25 February 2019

Our Machine + Human
Approach to Combating
Influencer Fraud

Influencer fraud continues to be an industry-wide issue. Everyone from tech giants to agencies to vendors and marketers are all trying to refine their models and algorithms to help eliminate fraudulent activity. As we all work to come up with the best solution, the below is our current process on how we’re tackling the issue on our end. And, as technology and the industry become smarter, we continue to adapt and refine our process.

Machine + Human Approach

Machine: we leverage the machines, automated technology, to alert us of any suspicious patterns that come up in the data when analyzing influencers. Some of the most insightful data points like the below are already baked in within our algorithms:

    • Follower Growth: How quickly has the influencer’s audience grown? Did the growth follow normal patterns? Are growth spikes accompanied by the expected natural spikes in engagement?
    • Audience Location Percentage: Where is the influencer’s audience based? Is that distribution normal for influencers in the same region?
    • Engagement Rate: How does the influencer’s rate of likes and comments compare to the expected rates for a particular social network?

Human: We believe that people need to manually perform audits and really dig into the details in order to truly assess if anything is fraudulent. For example, a machine may alert us that the engagement levels of an influencer doubled over night and while that seems unusual and typical of fraudulent activity, when checking the influencer’s page, we may notice they were tagged by a celebrity in a post the day before and that’s what caused a huge spike in followers.

The types of things to check manually prior to/after any alerts;

    • Content – what is the quality overall? Do the engagement levels seem in line with the amount of followers they have?
    • Audience – Double click into their followers. Are they primarily located in one area vs another and if so, does it make sense? Do their areas of interest line up with the influencer’s?
    • Comment quality – Do the comments seem authentic or are they repetitive and vague

A Machine + Human approach is robust and thorough and while it can be intense in certain instances, we believe that it is truly the best way method for assessing influencer fraud in today’s environment.