Guide. Everyone translates their own way: organising legal translation in a Swiss law firm

By Neur.on's legal and language team · 15 September 2026 · Updated on 15 September 2026

A Swiss law firm translates every day without realising it. A partner reads a judgment from German-speaking Switzerland with an online tool. An associate retypes in French a clause received in Italian. A trainee prepares an English version of a memo for a foreign client. None of these operations appears in a budget, and all of them commit the firm.

1. Taking stock: count before you decide

Any attempt to organise translation that starts with the choice of a tool fails. It fails because it answers a question no one has asked, and because it ignores the real volume. Start with a survey, over four weeks, without judgement and without sanctions.

Ask each person to record, for every translation done or ordered: the date, the language pair, the type of document, the final recipient, the tool or provider used, and an estimate of the time spent. A shared spreadsheet is enough. The aim is not accounting precision: it is to reveal an order of magnitude.

Three findings come up almost every time.

• In-house work dominates. The volume entrusted to a provider, the part the firm knows about, is only the visible part. The bulk is produced internally, in a hurry, and billed nowhere.
• The tools are a patchwork. Three or four different services coexist, chosen out of individual habit rather than by decision.
• No one knows what has already been translated. The same clauses, the same definitions, the same general terms and conditions are translated again from one matter to the next.

2. Mapping needs: understand, produce, file

The quality of a translation cannot be discussed in the abstract. It is discussed in relation to a recipient. So classify your needs into three families, and a single question is enough to place any document: who is going to read it?

This grid immediately makes clear something that internal debates blur: most of the volume belongs to the first family, which needs neither stylistic review nor certification. It is also the family that, for lack of a tool, ends up in unauthorised services. The second family is where a law firm is judged. The third is rare and costly, and it has to be planned.

Then add the expected review level to this grid. The three levels of external review commonly offered map onto it naturally: a light review to understand and circulate internally, a full review for what leaves the firm, a certified double review for what is filed. These levels are explained in detail in our article on reading a quote.

3. Who decides what

The most frequent bottleneck is not technical: it is about decisions. Anyone can request a translation, no one is responsible for the level chosen, and the trade-off is made over the phone, matter by matter. Three roles are enough to get out of this.

The partner or legal expert responsible for the matter

They alone know the recipient and the risk. They therefore decide the family of need and the review level, and they approve the terminology specific to the matter. So that this decision costs them nothing, the written rule must give them a default answer: they only decide the exceptions.

The secretariat or case assistant

They carry out the work: uploading files, choosing the target language, tracking deadlines, retrieving and filing the deliverable. They do not decide the review level, and they do not have to interpret a confidentiality policy. Give them a procedure, not a judgement call.

The translation coordinator

One person, not a committee. They keep the firm's glossary, track the indicators, settle terminology disagreements that cut across matters and remain the provider's contact. In a medium-sized firm, the role takes a few hours a month, and its absence is the main reason an otherwise well-designed process is abandoned.

4. The rules to write, and those not to write

A useful translation policy fits on one side of a page. Beyond that, it is not read, so it does not exist. Here is what it must contain.

1. The default family of need and its review level, for the five or six most frequent types of document at the firm.
2. The confidentiality mode to use according to the sensitivity of the matter, worded as a choice and not as a prohibition: in this case, the strictest mode; in that case, the standard mode.
3. The approved tool, named, with the name of the person who grants access.
4. The terminology approval process: who decides, how quickly, and where the decision is recorded.
5. The minimum delivery time to allow for each family of need, so that translation stops being an emergency every time.
6. The point of contact when a case does not fit any box.

What not to write is just as important. Do not write a list of prohibited tools: it is out of date the following month, and it turns the policy into a policing document. Do not write a rule that starts with ‘it is strictly prohibited to’ without saying straight away what to do instead. And do not write a numerical threshold that no one will be able to check.

5. Stopping shadow translation

Shadow translation means pasting an extract from a case file into an online service that has not been approved. It is a law firm's most concrete translation risk, and sanctions do not solve it, because it stems from a legitimate need: to understand quickly.

Three levers work, in this order.

• An approved tool at least as fast as the alternative. If the firm's tool requires a written request and two approvals to translate a paragraph, it will not be used. The comparison is not with a provider: it is with a free service open in a browser tab.
• The rule displayed at the moment of choice. A policy shown in the interface, at the moment the user selects an engine, gets read; the same policy on an intranet does not. We describe this mechanism in our article on where data is processed.
• An explicit way out. For cases where the tool is not enough, say whom to call. The lack of a documented fallback is what drives people to improvise.

A fourth lever, often neglected: explain once, seriously, why the question arises. The difference between storage and processing, what pseudonymisation really covers, the notion of subcontracting: a thirty-minute presentation to the whole firm does more than a year of email reminders.

6. Building on terminology

A law firm keeps translating the same things. The definitions in its contract templates, the headings of its usual clauses, the names of its institutional clients, the statutory references of its core practice areas. Each new translation is lost time and a chance for inconsistency.

Building on terminology does not need a sophisticated tool to get started.

1. Open a shared file with four columns: source term, target term, official source for the decision, date. Nothing else.
2. Add to it only when a decision has been taken, not in advance. A glossary built in advance is never consulted; a glossary that records real decisions is.
3. Require the source. An entry without a verifiable reference will be challenged at the first disagreement, and the discussion will start all over again.
4. Review it once a quarter with the translation coordinator, to remove duplicates and resolve contradictions.

Swiss legal terms are checked against official sources: the federal law platform, which gives the same article in all three languages, published decisions and the publications of supervisory authorities. A bilingual concordancer based on these sources, such as Fast lookup CHnell, makes this check immediate rather than tedious. The reasoning behind it is set out in our article on Swiss legal multilingualism.

7. Measuring, with six indicators

A process that is not measured slides back into a sum of individual arrangements within two quarters. Six indicators are enough, recorded at the same point each quarter, on a single page.

1. Volume handled per family of need (understand, produce, file), in standard pages. It is the only unit that stays stable when formats change.
2. Share handled in the approved tool. The governance indicator par excellence: if it does not rise, the rule is not workable.
3. Actual time between request and delivery, as distinct from the delivery time announced. The gap between the two is what drives workarounds.
4. Internal review hours given a value. As long as this line is at zero, every trade-off between internal and external review is distorted.
5. Rework after delivery: number of documents sent back for correction, and the main reason. Three reasons are enough for classification: terminology, omission, formatting.
6. Share of matters handled in the strictest confidentiality mode, compared with the share of matters that required it under the written rule.

Do not track declared satisfaction: it is stable and tells you nothing. Do not track unit cost on its own either, because it falls mechanically when quality falls. Indicators 3, 4 and 5, read together, give a far more faithful picture of what translation really costs the firm.

A realistic timetable

• Month 1: survey of current practice and mapping of needs.
• Month 2: drafting the page of rules, appointing the translation coordinator, choosing the approved tool and granting access.
• Month 3: presentation to the firm, opening of the glossary, first indicator readings to serve as a baseline.
• Following quarters: readings, glossary review, adjustments to the page of rules. One hour per quarter, no more.