Machine learning has certainly been a “buzz word” within the marketing landscape over the past three-plus years. The difference today is that several brands are implementing machine learning strategies and are seeing positive results. In fact, the overall global artificial intelligence revenues will see a massive growth from just 643.7 million in 2016 to an excess of 36.8 billion in 2025.
As consumers, many of us have experienced machine learning firsthand. Whether it be experiencing a “smart dressing room” at a retailer or interacting with your local grocer’s robotic digital assistant, machine learning surrounds us. And as marketers, it’s important for us to understand how data factors into machine learning technology.
The role of data within machine learning:
- Data adds quality to your machine learning program. Remember, it’s about data quality versus quantity—ensuring you have the right data. Many marketers think the more data, the better. This is not always the case. There’s something to be said about quality over quantity. It seems obvious, but it’s not uncommon for a marketer to line up their machine learning program and have it ready to go, to only find out that the data is not connected because there is no common ID between the disparate data sources. So find that common link and bring it all together.
Next, you have to assess the value of the data. The buzz of “Big Data” has lost some of its steam over the past few years as not all data can have an impact. We see it ourselves when we test additional data attributes, theorizing that it will provide a lift in performance, and it has no significant change. Be sure to test, test and re-test so you can feel confident knowing your investment in data will give you the results you want.
- Data fuels machine learning algorithms. Prior to beginning your machine learning analysis, it’s important to ensure you have qualified data to “fuel” the machine. Often the term “machine learning” can be intimidating but as one of our own data scientists explains it: If regression is algebra, machine learning is calculus, and the machine can run calculus on hundreds of thousands of elements to make that equation more precise. For example, Amazon applies machine learning to customer data to make an accurate forecast for many products, detect fraudulent activities and offer customer-specific product recommendations. Or you might have experienced this as a customer; you receive a call or text from your bank checking on a suspicious charge on your credit card. This fraudulent activity is identified by a machine learning model.
- Data enables one-to-one communications. When it comes to machine learning, one of the key desired outcomes by marketers is improved personalization. Machine learning is leveraged to create experiences, but enabling these experiences requires data. To achieve personalization, marketers need to train the machine to predict the personalized messages each consumer should receive, which could be a promotion, special landing page or product recommendation. Marketers need to identify those who have responded to a personalized offer and train the model to find “lookalike customers” and deliver that offer or creative when they visit your site.
At Epsilon, we continue to innovate our data offerings and have incorporated machine learning techniques to take our modeling solutions to the next level. We work with brands to help them:
- Activate the audiences in any channel for integrated omnichannel campaigns
- Reach new, incremental prospect audiences to grow their brand
- Access more performing names to replace underperforming universes
- Drive stronger performance while improving return on marketing spend
Let’s put it in the perspective of a retailer. The retail industry will invest more than 8 billion in machine learning by 2024. It’s important to understand how retail brands are managing their machine learning initiatives regarding their data strategy. The majority of retailers are working within their customer file, which essentially means they’re connecting with existing customers for cross-sell/upsell opportunities and prospect within their own file. As it relates to their prospecting efforts, retailers are focused on reactivating former customers. But as I mentioned above, retail brands must ask themselves if they have all the right data to succeed with their machine learning goals.
Retailers need to consider the value of third-party data as it relates to their machine learning efforts. Sure, the web-based activity data they are using is effective, but retailers need to think of how they can marry demographic and third-party transactional data to get a boost in their machine learning model. For example, a data set like niches can help retailers gain additional insights into the customers’ life stages and message them based on their individual needs.
As you’re evaluating your machine learning strategy, think data. Research all your data options so you can best understand what’s the right data for your program. And don’t be overwhelmed by the thousands of variables you’ll find in some models. A marketer with a large customer file, like Walmart or Best Buy, might have more than 10,000 variables. Embrace the options and don’t dismiss the benefits of data mining to help you achieve your goals. Machine learning will continue to transform our marketing efforts, and during this transformation, make sure you’re data-ready.
To learn more, download our e-book, how to access data quality in an omnichannel world.