By Ryan J Barr, Andrew Williamson, Emily Finnegan, and Carter Andrues
Three Ways in which Machine Learning is a Competitive Advantage.
Machine learning can be used in myriad ways to power the future of digital marketing. In the prior blog post, we introduced you to machine learning. In this entry, we share three powerful examples of how machine learning may be used to power marketing campaigns.
- Using Machine Learning to Make Hundreds of Personalized Email Campaigns to Maximize Response Rate and Conversion Rate
In his ebook, “The 5Ps of Marketing Artificial Intelligence,”[10] Paul Roetzer explains how machine learning can personalize email campaigns in conjunction with with gated content. Gated content is that which is accessible only in return for giving something up, such as providing your email address in exchange for an ebook. As a marketing agent or business manager, you have probably seen or run a gated content campaign before.
Let’s use the same principles as Roetzer did with his email campaign example, except with trip photographs instead. For this example, let’s pretend your company provides trips and experiences for guests/tourists, maybe surf lessons, rafting, or a zipline. A customer goes on an experience with your company, and an employee-photographer takes pictures of the action. The customer then returns home and goes back to your website. This customer wants pictures from the trip, so they input their email in exchange for a few of their trip photos for free.
When that email address is received, you or your marketing agency puts that customer on an email marketing list, and email automation will automatically send that person a series of emails over a period of time. These emails can be just about anything. Some examples are: how to get the rest of their pictures at a particular price, review follow up, discounts for a return visit, etc. These automated email campaigns are fairly easy for marketers to set up, say, in MailChimp or other email marketing software. This concept of setting up an automated campaign is a set of rules, and that set of rules is known as an algorithm.[11]
But, what if your company received hundreds of requests for gated content? Say, for example, that all your summer visitors are inquiring, via phone or email, at various times during the summer and through the fall, sending in requests for their pictures. Easy to handle the responses, right? No! Not if you want to personalize each email. Sophisticated tour operators might already have obtained the customer’s email address when the customer signed your risk acceptance waiver. Wait, what?
To put it simply: When potential customers decide to go on a whitewater rafting trip, for example, they have to sign a waiver that states: “I understand this activity is dangerous and I won’t sue your company if I get hurt.” In the perfect world, your company should make that waiver digital, so when the customers signs it at home or on her mobile device, she can “Sign in with Facebook” or “Sign in with Google.” So, when the customer wants that picture and input her email address, your computer matches up her request with the previously-signed digital waiver that contains information about her age, Facebook account information, Google advertising profile, where she’s from, etc.
In our agency, this is the process we use to capture not just the customer’s name, but Facebook account information, Google advertising profile, and a lot more. Everything Facebook and Google knows about that person, we now know about that person—and so do our clients.
After the trip, customers might reach out to your company again. However, if they don’t, that’s okay, because you should email them again anyway. For example, you might want to thank the customer for her business, offer her a discount to refer her friends, etc.
Note that it quickly becomes rather impossible for a human to personalize that many emails. However, such personalization is not difficult for a machine learning algorithm. “It takes data-driven, complex tasks and makes them look easy”.[12] And by using a digital marketing agency that has access to more data than any one company alone does, the agency’s machine learning algorithm will be more finely tuned.
The machine learning algorithm “listens” as it sends these emails. It watches for what emails are working and which ones are not. Each blast of new emails is different from the last. Through this learning, it tweaks and makes adjustments, further fine-tuning its efforts for maximum effectiveness. “In other words, it learns, it gets smarter, and it creates its own algorithms”.[13] In addition, the efficiency of machine learning allows your company and your personnel to tackle other projects.
About now, if you are in marketing, you are probably a little scared that machine learning might replace you, take your job, as it were. According to the Marketing Artificial Intelligence Institute, artificial intelligence, predictive marketing analytics, and machine learning will not replace marketers. Instead, it will “largely enhance knowledge and capabilities, in the near term”. In fact, rather than replacing marketers, it will empower them. These technologies literally offer superpowers.[14]
- Facebook Can Make over 6000 Different Versions of an Ad
Most companies these days make a few different variations of an ad for their campaigns. Maybe a birthday ad for males has a blue background and one for females has a pink background. A marketer can set the campaign to target people who have upcoming birthdays. Is that enough variation to maximize responses? Well, maybe….
Facebook itself is using machine learning to get a little more specific.[15] Facebook has a new tool it’s calling “dynamic creative”.[16] Your company or marketing agency feeds the system a range of components (that you hopefully obtain from your marketing analytics) and specify who you want to reach. Then Facebook’s algorithm will do the rest to target the right audiences to maximize the return of your ad spend. If you spend $250 on an ad over two weeks, you probably want to see enough products/experiences sold to at least get $250 back, right?
