A Comprehensive Guide to Machine Learning
Now that we know the language of Machine Learning, we’re ready to look at specifically what marketers can do using Machine Learning. The ad tech space is full of companies promising the next silver bullet for marketers. Armed with your new knowledge of Machine Learning and related concepts, we can begin to look past the veil toward what makes these tools, process, and marketing services tick.
It’s highly unlikely that if you’re reading this guide you’ve not worked directly with the concept of marketing automation. It’s even highly likely that you’ve played around with one of the industry’s leading marketing automation platforms such as Marketo or HubSpot. Even niche tool providers like MailChimp have created automation in their platforms to help marketers scale their efforts.
These automations often help marketers and sales teams streamline and manage tasks that human hands once did. Drip emails are a great example. You can now build an email list, generate a number of templates, and the system will email your recipients at the time and on the conditions you instruct it to.
While these automations are highly valuable to marketers, the next iteration for these systems will be layering in Machine Learning.
Using the above example of email drips, software providers like Mautic are already offering mail automation that relies on a logic true. If your recipient Z takes action X, then it sends email Y. This supervised learning system (remember that term from Chapter 2?) is a basic one.
The next evolution in such a system would come from Mautic learning how long it takes for recipient Z to respond to emails and instructs your sales team on the best time to follow up an email with a call based on an unopened email. Going even further, the system helps you to pick the best adjectives and verbs for your subject lines based on previous subject lines and open rates. You’ve likely seen or reviewed tools with these types of features, since they’re becoming more available and can greatly impact the value of human capital working on marketing and sales initiatives.
Sending frequency optimization can also have a substantial impact on both your standard email marketing initiatives and your drip campaigns. Machine Learning can help you answer the following questions:
1. How often should you pay attention to specific recipients and segment new marketing messages?
2. When is the best time to send email to specific recipients and segments?
Before Machine Learning, marketers would leave frequency optimization up to testing and ROI analysis; however, frequency optimization was all but a measurement on full lists in most cases. Machine Learning allows marketers to carve lists into precise segments and to neatly personalize sending frequencies for individual recipients.
How much time are you spending on administrative tasks, such as asset tagging, versus content creation? Tagging assets with relevant keywords is essential to making them searchable, but it’s also a tedious, timeconsuming task that most marketers would rather avoid.
Machine Learning technology can smartly include the most valuable or least expensive keywords in copy. This technology can associate similar or related assets, making it easier to combine relevant copy, images, and videos for target audiences. If your audience has consumed one bit of content, then it can smartly recommend what to offer next for the highest conversions. Machine Learning technology can also help predict what content will lead to the behaviors — sharing or engaging with content, increased sales, and improved customer loyalties — you’re trying to gain from customers. Adobe’s Smart Tag technology is available now to automate metadata insertion so that you can get better search results while slashing the amount of time you spend on this task.
Each marketing channel presents a special set of requirements for the size and resolution of marketing assets. When a new platform emerges — or if you decide to add a new channel — it could require the time and cost of redesigning existing assets. For example, if you have a piece of content delivered to a web channel or blog, Machine Learning can smartly crop that content for a mobile channel or reduce the copy in smart ways. With Machine Learning, you can shorten visual content and videos to optimize experiences for different channels based on the patterns by which people are consuming them.
Machine Learning will either offer recommendations — or provide a first draft of the new content — that can then help accelerate the pace by which you get those different pieces of copy, creative graphics, or videos published to various channels and distributed to the selected audiences. You don’t want to have to create massive volumes of content and hope that only some of it will be effective. What’s important is being able to create the right content that’s effective in your channels, learn from that content creation, and then develop more content based on those insights as you expand from that point.
Machine Learning can give you the intelligence needed to quickly determine what’s working as well as recommend what’s needed to amplify your strategies that might better connect with your audience. The learning part of Machine Learning means that, over time, the machine becomes smarter. We’re still in the early stages with this knowledge evolution, but machines could potentially learn so quickly that you could remix, reuse, and adapt content almost instantaneously, test the content you’ve created, and learn whether what you’ve created will be an improvement over your previous campaign or whether you need a different approach.
