day, and whatever they mishear goes into the records with no trail |
Back-of-the-envelope maths, to redo with your own numbers
What follows is back-of-the-envelope arithmetic, not survey data. Swap in your practice numbers. Eleven voice notes a day at an average of 1 min 40 s is 18 minutes of pure listening. Replaying the part that was not clear, on two notes out of five, adds 7. Typing what matters into the task tracker, 40 seconds a note, adds another 7. That closes at 32 minutes a day, near 12 hours a month, close to a full working week of one person per quarter.
None of those hours shows up on a fee note. They show up on Friday night, when the partner finds the certificate that was due on Tuesday.
From voice note to logged request, without leaving the chat
The swap is easy to describe: instead of carrying the audio out of WhatsApp, you let the text appear inside the conversation. ZapVox is an extension that works in WhatsApp Web, on Chrome, Edge and Firefox. In an accounting practice the route looks like this:
- Automatic transcription. New voice notes in that chat turn into text on their own, under the message. No download, no renaming.
- Bullet-point summary. The 7-minute note becomes a few lines with what was actually asked. That is what gets pasted into the task tracker, not the full transcript.
- Glossary and names your way. You teach it the terms your practice uses and the name of every client, so the transcript stops tripping over the one word that mattered.
- Notes per contact. The note stays pinned to the chat and visible only to you. Useful for the thing that is not a task today but will be billed in March.
- Search inside voice messages. When the question lands in November, you search the word certificate across the transcripts. Results come grouped by person and open the chat at the right point.
- Export the chat transcriptions. It comes out as .txt, ready to attach to the client folder or send to whoever owns that team.
- Schedule messages and files. The filing goes out on the right day, with the recipient checked before sending, instead of depending on someone remembering at 11pm.
Bullet-point summary, glossary and export sit on the paid plans and the 5-day trial. Automatic transcription, scheduling, notes and search exist on the Free plan too, with a smaller quota.
Notice what did not change: nobody stopped listening to clients. What stopped existing is the trip from the voice note to the downloads folder and back as text typed by ear.
Where this stops
Worth saying what the extension does not do, because a practice that changes its process on a promise changes it twice.
- It is a browser, not a phone. It works in WhatsApp Web, with the computer on and the tab open. The note that lands at 10pm on the partner's phone only becomes text when somebody opens WhatsApp Web.
- There are quotas. The Free plan gives 8 cloud transcriptions a day and up to 15 minutes of audio a day, with 3 scheduled messages a day, notes for up to 10 contacts, settings for up to 10 chats and search across the last 7 days. The trial unlocks the Pro features for 5 days, with 15 operations a day and up to 30 minutes of audio a day. A practice taking eleven notes a day should check the quota before building a process on top of it.
- Transcription gets things wrong. It mishears proper names, it mishears numbers spoken fast, it struggles with audio recorded at a noisy counter. Amounts, dates and document numbers are always checked against the document. The glossary lowers the error rate, it does not remove it.
- It is not evidence. A transcript is an internal working tool, for sorting, prioritising and routing requests. It does not replace the document and should not be presented as proof of anything.
- Processing happens in the cloud. Audio is processed on the spot and not stored, but it is sent off for processing. If you would rather run it through your own AI account, you can connect a Groq or OpenAI key, free on every plan.
Test it before you change the practice process
Install it, point it at the three chats that send the most audio, and watch for a week. If the summary is not good enough to paste into the task tracker, you find out in two days instead of two months. Every signup starts with 5 days of trial with the Pro features, no credit card. After the 5 days the account drops back to the Free plan on its own and nothing is charged.
Try 5 days free, no card. Then count how many client requests came in as voice notes, and how many became a task the same day.