Ad Variations need to collect enough statistics

I made several variation for an ad I use for my most effective ad group in AdWords. Now I've found that I accidentally made a duplicate. At 23k expressions one copy has CTR 0.69% and another one – 0.56%.

So the 0.13% CTR difference is statistically irrelevant at 23k expressions. This is unexpected. How much expressions should be there before one can make conclusions about ad variation effectiveness?

PS The difference is 0.06% now at 40k expressions for each of the copies.