Each company uses its own reference populations, databases and algorithms, so it is normal for assignments to vary. This does not necessarily mean that any company is wrong: each takes a different approach.
Small percentages of unexpected populations may also appear due to the statistical margin of the method. These trace percentages are not significant.