When real-time translation is the wrong answer: building async-first multilingual workflows

Real-time transcription and translation is the obvious answer for multilingual meeting problems. Someone speaks, the translation appears, everyone understands. Problem solved.

Except for a lot of workflows, sync meetings are not the right starting point. The translation tool is solving the right problem with the wrong medium.

MangoFinch is a real-time transcription product. We benefit commercially from people having more sync meetings. It would be against our commercial interest to tell you this. But two years of working with multilingual teams has made something clear: some of the best multilingual workflows we have seen use almost no synchronous meetings at all.

The synchronous meeting assumption

The default fix for multilingual communication problems is "better sync meetings." Add transcription. Add translation. Add closed captions. Run better facilitation. Speak more slowly.

These are all useful adjustments. But they assume that synchronous communication is the right format for the information being exchanged. Often it is not.

Think about what happens in a typical multinational team's weekly call. There is an agenda with eight items. Items 1-4 are status updates. Item 5 is a decision that two people need to align on. Items 6-7 are informational. Item 8 is a planning discussion.

For a native English speaker in New York, that meeting is fine. You attend, listen, contribute when relevant, and leave with updates and a decision.

For a Japanese team member who joined the call at 10 PM Tokyo time, and who processes complex English sentences about 20% more slowly than a native speaker, that meeting is exhausting. The translation appears on screen, but it is catching up to the conversation. By the time you have processed item 5, item 6 is already being discussed. You can follow the thread or think about the decision, but not both simultaneously.

The problem is not translation quality. The problem is the time pressure of synchronous communication.

Where async wins

GitLab runs 2,000+ employees across 65+ countries with a culture that is deliberately async-first. They publish their internal communication guidelines publicly. One core principle: if the meeting can be an async update, it should be.

This principle is more powerful for multilingual teams than for monolingual ones, for a specific reason. Async communication removes time pressure. The reader can stop, re-read, look up a word, think about what was said, and respond when ready. A non-native speaker reading a written summary has full access to translation tools, their own native-language notes, and time to formulate a precise response. A non-native speaker on a live call has none of those.

When Doist, the company behind Todoist, published internal research on their async-first model, they found that employees reported meaningfully higher quality of communication in async formats than sync formats. The reason was time. People had time to think before responding instead of reacting in real time. For non-native speakers, that time gap matters even more.

For a multilingual team, a well-written async summary of a meeting is often worth more to non-native speakers than the meeting itself.

The timezone problem that makes this harder

Accent bias and language processing are the obvious challenges for multilingual communication. The timezone problem is less discussed but often more damaging.

A team spread across New York, London, Nairobi, and Tokyo has a maximum overlap window of about two hours per day without anyone joining at an unreasonable time. In practice, most teams choose a time that works for one or two regions and asks everyone else to sacrifice. The Tokyo office joins at 9 PM. The Nairobi office joins at the same time as London but has already been working for seven hours.

Sync meetings in mismatched timezones have a known effect on participation quality. People who are tired, or who joined at an off-hour, contribute less, ask fewer questions, and retain less of what was discussed. For non-native speakers already managing the cognitive load of operating in a second language, the added fatigue of late or early hours compounds the problem.

Async communication solves the timezone problem directly. The engineer in Nairobi records their update at 9 AM. The engineer in Tokyo watches it at 9 AM Tokyo time, 12 hours later. Nobody sacrifices sleep. Nobody zones out during hour three of a call that started at midnight.

When you combine timezone distribution with language diversity, async-first stops being a preference and starts being an equity issue.

What the stack looks like in practice

One of our beta teams is a 14-person product group spread across Brazil, Japan, Germany, and the US. They run what we call an async-primary workflow. Here is how it works.

Status updates happen in Notion, in writing. Each team member updates their section daily. No standup call. The Brazilian members write in English but keep sentences short and avoid idioms. The Japanese members write in English or Japanese; the others use DeepL or Google Translate to read the Japanese updates when needed.

Weekly alignment happens on a Zoom call with MangoFinch running. The call is 40 minutes instead of the 90-minute weekly meeting they used to hold. They come prepared because they read the async updates, so the sync time covers only what genuinely needs real-time conversation: decisions, blockers, anything requiring back-and-forth.

