I'd say that it depends very much on where you are in the project and what time pressures exist on the current development. Also, how likely is the need?

If you are at the beginning/design phase of the project, make finding/designing your i18n library he immediate priority, remembering that simple is best, and have your developers get used to specifying text in terms of message numbers and using the appropriate apis for outputting them etc., from the get go.

If your project is well along, under time pressure or in the mintainance phase, realise that conversion to i18n is a major undertaking, and defer it for a major release. Fork your codebase and run the i18n refactor in parallel to the main thread with one senior developer and as many juniors as you can spare/get budget for. Only merge it back to your main thread once you have a good idea of what the effects and costs are going to be and how much change is going to be required.

If the requirement is only speculative, still do the fork and put a senior developer in charge, but have him consentrate on costing the change rather than doing it, with a view to presenting the speculating management with some hard, profit/costs-based information upon which to base a decision.