We take a combination of multiple variables into account when calculating author profiles, including, for example, existing person IDs, affiliation data, and years. Person disambiguation in Dimensions is naturally conservative, meaning that where there is any doubt about whether two name (and related affiliation) variants are the same or not, we will not merge them, as we believe it is better to try and ensure that two people are not combined when they should not be, even if this means that a researcher may end up with several profiles in Dimensions, or that some publications are not matched to any profile.
The technology and data sources we are using to disambiguate authors are improving all the time. A well maintained ORCID profile - which needs to be set to 'public' - significantly increases the probability that related profiles will automatically be merged in the future, as ORCID data is taken into account when performing the disambiguation process.
If you are searching for a person in Dimensions and find that there is more than one researcher profile for that person you can combine these for the purposes of your search by using the "OR" search functionality. Simply select the circular check-box next to each researcher in the search filter, and then click "limit to" at the bottom of the filters (see screenshot).