Ask a company with a multilingual website how translation gets done and one answer comes back more than any other: we handle it internally.

It is a reasonable answer. It is also three different answers wearing the same sentence, and only one of them describes a localization team.

The three setups people mean

A dedicated localization function. Professional translators or localization managers, on payroll, with terminology assets and a defined process. This is real capability. A vendor adds capacity and language coverage here, not competence, and the decision is about elasticity rather than quality.

Bilingual employees doing it alongside another job. The regional sales lead who also maintains the German pages. The support engineer who checks the Japanese release notes. This is the most common setup by a wide margin, and it is the one people describe using the language of the first setup.

A coordinator who routes the work. Someone owns the process but does not produce the content; the actual translating happens somewhere else, often inconsistently.

The three fail in completely different ways, so an honest answer to “should we outsource” starts by naming which one you have.

When in-house is genuinely the right answer

This gets skipped in most vendor writing, which is why it is worth stating plainly. Keeping it inside is the better call when:

  • You run one or two languages and the content is stable. Coordination overhead is what external providers earn their margin on. With two languages and quarterly updates, there is not enough coordination to be worth buying.
  • The content is internal. Process documents, internal wikis, team announcements. Nobody outside reads it, so the cost of an awkward sentence is close to zero.
  • Domain knowledge outweighs writing craft. For dense engineering documentation, an engineer who speaks the language often produces something more useful than a translator who does not know the product — provided somebody edits it.
  • You need speed more than polish, and you can live with the variance that comes with it.

If your situation is on that list, a vendor will mostly add process you do not need. The honest recommendation is to keep it where it is.

The four thresholds

In-house arrangements rarely fail because someone decides they have failed. They cross a line and keep going.

Threshold Roughly where it sits What breaks first
Language count past three or four coordination becomes a job of its own — chasing versions, not translating them
Update frequency monthly or faster it stops being a project and becomes an operation, with nobody assigned to operate it
External exposure claims, contracts, safety, regulated labelling the person who translates is also the person who approves, and no one else reads it
Cost of attention the person is senior the language work is being done instead of the work they were hired for

The last one is the most expensive and the least visible, which is why it deserves its own section.

The cost with no invoice

A regional salesperson spending half a day a month on the local site costs roughly six working days a year. Nobody books those six days anywhere. They do not appear in a budget line, a project plan, or a cost centre. The work is real and the number is not recorded.

Then a vendor quote arrives with an actual figure on it, and it gets compared against a cost that has never been measured. The internal option wins that comparison automatically, because it is the only one of the two without a price attached.

This is not an argument that outsourcing is cheaper. Often it is not. It is an argument that the comparison being made is not the comparison people think they are making. Before deciding either way, the number worth having is hours actually spent — not estimated, counted for one month across everyone who touches a language version. The result is usually higher than expected, and it is frequently concentrated in one person nobody realised was carrying it.

The related risk is single-point dependency. One bilingual employee, no documented terminology, no second reader. When that person changes role or leaves, the language line does not degrade gradually — it stops, and whoever inherits it starts from nothing.

Fluency is not the qualification people assume

A native speaker will reliably catch what reads wrong. That makes them an excellent reviewer.

Writing publishable copy in a register you do not work in is a different skill. Someone fluent in Spanish who works in engineering will not necessarily write good Spanish marketing copy, for the same reason a fluent English speaker who works in engineering does not necessarily write good English marketing copy. Being a native speaker is a prerequisite, not a qualification.

This is why the most common in-house failure is not mistranslation. It is text that is accurate and reads like it was translated — technically correct, tonally wrong, and nobody internally is positioned to say so, because the person who wrote it is a colleague and the person reviewing it is the same person.

The configuration that usually wins

For most companies past the thresholds, the answer is not in-house or vendor. It is splitting the work along the line where each is actually better.

Keep review inside. Your people know the product, the customers, and what the company is not allowed to claim. That judgement cannot be bought and should not be delegated. Their time is also far better spent approving than producing — reviewing a page takes a fraction of the time writing it does.

Move production out. Volume, coverage, consistency across updates, and the terminology maintenance that nobody internally has time to do.

This configuration fails in exactly one way, and it is worth naming: if review has no named owner, no written criterion, and no decided default for what happens when nobody responds, the reviewing side becomes the bottleneck that the whole arrangement was meant to remove. That is a solvable problem, but it has to be solved deliberately — see enforcing translation review before publish.

Where this sits

This is the question before vendor selection, not part of it.

Once the decision is to bring someone in, translation agency or TMS covers the next fork — whether the gap is capacity or workflow — and what enterprise teams actually need from translation partners covers what to hold a provider to.

If the internal arrangement is holding but versions keep falling out of sync, the problem may be structural rather than a staffing one: how to prevent locale drift.

If you want to run the hours count before deciding, or you have crossed a threshold and want the split described above scoped properly, start here.