When filmmaker and photographer Sam Ciurdar was hired to post about his experience with the Mazda brand at SXSW this year, he never actually drove the sporty CX-5. But that was OK with him.
"It didn't make sense at SXSW because it's all events. There wasn't a lot of driving involved," he said.
He will be behind the wheel of one of the new Mazda crossover SUVs this weekend, however, when he drives one in the LA area as part of another social media influencer campaign. Ciurdar was one of a handful of people chosen by Influential, an agency that pairs brand clients with so-called influencers, with the aid of data analysis and natural language processing.
As brands burn out on celebrity endorsements and the legion of social media power-personalities multiplies, it's become increasingly difficult to find so-called influencers for their campaigns. A handful of tech-centric agencies and platforms are using artificial intelligence-informed approaches – similar to those employed to model audience segments for ad targeting or enable chatbots -- to pluck potential influencer mates from the dating pool on behalf of brands like Mazda and Gerber.
Using artificial intelligence and machine learning to discover and link social media's popular kids with brands "is not something the majority of influencer marketers are doing," said Kristin Hersant, VP of marketing at Linqia, an influencer marketing firm that has built various components of its data-crunching tech platform in-house over the past couple years.
AI is "definitely a sexy thing to talk about," said Jonathan Pollack, Linqia's VP of product.
Influential runs on IBM's Watson AI technology. Mazda enlisted Influential to help brands find influencers to to send to SXSW to promote the new CX-5. The technology parsed the posts in the social ecosystem to determine common words used alongside the Mazda brand to find appropriate influencers. The brand was looking for artsy extroverts with a flair for excitement -- signaled by exclamation points and emojis, for example.
For the SXSW effort, four Mazda influencers were selected to cruise around Austin in the CX-5 and hang out in a branded Mazda Studio, then post about the experience on Twitter, Instagram and Facebook. The data-driven approach gave Mazda the ability to refine its influencer marketing targeting in a way that aligned better with the brand's values, suggested Eric Watson, director, marketing operations at Mazda North American Operations.
Now, Mazda is finalizing the traits it will seek in a new group of influencers who will participate in ride-and-drive events planned through July in cities such as Dallas, Los Angeles and Philadelphia.
Ryan Detert, CEO of Influential, considers the company's data-centric search process for finding brand matches the "opposite" of a traditional talent search through agencies like William Morris or Creative Artists Agency. Rather than charging clients based on the number of followers an influencer has, or on a fee-per-post basis, Influential uses metrics reflecting engagement from among the targeted audience defined by brand clients to determine cost.
Influential said payment per influencer can range from $500 to $10,000 or more, based on a stock price that changes dependent on Influential's brand match score. That score is determined by the demographic, contextual and psychographic relevance between the influencer and brand.
Better Matches, but Risks Remain
Big-name celebrity-focused brand endorsements are under scrutiny from regulators. The Federal Trade Commission sent letters to 45 celebrities including Jennifer Lopez and Sean "Diddy" Combs alleging that they promoted brands in social media without disclosing paid relationships with the advertisers. J-Lo, for example, posted a photo of herself in a shimmery gown perched on a table alongside bottles of Beluga Vodka on Instagram, thanking the brand.
Not only did the post catch the eye of watchdogs who complained to the FTC, it raised questions from her social media followers who wondered, why someone who had previously stated she does not drink alcohol would appear to be endorsing a booze brand. "Thought you didn't drink alcohol? It was one of the reasons you're aging well," wrote one commenter.
"We realize that some of the most engaged influencers work with various brands and we completely understand and respect this," said Mazda's Watson. "In fact, the showcase of an older vehicle that they love to drive shows that they share similar attributes with our target audience."
Because brand clients sometimes have a nebulous idea of the types of people they'd like to work with, incorporating data analysis can help pinpoint people or groups with similar characteristics once there's a general idea, said Linqia's Pollack.
"The way they articulate who they are looking for is very subjective," he said, regarding some clients. "They'll say things like 'edgy' or 'fashionable.' It's very hard to build search functions based on those criteria."
Linqia applied the approach to find influencer matches for its client Gerber, to help drive awareness of its Lil' Beanies product. People chosen for the campaign posted stories and photos of their kids nibbling on the snack made from navy beans.
eHarmony for Brands
Creative tech consultancy Ayzenberg struggled to manually identify people for influencer marketing efforts and worried that when they did, they didn't have a strong sense of who those people really were. So, when the firm decided it was time to apply data and technology to the process, they went straight to the source of digital matchmaking: Dr. J. Galen Buckwalter, a behavioral scientist and the original chief scientist of eHarmony, the online dating service that famously promises to foster long-term relationships based on 29 dimensions of personality.
"What we did at eHarmony was pretty much exactly what we're doing here, at least on the psychometric perspective," said Buckwalter, who helped the consultancy develop its Soulmates.ai technology.
The firm currently uses the technology internally to surface people who -- even if they don't typically work with advertisers or post sponsored content -- might be an appropriate companion for a brand.
To uncover potential partners for brand clients, many of which are in the gaming industry, the system deciphers patterns in language used by Instagram, Facebook and Twitter users with relatively robust followings (usually a minimum of 10,000 followers) to determine whether they fall into categories associated with personality traits sought out by the advertiser.
The algorithm factors in six traits: honesty and humility, emotionality, extraversion, agreeableness, conscientiousness, and openness to experience (based on the "HEXACO" personality inventory). So, someone posting a photo of packed luggage with the caption, "I spent all night getting suitcases organized to go on my trip!" might be considered conscientious, for example. Someone else with a similar vacation packing related photo accompanied with the text, "Threw my clothes together – almost missed my flight!" might not.
Buckwalter and the team labeled thousands of words according to that category structure in order to determine whether a social power-user's persona makes sense for a brand. Rather than dumping a database of language from social posts into a learning system, Ayzenberg "used the existing understanding that science has of personality and emotions and of values and used that to structure our system as we developed it," said Buckwalter.
It also lets the agency track the status of negotiations with social influencers, view their past content and estimate the projected number of views they might get if they mention a brand in a sponsored post.
"We're looking into the people that have a very authentic following who are just a good match," said Kai Mildenberger, Ayzenberg's chief technology officer. "Not because they do it because they want to make money with it."