Unless things have changed dramatically in the last couple of years, those percentages are typically going to be “best guesses” based on some number of locally varying gene “flags”. It’s definitely not meant to be used like this.
I’m not even sure using it to deduce your parents would be possible, based just on that. For example let’s say we’re only talking about two areas, just to simplify things. To make it even simpler, we’ll say your parents are found to be
Mom – 40% Region A, 60% Region B
Dad – 60% Region A, 40% Region B
What would your results be? Not 50%/50%... maybe 70/30 or even 70/25/5 if there are genes both your parents are heterozygous for. Your genes aren’t an average of your parents. This can be displayed in any genetic trait, your skin tone, eye color, hair color, height, etc wouldn’t be an average of your parents and you may have a large difference to both without any outside influence as is being inferred here.
So what they ARE saying with those percentages is they have certain genes that show up in certain populations with regularity. Some, well studied populations probably have hundreds of “flags” that can indicate a connection, some lesser studied probably only have a handful. They may even be using reference populations to give you a best guess where you fit.
It could be one flag that shows a 3% difference in an area. It could be one missing from a parent to make the same difference. It could be they weight certain flags for a region and surrounding regions using some kind of overlaying vectors.
In his shoes I would track down what flags are being used and how based on the company doing the testing, but I certainly wouldn’t be shocked by a couple percentage points here and there.
ETA: here’s 23andme’s explanation of how they get similar results it’s a lot of good science that ends up with a good guesstimate.