We can't know the real conversion rate of B unfortunately. Yet if this real conversion rate of B is in fact equal to the conversion rate of A or smaller than A's, then over 10% of the experiments of the same size would show as extreme fluctuation as we observe now. It is not rare enough to conclude that B converts better than A.
If we base our business decisions on many similar tests (we implement option B when it wins with the significance level 90%), we will be making wrong decisions once in a while yet we can limit the rate of wrong decisions by 10% (it could be smaller, but we can't guarantee it). Apparently we consider such risk of wrong decision being too high.
There is a possibility that B does convert better than A, but improvement is so small that this test failed to notice it with the current number of trials (test had too little test power). Use test power calculators to estimate the number of trials needed to detect the improvement of this or that size reliably.