Explore how Influential integrates Watson services and Bluemix into its influence marketing platform with plenty of developer insights from Piotr Tomasik, CTO and Co-founder of Influential.

In this video:

Influential does influence marketing – the company analyzes influencers psychographically, demographically, and contextually to try to match them with the best brand – and it uses Watson and AI in its Influencer One platform to make the matches on Instagram, Facebook, and Twitter. Similar to what Nielsen or Arbitron does, Influential essentially provides brand managers with the ability to hit their demographic target when they place an ad insertion order. They’ve created an invite-only network of more than 10,000 leading social influencer accounts on Twitter, Instagram, Snapchat, YouTube, and Facebook, people who are in the top 1 percent of engagement on each platform and who collectively reach over 5 billion followers.

The Watson part in all of this comes in helping train the machine learning components of the platform to recognize the influencers who have an audience with a passion for what an individual brand or advertising campaign represents, then to refine that recognition ability through the processing of lots of incoming, often real-time and unstructured, data. Agencies can monitor the campaign as it happens across three key dimensions – demographics, psychographics, and context – and make adjustments to their strategies and messaging on the fly.

With Watson, the system can:

  • See how an influencer matches with a brand on a score of 0-100, based on demographics, psychographics, and contextual relevance
  • Employ a technology that identifies personality traits and archetypes
  • Engage a system that checks billions of data points to determine whether an influencer has used profanity, referenced a competitor, is talking about a brand organically, or even perform felony/DUI checks

Piotr gives an example of Sony Pictures’s movie The Shallows:

“We went into our network, found people who’ve talked positively about horror movies and cross-referenced this with those following our network of influencers. That was through Bluemix.”

Piotr discusses the pieces of Watson that are in Influential’s product. He notes that the key phrase to remember is enrichment: “That’s what we use the Watson APIs for – enrichment.” Top of the list is Personality Insights. The company takes the entire corpus of tweets from an influencer and runs it through Personality Insights to get the 52 traits and consumer behaviors results back.

The platform also uses AlchemyLanguage’s (now Natural Language Understanding) keywords, taxonomy, and concepts components. They combine the two sets of results:

“Personality Insights gives us a great understanding of the influencer’s personality and then the NLU components help us associate that influencer with certain concepts. Agencies really really like to know if an influencer will be talking about a topic that matches their brand.”

Piotr acknowledges that AI allows them to start thinking of traditional demographics, such as the age of the audience, as the starting point of an effective analysis. One you identify the baseline demo you want to present a message to, then you look at the psychographics – personality, values, opinions, attitudes, interests, and lifestyles – to further refine your campaign’s targeting and messages.

Piotr and Ryan also discuss what makes a good, innovative developer (someone who’s flexible with changes in direction) and emerging coding languages and practices like

  • Node.js and Javascript,
  • ELM, a domain-specific programming language for declaratively creating web browser-based GUIs that has an explicit-type compiler that eliminates faith-based coding
  • Microservices over monolithic APIs

Piotr also discusses dealing with data. The unstructured data lives in ElasticSearch, RDB-type data in MySQL variant MariaDB, and Redis – the in-memory open-source database that implements a networked, in-memory key-value store with optional durability – lets the company cache information from an influencer’s timeline immediately and start analysis.

So what’s tricky about grabbing and storing data, sending it to Bluemix and Watson services for enrichment, then getting it back to plug into the Influential platform? The trickiest part, according to Piotr, is deciding what you want to keep and what’s OK to let go because if you’re trying to track hundreds of thousands of social accounts, there’s a fair bit of noise in there. Also, you have to consider how to keep the data you retain performant.

Resources for you

Cognitive Computing

Create apps that accelerate, enhance, and scale the human experience.

Join The Discussion

Your email address will not be published. Required fields are marked *