Listening to Hugo Larochelle, researcher at Google Brain, talk about the relationship between artificial intelligence, deep learning and the, apparently, much-maligned neural networks, two things become clear. First he is super smart. Like PhD+ smart. Like unattainably smart. The second is that Canada is where all the neat AI and deep learning breakthroughs happened – speech recognition that actually works, image recognition and more.
When I challenged him jokingly on the Canadian thing, I was expecting a light-hearted response like the character Chekhov in the original Star Trek series who always claimed Russia came up with everything – “It was inwented by a leetle old lady in St. Petersburg.”
Hugo is serious and backs up his claim about the Canadian heritage via smart folks like Geoffrey Hinton and others (see a great story from MIT Technology Review on Hinton and Deep learning here ) Then Hugo gave us a “deep learning” demo-for-dummies. It was almost impenetrable. What he took for obvious and self-evident gave me a headache. Still, it left me with a clear sense that deep learning and AI will radically change many aspects of marketing someday.
What is Deep Learning
With all apologies to Hugo, my simplified take is that deep learning is computers using human brain-like neural networks to find or apply meaning to data. Here’s how Bernard Marr (@BernardMarr ) puts it:
“…it’s probably most helpful to think of Deep Learning as the cutting-edge of the cutting-edge. ML (machine learning) takes some of the core ideas of AI and focuses them on solving real-world problems with neural networks designed to mimic our own decision-making. Deep Learning focuses even more narrowly on a subset of ML tools and techniques, and applies them to solving just about any problem which requires “thought” – human or artificial.”
Four Ways Deep Learning May Change Marketing
Pulling meaning from quantitative and qualitative customer research
Think about the last quantitative survey you did and the effort to derive insights and meaning from it. What a laborious process. There’s all the mechanical procedures and strategic choices that go into creating the test instrument and then administering it. At the end of the day, humans review that data and assign meaning to it.
Deep learning and AI may not only be able to sort through the data to more insightful and defensible insights, it may be able to detect better inputs than self-reported responses. Hershey did this via Affectiva with their smile-detector – technology that reads and analyzes human emotion from a single expression, in this case a smile. AI or AI-like image recognition played a role here to better understand the nature of the smile delivered.
Understanding customer intent more precisely
Think about how we use search engine optimization to understand intent and then deliver something valuable. By sorting through keywords, phrases and the complete questions people now type into Google, we can draw conclusions about what the person is trying to accomplish. Pretty crude and straightforward.
With deep learning and big data sources from search engines, social media and where people go online next, it may be possible to better understand (infer?) what people are trying to accomplish. Marketers can then be more effective at serving their needs. How to accomplish this and stay on the better side of privacy is key as is the potential misuse of this hypothetical knowledge.
Deep learning would be seeking “patterns within patterns” of data over time. Rather than being able to recognize visual or auditory patterns, this might give us a decipherable view of an elongated customer journey. What happens over the course of 4 months, 8 months or two years as someone approaches buying a new house, car or, for that matter, insurance? Deep learning may finally make the complex and long buyer journey into a navigable course where marketers can reliably be more useful to shoppers at different points along that journey.
Preference recommendation systems
We all experience preference recommendations today. Amazon product recommendations, Spotify’s related artists, publishers suggestions of related content are examples of recommendations fueled in part by AI.
“Like Flipboard, other publishers around the world are tapping into AI to improve their content recommendations. Some 60% of the 184 publishers in the Reuters Institute poll said they are using AI to improve content recommendations. Significant numbers were also using it to automate workflows and to target ads. (Flipboard, too, uses AI for ad targeting.)” (see more)
Current systems seem to be blunt instruments. How often do you take a Netflix movie recommendation? Still, they are terrific discovery “engines.” I have discovered plenty of music artists via iTunes recommendations. If I am in a ‘discovery’ mood – willing to click around a bunch – this can be a much better alternative to randomly wading into the haystack of online music.
Deep learning may provide ever-more precise recommendations as multiple data sources combine together on individual users (blinded, of course). Browsing history, demographics, geolocation, past shopping history, content choices, other users’ behavior – may all be combined in some way that provides ever-better recommendations via constant learning.
Deeper knowledge will also lead to more personalized messaging about recommendations. Product choices can be presented with just the right headline that hits on the concerns of that individual. (image from Adelyn Zhou's great presentation)
Fuel effectiveness and efficiency in account based marketing
The two super-powers of account based marketing (ABM) is reaching the individuals from a particular company or organization and delivering personalized messaging that drives action.
A third super-power is finding ‘look-a-likes’ based upon the original prospect list and a ton of relevant data.
Deep learning may allow marketers to analyze and growing data set of the users from a particular account/company, understand their individual intent and deliver the right messages and content that meets that person’s need. This goes beyond wedging the company name into the banner ad as a nod to personalization. Of course, these messages and the sequencing of content would all be informed by what causes a prospect to convert.
Final word
How much of any of these applications are fueled by artificial intelligence, writ-large, versus machine learning versus the particular capabilities of deep learning? I don’t really know. I would have to ask Hugo Larochelle and I wasn’t quick enough to put that question together when I saw him.
AI, ML, and DL will change marketing. Time to learn.
AI resources:
AI Can Drive New Insights to Drive SEO
10 Ways Machine Learning is Revolutionizing Marketing
AI, Machine Learning and Their Application for Growth
What Marketers Can Expect from AI in 2018
AI - Closer to Relevance at Scale
Comments