Monday, before. A business serves one market properly and receives visitors from another that it does nothing for.
Those visitors arrive, read pages in a language many of them are working through rather than reading comfortably, and leave at a noticeably higher rate than domestic visitors. A handful enquire, in their own language, and wait two days for a reply in English.
Nobody has measured any of it, so the decision about whether to support that language is made on impression rather than on the numbers already sitting in the analytics.
Monday before: the baseline nobody recorded
Traffic from the market, enquiries received, conversion rate and revenue. Four numbers, available before anything changes, and almost never captured.
Monday after: the same four numbers, recorded
Which is the entire basis for knowing afterwards whether it worked, and takes twenty minutes to establish.
Tuesday before: the enquiry waits two days
Arriving in a language nobody reads, translated when somebody has time, answered in English on the third day.
Tuesday after: the enquiry is answered immediately
In the language it arrived in, with the routine questions handled and only the genuinely difficult part reaching a person.
Wednesday before: the visitor leaves the pricing page
At a rate well above the domestic figure, which is the clearest available signal and which nobody was looking at.
Wednesday after: the visitor asks a question instead
Because there is now something that answers in their language, and the question they had was small.
Thursday before: no follow up is possible
Nobody can write a second message in that language, so anything that went quiet stays quiet.
Thursday after: the follow up happens
Which recovers a share of enquiries that previously went nowhere purely for want of anybody able to send a second message.
Friday before: the market is judged on impression
Difficult, low quality enquiries, not worth the effort. All of which is assumption rather than measurement.
Friday after: the market is judged on numbers
Enquiries received, answered, converted and their value, against the cost of supporting the language, which is a decision rather than an opinion.
What to measure specifically
Enquiries in that language, response time, conversion rate against the domestic figure, average value and the proportion needing a person.
What not to measure
Traffic alone, which rises with any translated content and proves only that people arrived, not that anything was served.
When to judge it
After a complete buying cycle for that market, which is the discipline set out in measuring whether automation actually worked.
What the second month usually shows
Volume in the new language typically starts low and climbs, because people discover over weeks that the business now answers them. Judging it after three weeks consistently produces a decision to stop just before it would have worked.
What a negative result is worth
A market that produces nothing after a full cycle is genuinely useful to know, because it releases the effort for a different one. The point of measuring is to be able to stop as well as to continue.
Measure the enquiries, not the visits
An assistant answering in each language produces exactly the numbers needed, by language, which turns the decision about a third market into arithmetic rather than a meeting. The measurement detail sits in measuring whether the other languages are working, the choice in deciding which languages to support first, and the approach in multilingual chatbots how ai answers in your customers language. See pricing, run the numbers on the chatbot ROI calculator, and the sector view sits under small business.