Marketing applications are undergoing a dramatic boom, with the “Martech 5000” ballooning to 7,040 entries. Machine learning benefits from this rising tide in no small way, with almost 80% of marketers electing to use machine learning in 2018.
But with any rapidly growing technology, many marketers are struggling to bring machine learning into their programs. Some are achieving significant goals through machine learning, some haven't found success yet for various reason and some don't know where to start. If you all into the latter, read on for a foundational overview of machine learning.
What is machine learning?
Here’s a layman’s introduction to how machine learning works:
- First, get historical data that describes that you want to accomplish.
- Example: a list of prospects who have visited your website.
- Pick a machine learning model and load your data. Ask the question you want.
- Example: Of the prospects who have visited our website, who will convert?
- The machine will make its predictions, and probably get them wrong.
- Show the machine the correct answers – here are the customers who converted.
- Iterate until the machine gets it right to within a decent confidence interval.
It's as simple as that. But in practice, it's not really all that simple.
Here’s the thing: What marketers consider machine learning often isn't. In fact, many marketing companies produce applications that are called machine learning, but don't learn and contain no smart features.
According to Paul Roetzer, founder of the Marketing Artificial Intelligence Institute:
“There may be teams of 300 engineers, but they’ve only got 3 people who actually know what machine learning is, how to use it, and how to identify and solve machine learning problems. Hence, platforms are built by folks who don’t know what AI is and who continue to “improve” it with capabilities that they don’t know aren’t intelligent.”
In short, making machine learning into a successful part of your stack will require a great deal of careful preparation.
Where to start: Clean data
In order to start putting marketing in machine learning, you need to start with marketing data – clean data, specifically. Garbage in, garbage out, as they say. If your data sets are out of date, full of errors or riddled with duplicates, then the resulting outputs will be less than optimal.
The bad news is that your starting point may also be less than ideal. Up to 25% of marketing data is incorrect, with errors in terms of demographics or their status within your pipeline.
Fortunately, there are shortcuts that don’t involve going through your database line by line. If your database contains only a few entries that are missing values, you can delete them.
You can also train your sales staff and marketing personnel to create better records going forward. Since bad data leads to bad decisions, it’s worth putting the time in to solve this problem up front.
Putting machine learning in marketing
Once you have a good foundation, it’s time to integrate machine learning into your marketing stack.
The good news is that there are a lot of pre-existing single-purpose AI/ML platforms for marketing that will snap into your existing platforms right away. Social listening tools will help you design ad campaigns or even write limited ad copy. CRM tools will generate personalized ad copy based on your buyer personas. CMS tools can automatically A/B test your landing pages. ESP tools can incorporate dynamic copy and images to improve relevancy, and thus clicks. And end-to-end partners can help you clean up data, solve for identity issues in your database and implement machine learning in meaningful ways.
All of these solutions are polished and market-tested, but there are two problems you must look out for.
- Is this something you need, or does it just look shiny? Marketers are known to waste about a quarter of their budgets. You may be tempted by a new digital marketing tool, only to find that it rusts in your toolbox – try it to narrow it down to something you really need.
- Are you using it correctly? Machine learning algorithms respond to bias. In terms of social listening, for example, an exclamation mark can mean either elation or anger. If you don’t train your model correctly, you might focus inordinately on the needs of satisfied customers while failing to take input from those who are about to churn.
With all of the pitfalls along the way to making machine learning a success, it’s worth identifying partners with proven success and taking some direction from companies that have been able to do it right.
Machine learning success story
What does it look like when marketers make machine leaning a success?
The cosmetics company Glossier started from a blog called Into the Gloss. They thought that readers of the blog would naturally shop their cosmetics. It wasn’t that simple.
While blog commenters were extremely engaged and likely to purchase products, those who only read the blog were less so, and those who only visited the ecommerce site were least likely to purchase a product.
Using machine learning, the company was able to trace the most popular customer journey. Purchasers would click a link from the blog to the ecommerce site on mobile, add a product to their shopping cart and then complete the transaction on their browser. Glossier was able to use this information to smooth the customer journey, making it easier for customers to enjoy an omnichannel shopping experience.
Although your own journey to machine learning in marketing may look different, it will contain the same elements. If all goes as planned, you’ll use accurate data in order to pursue relevant goals and receive an unexpected yet actionable result.
If you want to learn how you can use machine learning to power more personalized and relevant customer experiences, contact us today.