[popinman322]

Run option A in one location, and option B in another, then see which works better? Of course the key would be to replicate the results in other locations to make sure the "better" option actually pans out.

Are there any policies that were built this way?

[NoGravitas]

In a sense, the US states are supposed to work this way. Supreme Court Justice Louis Brandeis described them as "[Laboratories of Democracy][0]". For example, the health care system governor Romney instituted in Massachusetts (Romneycare) was moderately successful, and was adapted to nationwide use as the ACA (Obamacare).

However, you can never really control for differences between the states, which means weak predictive value, and it's easy to argue about results. And given that the policies we're talking about are inherently political, facts and evidence don't necessarily contribute to policymaking.

[0]: https://en.wikipedia.org/wiki/Laboratories_of_democracy