One of the first platforms to incorporate Machine Learning into systems and processes that marketers use every day are media-buying software sets. As an example, Google’s quality score helps determine which ads are most relevant to searchers and the cost per click of ads by using factors such as click-through rate, ad copy relevance, and landing page quality.
With the growth of mobile internet consumption, banner ads and other display advertising have had to undergo a major change in order to retain their value. Further, with the rise of better and better technology, marketers are looking for ways to segment their message in a more precise manner and not broadcast to a large audience in hopes of motivating a few people. Programmatic advertising allows marketers to take matters one step further by measuring and pivoting on advertising strategies in near real time. Where big data has allowed for segmentation, programmatic advertising has allowed for marketers and advertisers to see how these segments perform on specific ads at specific times on specific devices, for example. This type of advertising also allows for more control over spend and higher ROI on display ads.
Google has developed much great functionality in its system for ad control and information in managing ads. Its resources will help you find new segments that you should be reaching and give you information on underperforming ads, groups, and keywords. However, tasks and processes are available to help an Google Ads practitioner in ways that Google Ads doesn’t offer: Google Ads Scripts fill this need.
Predictive analytics are used in many verticals and professions. Essentially, these analytics are models that help find and exploit patterns in data. These patterns can then be used to analyze risks and opportunities. In the marketing world, predictive analytics layers on top of our standard analytics data to give us more insight and help shape marketing actions. Previously, we’d look at raw data like “sessions” that would give us insight into our analysis of ROI based on base metrics of lifetime value for a session. Now, we can predict with more precision the exact value of each individual session based on their onsite actions and the sources of their referrals. As with programmatic advertising, we’re able to interact with the potential customers represented by those sessions in highly tailored ways, including chatbots, which we’ll discuss in more detail in a later chapter.
Keeping customers is as pivotal to growth as getting new customers. Analytics can help understand behaviors that lead to customer churn and help marketers craft strategies to reverse churn. Predicting customer churn is a valuable piece of this puzzle. Through Machine Learning, behaviors of past users that are no longer customers can give us a data set of actions to apply to the current customer base to see which customers are at risk of jumping ship on you. By taking advantage of certain Machine Learning models, like random forest, we can not only predict customer churn based on features, but identify the importance coefficient of each feature
Computer vision is exactly what the term sounds like — it’s how machines “see.”
Machine Learning, especially reinforced learning, has allowed machines to have ever-increasing capabilities related to image recognition. A great everyday example is the facial recognition that Facebook uses when suggesting people to tag in a photo. Facebook’s facial recognition technology has “learned” the faces of users over time as they’ve been tagged by the endless amount of photos that make their way into the Facebook universe.
Computer vision has practical uses in marketing.
For example, Sentient Aware offers software that lets its users serve customers with products that are visually similar to the products that they choose. This style of product serving could have a significant benefit over the traditional use of tagging, especially when dealing with new customers whose buying habits are not yet known.
Snapchat and Instagram have made visual-based social listening increasingly important. Ditto and GumGum provide social listening tools that can enhance reputation-management efforts. For example, brand marketing folks can receive alerts to tell them when their company’s logo appears in a meme or image that might need their attention.
Audience segmentation has always been an important part of advertising. Knowing the members of your audience and where they’re coming from offers marketers incredibly valuable information. Until the invention of the internet, gaining that data was never done in real time. Now marketers can gain almost real-time access to the demographic-based data of their consumers and create actions that interact with them.
Only a decade ago, marketers rejoiced when they gained access to data such as age, sex, location, and the length of time that users interacted with their messages. Now marketers can create micro-segmentations as well as measure and compare how each segment reacts to different messages.
To implement audience segmentation, marketers are able to use different clustering models based on how your dataset looks like. For shopping history datasets, the K-means model is the one you should use; if you have customers’ demographic data or survey data, K-modes would be a better choice. After labeling each customer, since Google Analytics offers behavioral-based demographic data such as affinity groups, marketers can create different user groups based on the result to help them generate more in-depth reports.