“We’re using AI to give creativity back to the marketers.” So said Market Logic CEO Kay Iversen this week at I-COM, the industry forum for Marketing Data & Measurement Strategy.
Kay was invited to address I-COM’s Chief Analytics Officer Council about AI best practices to organize research and help the industry move forward. The Council tasked him to share practical examples of the ways marketers are using AI, as a follow up on their broader discussions about AI opportunities generally.
And he did so, with a rapid-fire description of our own marketing insights platform, and three hot applications: cognitive search, self-service intelligence and cognitive marketing assistants.
Cognitive search
The ability to ask a question and get the answers from all your unstructured research and data (primary, secondary, social) is extremely important.
Kay explained how Unilever has unleashed the power of cognitive computing to do this on their PeopleWorld insight engine (see Harvard Business Review)
This involves teaching the machine the language of marketing with examples so that document miners can analyze text and annotate entities and fundamentals (needs, segments, functional values, emotional associations, RTBs, etc.) that can then be connected in the Market Logic knowledge graph.
Then, discovery miners can classify information, and the user can access all those insights in on-the-fly generated dossiers which summarize those for easy consumption
Self-service intelligence
The first Market Logic clients are now leveraging AI to deliver personalized, low-noise dashboards to marketers with self-service news and topic feeds, so there’s no need for any hand-holding by the central market intelligence team.
This self-service intelligence experience is enabled by unsupervised learning over unstructured data from secondary sources and RSS feeds – where the algorithms automatically figure out from looking at all content, what topics exist, while a user (by selecting a few news items of interest) can tell the machine what sorts of topics they are interested in.
This process of understanding the topic fingerprints across all news feeds is done by discovery miners, so that a news module can simply fetch all similar news items per topic, without having to rely on the exact words that appear in each of the news feed items – which, as we all know, can change any minute.
With this setup, noise in automatically filtered newsfeeds is reduced by over 90% and does not require a user to constantly modify the topic selection – the machine solves that automatically.
Cognitive marketing assistant
The ultimate application we see on the market and a likely reality for corporate marketing teams by 2020. Cognitive marketing assistants understand what the marketer wants to achieve from the questions they ask and topics they discuss, recommend the company’s best practice process, step by step, and suggest the right information at the right time in the process.
This entails reinforcement learning, where the machine learns how to solve a business problem by analyzing implicit positive and negative feedback, the complete AI infrastructure mentioned above, and cognitive services to guide the user to the right result with a conversational interface. Watch this space!
Q&A session
In Q&A, discussion amongst council members from iHeart Media, Intel and Dentsu focused on the deployment effort needed to get these applications up and running.
Here, Kay emphasized the importance of going broad with data inputs across all sources ASAP, and the typical 12-week supervised learning cycle to teach the machine the language of a business.