How machine learning can amplify personalized marketing
By Steve Renshaw, Co-Founder, Ratio
The biggest challenge marketers face with personalized marketing is managing it at scale. Setting up accurate personalization rules relies on a mixture of data analysis, hypotheses, content curation, and manual configuration. Managing multiple rules across multiple pages for multiple scenarios—and knowing what is working—can quickly become daunting, especially with smaller marketing teams.
Machine learning adds value to personalized marketing by eliminating a lot of the guesswork. Using “self-learning” modeling to analyze successful behavior automates the process of serving relevant content—almost bypassing the need to set up rules—by surfacing targeted messaging, such as “related products” offers, on e-commerce sites based on earlier successful user behavior.
Cloud is a big factor in scaling up, providing the immense resources to manage machine learning and background processes. I’ve read about machine learning running on-premises because of data privacy concerns, but that requires serious server power. More commonly, data stores are hosted on-premises and then data is ported to a cloud platform for machine learning training and learning processes. Why make life difficult when the tools already exist?
Cloud-hosted machine learning helps scale personalized marketing
With limits on how many concurrent requests you can make to machine learning services, a machine learning based experience cloud makes it easier to scale endpoints manually or programmatically to help react to peak periods of personalized marketing traffic.
For example, retailers need to scale up and down to accommodate seasonal peaks. Plus, cloud services make machine learning more cost effective. This especially helps mid-size companies compete with larger rivals. But effectiveness depends on who sets it up and the quality of the data and data modeling provided by the tools you use and what data they are ingesting.
When it comes to machine learning, it’s all about setting parameters and then feeding it data so that it can train itself. This only works if you set up experiments intelligently based on existing analytics and user data. Done well, it will enhance your user experience efforts, giving you an edge over competitors.
Out-of-the-box machine learning for personalized marketing experiments
Machine learning doesn’t have to be complicated—Microsoft Azure provides a Recommended Products ML engine out of the box, to start running personalized marketing experiments.
But remember machine learning is a framework. So as you get smarter, there is potentially more up-front configuration. ROI will vary wildly but is more easily measured in monetary terms within a transactional scenario. It all comes down to how much thought has been given to the data, planning, objectives, and setup.
Machine learning really comes into its own when it communicates with other platforms and data stores to digest data and then share learnings back to your website or content application. So, yes, you can integrate with your existing website, automation tool, and email marketing platform. A sandbox environment is a great way to get a simple scenario running to help bring machine learning to life for the business. But at some point, you need to feed it real user behavior data—you can’t simulate that. You only see results when it’s in production. That’s why analysis on your live data is so important to make sure you’re feeding machine learning the right data from the user journeys and conversion points you want to affect.
Personalized marketing requires ongoing investment in machine learning
Personalized marketing requires ongoing investment in machine learning in order to see results. Therefore, executive sponsorship is important to ensure it doesn’t just launch then go stale. Any initial approach must directly align with business objectives.
You need a team of strategy, business analysis, data analysis, development, and project management professionals to plan and deliver a robust and measurable machine-learning-for-personalized-marketing program. Because if it’s considered a side project, you’re doomed to failure.
What role does automation play in all this? For example, one of the use cases is auto populating or changing rules programmatically based on machine learning data. A constant rule automation engine is running based on real-time data.
A completely different automation scenario uses machine learning to power bots that can answer user questions and help people perform tasks quicker. For example, inquiring about a holiday and clicking “book” all within a single chat session. Personalized marketing should enhance the user experience—not just sell products.
Growing machine learning skillsets to sustain personalized marketing projects
It could only be days to spin up a proof of concept for a chatbot or machine learning recommendations engine using out-of-the-box tools and templates. However, as you get smarter, it’s likely to be a larger investment in time, skills, and outside assistance to achieve your objectives.
So you need a range of disciplines to properly plan and execute a sustained personalized marketing project—that only comes with buy-in across the involved business units. I recommend setting up a dedicated team with a mandate, plan, and buy-in, aiming to rapidly show value, which will justify investment in a broader personalized marketing rollout.
Data is your friend once you implement machine learning. Ideally, you’ll build a measurement framework that will dictate your goals and targets how you’re measuring the success of your machine learning modeling and approach. And don’t be afraid of bad results—you can learn as much from them as good personalized marketing results.
Run a series of tests or models every month in a test-and-iterate methodology so that it’s baked into your ongoing optimization agenda. Then report its effectiveness back to the business.
Done properly, a carefully planned and managed machine learning strategy will grow your personalized marketing conversions through a process of continuous optimization.