Complex explanations use async video. When a team member in Brazil needs to explain a new API design to the Tokyo team, they record a Loom. The Loom auto-transcribes in English. The Japanese team members get a translation of the transcript and can watch the video at their own pace, pause when needed, and re-watch any section. No one needs to be on a call at the same time.

What MangoFinch handles for them is the weekly alignment call, not the entire communication stack. The tool works at the point where real-time conversation is genuinely necessary.

The four categories and which medium each needs

After working with about 40 multilingual teams over the past year, we see four categories of work with distinct communication needs.

Status updates — what happened, what I am working on, what is blocked. This almost never needs sync communication. Written async works better for most teams. For multilingual teams specifically, written async is better because non-native speakers can write more precisely than they can speak under time pressure.

Decisions that need alignment. Two to four people need to reach agreement on something specific. This is worth a short sync call, and this is where real-time transcription and translation adds the most value. The conversation needs to be live. The stakes of misunderstanding are high. You want to ask clarifying questions in real time.

Complex explanations. One person needs to transfer knowledge to others. Async video is usually better here. The presenter prepares, records once, and everyone watches at their own pace. Non-native speakers can pause, replay, and supplement with translation tools. A 20-minute explanation video beats a 45-minute meeting where the presenter repeats themselves for the people who missed parts of it.

Brainstorms and open-ended problem solving. This is the hardest category for async. Brainstorming benefits from rapid iteration and building on each other's ideas. For multilingual teams, real-time translation matters most here, because the speed of exchange is part of the value. A focused sync meeting with good translation is the right format.

Categorize your recurring meetings by these four types. You will immediately see which ones should never have been sync calls in the first place.

The async video transcription problem

Here is where MangoFinch fits into the async stack differently from its main use case.

When someone records a Loom or a Notion video, the platform auto-transcribes it. For a native English speaker, this usually works fine. For a non-native English speaker recording in English, or someone recording in their native language, the transcript quality degrades significantly. The same accent bias issues we covered in the previous article apply here, with one added problem: async video has no fallback. In a live meeting you can ask someone to repeat themselves. In a recorded video, a bad transcription just stays bad.

We built an async video transcription feature into MangoFinch that handles uploaded recordings separately from live meeting transcription. You upload a recorded video, get an accent-robust transcript, and generate multilingual translations from it. The output is a structured document with the original transcript, translated versions, timestamps, and a summary that you can drop into Notion or share via Slack.

The quality is better than Loom's built-in transcription for non-native speakers, and significantly better for non-English source languages. If your async-primary team produces Looms or knowledge-base videos with non-English speakers, this is the gap we are trying to close.

What teams actually report

Three patterns show up consistently in beta teams running async-primary workflows.

Sync meeting time drops. Teams that shift to async-primary typically cut sync meeting time by 40-60%. The effect is larger for multilingual teams because the cognitive overhead of real-time interpretation is significant. When status updates move to written async, the remaining sync time gets focused on work that actually needs it.

Non-native speakers participate more in written async formats. This one surprised us. In sync meetings with real-time translation, some non-native speakers still hold back, particularly for complex contributions. In async text formats, those same people write full, detailed, carefully expressed updates. The time pressure is gone. The quality of their contributions goes up.

The sync time that remains is better. When you reduce sync meetings to only what genuinely requires them, the remaining calls are more focused. Less filler status content because everyone read the async updates. Decisions get made faster because people came prepared. The meeting is doing the thing only meetings can do, instead of trying to do everything.

Building the shift

If you want to move toward async-primary, start with status updates. Move your standups to written async for two weeks. Measure the time savings and the participation quality, specifically looking at whether non-native speakers contribute more in the written format than they did in the live standup.

If that works, look at your complex explanation meetings. Which ones could be a recorded video instead of a call? Record two or three and share them with the team. Get explicit feedback on whether the format worked.

Do not try to async-ify decisions or brainstorms. Keep those as sync meetings. The category distinctions matter.

For the sync meetings that remain, run them with real-time transcription. That is the appropriate tool for the appropriate task. MangoFinch works best when the meeting actually needs to be a meeting.

The teams that handle multilingual communication well all end up at the same architecture: fewer, better sync meetings for the things that genuinely require real-time exchange, and a structured async layer for everything else. The translation tool is one component. The workflow design is the other half.

If you are running 12 sync meetings a week for a globally distributed team, the first fix is probably not a better translation layer. It is figuring out which 7 of those meetings should be a written update, a recorded video, or a shared document instead.