Instead of using automation, let’s try to do this task manually, using real data from a watersports company in South Carolina; this company teaches people to surf, paddleboard, and kayak in their local waters. The company’s Google Analytics shows that, over the last two months, 35% of visitors range in age from 25-34, 22% are 35-44; 52% are male / 48% are female. Interests are… well, all over the place. Maybe Facebook Analytics data has some more specific interest targeting. Facebook Analytics provides more precise data to create and target its next Facebook ad. The data from their Facebook Analytics shows over 101 different varieties of “likes” with the highest percentile, 3.33%, “liking” food & culture; 45% customers likely have a household income of $75k-$99k. Your company and/or your agency continue to dig through these analytics… but, let’s not forget all the other tasks you are supposed to do today…
There is no way a marketing manager can make 101 different Facebook ads targeting different demographics and likes, and still have time for supper. A machine can, though! Facebook says that “dynamic creative” can automatically explore and deliver ads with variations of [its] choosing.[17] It will analyze which version of the ads perform best with each new audience it targets.
- Predictive Outcomes for Website Landing Pages to Reduce Bounce Rate. A Smarter way to Master Content Personalization.
A landing page is the very first page people see when they visit a website. This is not always the homepage. This page should tell a customer who your company is and what services or products you provide. It captures the customer’s interest and hopefully she begins to explore more of the website, which leads to a conversion.
In a perfect world, machine learning has optimized your website to be psychologically seductive so that people stick around and purchase. This isn’t always the case, though, as many unoptimized web pages have high (terrible!) bounce rates (people who leave the page quickly; if one hundred people visit a webpage, and all leave immediately, that would be a 100% bounce rate[18]).
Say someone is searching for “Backpacking Guides in Yosemite National Park” via Google, and then clicks through to your website; you probably want them to stick around and not bounce. Through landing page optimization, a marketer can study data and, based on research about what does and does not work, make changes to the website. It’s a time consuming, tedious process.
Using machine learning to train a system to study data about what does and does not work on websites, this process can be less tedious. “What we’re really trying to understand is how to make it better” says Mike Moran in a recent Search Talk Live podcast.[19] Machine learning can look at many features–things like button color, paragraphs being turned into bullets, adding more pictures, more links going to other pages, etc.– to understand what is working in order to keep doing it, and what is not working in order to change it. All of this and more can contribute to a beautiful landing page with a low bounce rate–and a high conversion rate.
Ideally, says Mike Moran, companies should run this analysis before a website is launched in order to have a perfect site out of the gate. If not, he continues, “feature analysis” of an existing page can reveal what aspects, or website features, are contributing to high and low bounce rates.
Training the machine requires access to different data–any data it can get ideally; mostly machine learning loves analytics. It looks through patterns in data and draws conclusions from that data. However, don’t feed it garbage data, because “garbage in equals garbage out.” To train your machine algorithm, “feed” it data that your company understands so you feel comfortable with the patterns it is seeing and recommendations it is making. Over time, it begins to understand the features that are associated with high and low bounce rates, and in time, it can make suggestions.
Here’s a pro tip: Don’t feed your machine learning the thank you page nor the shopping cart, as these two pages naturally have the highest conversions. Instead, be sure to monitor the data that the machine obtains from these kinds of pages. For example, to understand checkout behavior, look at abandoned carts and bookings.
Can the machine learning system create a website that is always adapting to each unique visitor’s persona and likes? Can it continually learn and make these changes? Yup! That’s where this technology is headed in the future. In addition, when digital marketing agencies use machine learning, they can leverage data from multiple websites/clients to more finely-tune the results for each client.
- Conclusion
Broadly targeting massive audiences like so many digital marketers are currently doing may be easy, but it won’t help them keep the competitive edge. Maximizing conversions with a finely tuned and specific targeted campaign is the next step in the search for better marketing strategies.
Using machine learning to automate digital marketing is the future of marketing. And, maching learning is not limited to email marketing, Facebook Ads, and website content personalization. These marketing tasks are only the beginning. Get a data scientist on your marketing team today! Discover the future of creating, communicating, and delivering your ideas, experiences, and products to your target customers.
Sources
[10] Paul Roetzer, The 5Ps of Marketing Artificial Intelligence, 2017, https://www.marketingaiinstitute.com/blog/the-5ps-of-marketing-artificial-intelligence
[11] Paul Roetzer, The 5Ps of Marketing Artificial Intelligence, 2017, https://www.marketingaiinstitute.com/blog/the-5ps-of-marketing-artificial-intelligence
[12] Paul Roetzer, The 5Ps of Marketing Artificial Intelligence, 2017, https://www.marketingaiinstitute.com/blog/the-5ps-of-marketing-artificial-intelligence
[13] Ibid.
[14] Paul Roetzer, The 5Ps of Marketing Artificial Intelligence, 2017, https://www.marketingaiinstitute.com/blog/the-5ps-of-marketing-artificial-intelligence
[15] Facebook, Two New Tools for Your Best Holiday Campaigns, 2017, https://www.facebook.com/business/news/two-new-tools-for-your-best-holiday-campaigns
[16] Facebook, Dynamic Creative, 2017, https://www.facebook.com/business/m/facebook-dynamic-creative-ads
[17] Facebook, Dynamic Creative, 2017, https://www.facebook.com/business/m/facebook-dynamic-creative-ads
[18] Google Support, Bounce Rate, 2017, https://support.google.com/analytics/answer/1009409?hl=en
[19] Search Talk Live, Mike Moran Discusses Using AI to Evaluate Search Landing Pageshttp://searchtalklive.com/previous-search-talk-live-shows/