My short answer: the hardest part is not language, it is assuming the same thing works everywhere. A message that lands in the US can read as foreign in Germany and overclaiming in Turkey. Translation does not solve that; it only hides it.
1. Creative and offer do not travel
What changes between markets is not just language: the reason to buy, price perception, shipping expectations, return habits, competitive intensity. Carrying the home market's creative and offer into a new market unchanged is one of the most expensive mistakes I see — because when results come back poor, the channel gets blamed while the problem sits in the offer.
What I do instead is rewrite the offer locally for each market and produce at least one creative variant built on that market's reason to buy. The home market creative stays as the control.
2. Measurement breaks per market
Currency, tax-inclusive versus exclusive pricing, shipping thresholds, payment methods — all of these affect conversion value, and in most accounts the setup was built for one country. A Merchant Center feed can show correct prices in one market and wrong ones in another. A measurement audit on a multi-market account takes several times longer, and skipping it puts everything else on the wrong foundation.
3. Learning scatters in one account
Putting every country into one campaign is easier to manage but blurs the algorithm's learning: budget flows to wherever clicks are cheapest, and the most valuable market starves. Splitting them fragments budget, and each fragment may fall below the learning threshold.
My criterion is volume: if a market can carry the weekly conversion threshold on its own, it gets its own campaign; if not, it is grouped with similar markets. That is a numerical decision, not an emotional one, and it gets revisited as the account grows.
4. Operations and time zones
The part most people leave out: when you need to pause something in one market, the team there may be asleep. Unless you write down the approval chain, who is allowed to stop what, and the reporting hours, a technically well-built account stays operationally slow.