Multilingual meetings have a minutes problem

A German engineering firm holds a quarterly review. Seven people, four first languages, one working language. The meeting runs ninety minutes and ends in apparent agreement.

Six weeks later two of the participants describe the decision differently, and both are certain. Someone opens the minutes. The minutes support one of them.

What almost nobody in the room thinks to ask is who wrote them.

The record is written by whoever is most fluent

This is the structural fact that European businesses have absorbed without ever examining.

In a meeting conducted in a second language for most participants, the written record is produced by the person who can produce it fastest. That is almost always the most fluent speaker present, frequently a native speaker, occasionally the most senior person, and essentially never the participant who was working hardest to follow.

Minutes are not a recording. They are a summary, and a summary is an interpretation. Whoever writes one decides what counted as a decision, which contribution was substantive and which was an aside, and what the meeting was fundamentally about. Those judgements are made through the filter of one person’s comprehension.

So the participant who made a careful point in halting English becomes a subordinate clause. The participant who made a mediocre point fluently becomes a paragraph. Nobody intended this and it happens in most multilingual organisations every week.

Over a quarter it compounds into something with a real cost: decisions that get relitigated, commitments that half the room does not recognise, and a slow drift in which the strategy the company believes it agreed on is the strategy the most fluent people understood.

The research has been pointing at this for years

Work on multinational teams has established the participation half of the problem fairly conclusively. Research on corporate language competence has documented how language barriers, both evident and hidden, suppress participation and sense-making in multinational teams, and how native speakers can end up dominating discussion conducted in a shared language, narrowing the range of input a company actually receives.

The documentation half attracts less attention, which is odd, because it is the half that persists.

A study of collaboration tooling for non-native English speakers surveyed 64 researchers working in a second language and grouped their difficulties into four areas: participation, comprehension, documentation and feedback. Documentation sits in that list as a distinct problem, not a downstream symptom of the other three. Being unable to fully verify what the written record says happened is its own barrier.

A meeting is transient. A record is an asset, and it is the version the organisation acts on.

Reading is not listening

Here is the part that changes the calculation, and it is a fact about language acquisition rather than about software.

For most people working in a second language, reading comprehension substantially outstrips real-time listening comprehension in a group setting. Reading is self-paced. You can stop, re-read the sentence, look up the term, and construct the meaning without a room waiting for you.

Live discussion offers none of that. It arrives at native speed, overlaps, uses idiom, and carries a social penalty for asking someone to repeat themselves a third time. Participants routinely follow perhaps seventy per cent of a fast exchange and reconstruct the rest afterwards from context.

Which means a verbatim transcript is not merely a convenience for these participants. It is the first version of the meeting they can fully understand.

That reframes what transcription is for in a cross-border business. It is not a productivity tool that saves someone the chore of typing notes. It is the mechanism by which everyone in the meeting gets access to the same meeting.

What this looks like in practice

The operational change is small, which is the main argument for it.

Record the meeting. Tools such as VOMO AI handle roughly 50 languages, label speakers automatically, and return a transcript alongside a structured summary with timestamped chapters, key points and action items. Accuracy sits around 95 per cent on clean audio. There is no cap on the length of a single recording, so a three-hour board session does not need splitting.

Three things follow from having that artefact, and only the third is obvious.

The summary stops being authored. It is generated from the full text rather than filtered through one participant’s comprehension and priorities. Anyone who disagrees with the summary can check it against what was actually said, which is a materially different situation from disagreeing with a colleague’s memory.

The transcript becomes searchable. Six weeks later, the question is not whose recollection is more confident. It is a search for a term across the record, which takes fifteen seconds and ends the argument. Some tools also allow direct questioning of the transcript, which on a long session is faster than reading.

Participants can review at their own pace. The person who followed seventy per cent in the room reads a hundred per cent that evening, and can raise a correction before the decision hardens. This is where the participation gap actually narrows, and it happens after the meeting rather than during it.

Speaker labels need a check on anything with more than three participants, particularly where people talk over each other. Ten minutes of cleanup on a long session, and the record is quotable.

The record travels further than the meeting

There is a second-order effect here that rarely gets connected to the first, and it is the one with balance-sheet consequences.

Most people who act on a meeting were not in it. Regional heads read summaries of reviews they did not attend. Executive committees read summaries of summaries. In regulated sectors, auditors and supervisors read the documented record years later, and in a dispute the minutes are evidence in a way that anyone’s recollection is not.

Every one of those readers is receiving a version of events shaped by the language competence of whoever typed it up. The filtering that happened in a conference room in Rotterdam propagates through the organisation and hardens at each stage, because each subsequent summary is made from the previous summary rather than from the meeting.

The practical result is that a firm’s senior management can hold a systematically distorted picture of its own operations, weighted toward the parts of the business staffed by confident speakers of the working language. Nobody misled anyone. The distortion is entirely an artefact of documentation.

For companies with significant operations outside their headquarters language, this is worth treating as an information-quality issue rather than an HR one. The question to put to a management team is straightforward: when the board reads a summary of a meeting held in English among people who mostly do not speak English natively, how many interpretive layers sit between the board and what was actually said, and who built each one?

In most European groups the honest answer is three, and nobody knows.

Retention becomes a live question

One consequence European firms should think about before rolling this out rather than after.

A recording and a transcript are personal data. They are subject to GDPR in the same way any other record of an identified individual is, and creating them at scale creates a retention obligation that most companies have not written down.

The baseline requirements are unremarkable: encryption in transit and at rest, a defined retention period, deletion that actually deletes, and an answer to what happens when a participant asks what is held about them. Providers vary considerably in how clearly they document this. It is worth establishing before the first recording rather than during the first complaint.

Works councils will also want to be consulted in jurisdictions where that applies, and in several European countries that is not a courtesy. Recording a meeting where employees are speaking is a different proposition from recording a customer webinar, and firms operating in Germany, France and the Nordics should expect to negotiate the policy rather than announce it.

What it does not fix

Transcription addresses the record. It does not address the room.

A participant who could not get a word in still could not get a word in, and now there is documentation of that. The fluent still dominate live discussion; the transcript simply makes the imbalance legible, which is uncomfortable and probably useful.

Nor does it repair meetings that were badly run. A ninety-minute session with no agenda produces a ninety-minute transcript with no agenda, and the structured summary of a rambling meeting is a shorter rambling meeting.

Cost is not the obstacle it once was. Free tiers cover roughly 30 minutes of transcription a week, which is one meeting, and unlimited plans start under two euros a week per user. The barrier to adoption in most organisations is policy and habit, not budget.

The strategic reading

European companies compete with a structural handicap and a structural advantage that are the same feature. Operating across four or five languages brings market access that monolingual competitors cannot replicate, and it introduces friction into every internal process that depends on shared understanding.

Most firms have optimised the customer-facing half of that trade and left the internal half to individual effort, on the assumption that everyone speaks enough English for it to work out.

It mostly does work out. The cost shows up in the margin: decisions taken twice, expertise that never reaches the record, and a documented history of the company written predominantly by the people who happened to find the working language easiest.

That is a fixable problem, and the fix is not a language policy. It is a better record.