Social media influencers are big news and big business. Many brands are investing more of their marketing budgets in paying those with big followings on social media platforms – particularly Instagram, YouTube and Twitter – to actively promote their products. Influencer marketing agency Mediakix estimates that globally, brands will spend up to $8.2bn on influencers in 2019. And while that’s only a little more than 1% of total marketing spend, it’s grown from just $500m – a 16-fold increase – in just four years. 
At first, brands’ rush into funding influencers to plug their products was driven by novelty and fear of missing out. But as with all new marketing channels ushered in by the digital revolution, once novelty wears off and finance starts asking tougher questions about impact, the focus inevitably turns to return on investment. That, and whether brands and celebrity endorsers are playing by the rules, making it clear that there is a commercial relationship between the influencer and the brand. For in the U.K. market at least, the Advertising Standards Authority has started to flex its muscles on this issue. 
In this environment and in response to demand from a number of our consumer goods, telecommunications, and retail clients, Ebiquity’s Analytics practice has developed a novel application of econometric modelling to measure the impact of influencers. The ‘exam question’ that more and more of our clients have been asking is this: “If we were to spend significantly more on influencers, how would we measure and quantify whether that budget has delivered meaningful ROI?”
In building our solution to testing the commercial impact of influencers, we considered and rejected a couple of approaches that some have suggested. The first was seeking to determine whether influencer marketing has an impact on brand affinity. This is typically measured in lift studies before and after an influencer campaign runs. Although useful, we rejected this for three reasons: (i) because brand lift metrics can be ambiguous in their commercial value if they’re not directly related to sales; (ii) because stated intent is rarely the same as the reality of purchasing behaviour; and, (iii) because the level of investment in influencers is unlikely to shift the needle on these metrics in any meaningful way.
The second approach we considered but rejected was running a national test using econometric modelling. The challenge here is that consumer goods, telcos, and retailers spend so heavily on other media lines – particularly TV, often with hundreds of TV ratings (TVRs) per week – that even the most sensitive model would find it hard to detect any signal among the noise. Influencer marketing simply cannot deliver the reach and frequency of TV.
Measurement challenges such as low levels of spend, and statistical challenges such as collinearity – when variables in a model are too closely correlated – are the reasons why our team has developed and advocates for regional test and learn models, based on our rigorous, proprietary testing tool TestMatch™.
An option we have is to boost the posts of influencers geographically, upweighting certain regions and controlling others. When influencers create posts and promote or boost them through paid social, those posts can – in the vernacular – be geofenced; we can be sure that they only appear on the social media accounts and platforms of potential customers in specific regions. The challenge here is that this quickly becomes a test of boosted paid social, with audiences not explicitly having opted to view content, and measured activity losing some intrinsic influencer value and power.
The solution lies in regional micro-influencers, those who have a specific (or at least biased) regional audience. Working in partnership with specialist influencer marketing agency Goat, we are able to identify those influencers who have up to 10,000 followers, who are mostly based in big cities, and who have localised audiences. By contrast, influencers with hundreds of thousands or millions of followers would both break budgets and have national rather than regional influence, rendering the test and control methodology insensitive to impact. We’ve established that it’s regional microinfluencers who have the potential to create the natural variation we need for a meaningful test of impact.
In this way, we can create natural, organic test regions and compare them with control cells where little to no activity takes place, without jeopardising the essence of the influencer activity.
Early tests of our application of TestMatch™ are encouraging – at least in econometric terms. We can determine if, when, and where micro-influencers have impact. And although that impact is often not detectable or only very small, what we’ve created already is a proof of principle that can detect the magnitude of impact with the same degree of rigour that we apply to other channels.
Influencer marketing has historically been a world of hype, snake oil, and vanity metrics. Applying the established principles of econometric modelling and the rigour of TestMatch™ to measuring the impact of regional, micro-influencers cuts through that hype. It also gives brands a much better and more realistic view of the power of influencers to shift brand performance.
What we’ve seen so far is that influencer marketing is typically more about marginal gains than wholesale transformation. But in the ever-more competitive, always-evolving digital marketing ecosystem, even marginal gains can be important. Influencer marketing doesn’t yet seem to be the magic bullet to brand success that many claim, but being able to judge with data rather than instinct is a significant step forward for brands.