As we can not get actual impressions from Instagram for any non-business account, the Mavrck platform estimates impressions based on published research done by Stanford University and Facebook's data science team called Quantifying the Invisible Audience in Social Networks.
The algorithm takes the follower count of the influencer, the engagements they have received, and a number of other factors into considerations, as described in the article listed above.
We are currently in the process of re-evaluating this algorithm based on recent changes made to the Instagram feed algorithm, as well as a move towards machine learning algorithms based on actual impressions captured from Influencers reporting from business accounts.
As part of our current Fraud Detection project, we are looking to make updates to account for various levels of automated and/or fake accounts that interact with the influencer. For example, if we determine that 20% of an influencer's audience is fake/automated, we would adjust the estimated impressions accordingly.