In my 25 years of working with large datasets, from developing early machine learning algorithms for multimedia systems in the 1990s to optimizing the email marketing infrastructure at GSI Commerce in the 2000s and now applying machine learning to big data to find actionable insights in real time, I've seen the convergence of machine learning and marketing firsthand.
This year, I'm excited to see how machine learning (ML), an artificial intelligence (AI) discipline geared toward the technological development of human knowledge, has impacted the marketing big data ecosystem. I'm also intrigued by how much room I see for growth in the future.
Machine learning techniques are being used to solve many diverse problems, and we stand to benefit as we move towards a world of hyper-converged data, channels, content, and context -- having the right conversation at the right time with the right person in the right way. For us marketers, ML is about finding nuggets of "predictive" knowledge in the waves of structured and unstructured data.
Prognosticators in 2016 picked up on the trends around natural language interfaces like IBM Watson, data growth acceleration with new sources like the internet of things (IoT), and maturation of technologies that make ML techniques efficient such as Kafka and Spark. I don't see any of these trends slowing down, as they represent the infrastructure that supports what we'll see in 2017 and beyond.
So what areas of the marketing big data ecosystem I we see being impacted by machine learning in 2017? I believe we will see a focus in four major areas:
- Automated data visualization (including ML results) will become more rich, and user-friendly.
- Content analysis (textual, lexical, multimedia/rich) will be used to drive better marketing conversations.
- Incremental ML techniques will become more prevalent, leading to real-time, not just on-going and automated, changes in marketing execution.
- Learning from ML results will accelerate the growth and skills of marketing professionals.
Automated Data Visualization
Being able to visualize relationships in data drives confidence. Confidence drives decision making, which in turn drives execution. Current tools, such as Tableau and Qlikview, give a rich palette of data visualization widgets that can be applied to structured and unstructured data. The challenge here is not one of structure, but one of understanding. Visualization tools are most effective when you understand the underlying data.
ML techniques find boundaries in data, which often belie our initial understanding, so finding the best visual representation is usually trial and error. Without the visual representation, confidence in the ML findings isn't as strong as it should be. In 2017, I think we'll see work on visualization automation -- and companies choosing the right widget for displaying its ML results -- increase substantially.
Content Analysis And Management
As marketers strive to engage in more meaningful conversations with their audience, understanding which words, phrases, sentences and even content formats resonate with particular audience members is key. Last year we saw progress in lexical analysis with the goal of finding content or text that drove overall marketing success. It did this by analyzing successful campaign content versus unsuccessful content. I believe 2017 will see that work get personalized by combining content analysis at the campaign level with content analysis at the individual level. The interconnected data makes it possible.
A simple example: A hypothetical A/B test can show that after a testing period for a campaign, the audience interacts best when “Welcome” is used instead of “Greetings.” Machine learning algorithms can be applied to the data from all campaigns to deduce the best textual introduction for emails sent to an audience, or even to an individual. This predictive power of data is incredible.
Incremental Machine Learning
Predictive models are best constructed and trained using the largest set of available data. Automation is used to generate a new predictive model from the currently available data and then deploy that new model. The challenge is that this work is performed outside of the flow of new data into the environment, while the current model is getting stale. Moving forward, I think we'll see this scheduled approach combined with an incremental technique that layers in data before the next model is scheduled to go live. This increases the predictive power without introducing too much error.
What we're really talking about is the ability to modify a solution that is already in place by introducing new data rather than having to stop using the current solution before building a new model from scratch. An example would be the addition of an extra lane on a highway, instead of closing the whole road and building a new road with more lanes right next to the old one. Construction processes and technologies have modernized to accommodate the need to add lanes to existing roads. I see incremental techniques for ML moving in that direction as well.
Unleashing Machine Learning’s True Marketing Potential
As ML techniques become more widely used, they become more widely understood. And as they become more widely understood, non-technical literature will become more widely available. I think this growing repository will lead to an acceleration of subject matter expertise around machine learning for marketing that will push the industry forward. One needn’t look any further than your favorite online education mechanism like Coursera and Udemy to find great resources for acquiring ML knowledge.
