Beyond translation
Decide which languages you actually owe
Selling into a market is not the same as supporting it. Start from where your customers are, what they buy, and where a language barrier is already costing you conversations — refunds handled badly, questions abandoned mid-thread, reviews written in a language nobody on the team reads.
Rank markets by the cost of getting it wrong rather than by revenue alone. A small market with complex products and strict consumer-protection rules can deserve native coverage before a larger one with simple orders.
Fluency is a service decision, not a nicety
There is a real difference between a language a customer can be served in and one they can be served well in. Complaints, cancellations and anything with a legal edge need someone who can hear tone, hedge carefully and be precise about what the company will and will not do.
Set the bar per contact reason. Order status can tolerate a machine-assisted reply. A dispute about a faulty product, in a market with statutory warranty rights, cannot.
- Native or near-native: complaints, retention, legal and regulated topics.
- Assisted: order status, delivery updates, simple account questions.
- Escalation route: every language needs a defined path to a fluent speaker.
- Written and spoken are separate commitments — do not assume both.
Give translation tools a defined job
Machine translation is genuinely useful for comprehension: letting an agent read an incoming message, or draft a first pass. It is much weaker at the things support actually turns on — idiom, politeness levels, and the exact wording of a commitment.
If translated replies go out, say so. A customer who knows they are reading a translation reads it charitably. One who believes they are talking to a fluent speaker reads every awkward phrase as carelessness.
Keep one standard across every language
The same question should get the same answer in every market. That means one source of truth for policies and macros, translated deliberately, rather than each language team writing its own answers and drifting apart over time.
Review quality per language, not only overall. An aggregate score is easily carried by your largest market while a smaller one quietly deteriorates.
Staff coverage around real demand
Time zones matter more than language count. A market served eight hours out of phase can wait a full day for each reply, which turns a two-message exchange into a two-day one and looks like neglect however good the answer is.
Publish the hours you cover per language, and be exact. Customers plan around stated availability; they lose patience with support that is theoretically available and practically absent.