Advertising has changed a lot over the years.
There was a time when machine learning, automation, and software-based marketing tech stacks weren’t a “thing.”
But now we’re past the days of just radio, outdoor, print, and a handful of channels on TV.
There are hundreds of channels across physical and print media and online at present, including social, mobile, and video. Even TV has diversified into hundreds of cable channels on your remote control. And yet, digital ad revenue has gone on to surpass that of TV.
The dominance of digital is nothing new. Paid search marketing is becoming more data-focused than ever before.
In fact, if you check the folders on your computer, I’m guessing some of you will find a few million-row spreadsheets full of cost-per-click bids, conversion rates, and return-on-ad-spend figures – along with countless other metrics for however many thousands, or millions, of keywords you manage.
So, What Do You Do with Big Data?
Because paid search is so reliant on big data – really big data, the kind that causes Excel spreadsheets to eventually crash for having too many rows – it’s my belief that the future of digital is inextricably tied to machine learning.
Is it because machine learning, automation, and software will completely replace savvy digital professionals and their creative ideas?
No. Far from it.
I believe that the future of digital will be a combination of smart marketers – like yourself – empowered by smart automation based on machine learning. As it happens, in a survey we recently ran on the subject, 97 percent of top digital marketing influencers (including speakers from AWeber, Oracle, and VentureBeat) agreed.
What Is Machine Learning & Why Is It Important?
Machine learning is the smart automation that can parse those million-row spreadsheets and pull valuable insights out of those mountains of data.
(To clarify, processing data to pull insights is something machine learning can help with… but actually taking those insights and doing things that are creative and smart with them? That’s still very much the domain of brilliant marketers like yourself, and why the ingenuity you bring to the table will continue to be so important when facing tomorrow’s digital challenges.)
As for why machine learning is important? For starters, digital advertising has a data problem. In addition, the face of marketing is changing due to the way your customers are becoming aware of, considering and purchasing your goods and services.
Digital’s Data Problem in Three Parts
Data is a challenge in modern marketing. There’s significantly more of it than there used to be, and as marketing technology matures, it becomes capable of collecting even more on top of that.
Data overload is a known problem. There’s too much of it – an overwhelming abundance of it already.
Yet Oracle points out that digital data growth is expected to increase globally by 4,300 percent by 2020. This problem isn’t going away anytime soon.
Despite the collection of data increasing exponentially, there’s a lack of centralized ownership with big data. You and your colleagues may be collecting CPC, CTR and CVR data in spreadsheets, but is everything centralized and standardized in a way that everyone in your organization can pull the data when they need?
Veritas reports that 52 percent of all business data is “dark” (of dubious or completely unknown value), and projects that mismanaged data will cost businesses $3.3 trillion by 2020.
There’s also a problem with siloing. Most businesses collect data in different buckets that aren’t necessarily integrated directly with each other, or indeed, with their own in-house marketing tech stack.
Accenture reports that while three-quarters of all digital skills gaps (the gap between a team member’s current level knowledge and the level of knowledge they need to successfully use new tech and tactics) come from lack of ownership, the remaining 25 percent of digital skills gaps come from a lack of integration.
And Then There’s the Changing Customer Journey
In addition to changes in the way data is collected and used in digital advertising, customer behavior is changing.
Advertising isn’t limited to a handful of channels. There are literally thousands of ways to reach customers, and pretty much all of them can be easily tuned out by an audience of increasingly demanding and disaffected customers who expect to have exactly what they’re looking for delivered to them instantly (and who will react poorly when it isn’t).
Research firm McKinsey breaks down the all-important consideration stage of the buying journey into four parts: “initial consideration; active evaluation, or the process of researching potential purchases; closure, when consumers buy brands; and postpurchase, when consumers experience them.”
The firm also finds that two-thirds of the touchpoints in the crucial evaluation stage are customer-driven, including browsing online reviews or soliciting word-of-mouth recommendations.
More to the point for those of us in digital, the use of ad blockers has increased 30 percent in the past year. And as you’ve surely heard, Google itself will be building in an “ad filter” in a 2018 version of Chrome to filter out “irrelevant” and “annoying” ads.
Effectively, as time passes, your ads are at greater risk of being filtered out by users who aren’t buying what you’re selling at this exact point in time.
How Does Machine Learning Solve These Problems?
Machine learning can be used to rein in the challenge of data, particularly when combined with disciplines such as probability-based Bayesian statistics, regression modeling, and data science. One of its greatest strengths here is the ability to take data-driven insights and build predictive models.
These predictive models can, in turn, be used to proactively address points of peak buying interest, attrition, or other key moments observed in the customer buying journey.
Examples of Machine Learning in Action
Let’s look at some examples of the way this technology is being used.
Chatbots & Voice Assistants
You may have noticed an increase in the use of conversational interfaces from major publishers such as Google, Amazon, Microsoft, Apple and Facebook in the form of chatbots and voice assistants (Alexa, Google Assistant, Siri and Cortana among others).
TOPBOTS notes that chatbots can have uses in unique, consumer-based contexts, such as event ticketing, health-related questions and the ever-important sports scores. These interfaces create a relevant and engaging user experience by supplying conversational responses based on historically-collected data – the most commonly-used or highly-searched terms.
Predicting & Preventing Customer Churn
A significantly deeper-funnel strategy at the post-purchase stage is to use machine learning to forecast common points of customer attrition.
Microsoft Azure and Urban Airship have both built predictive analytics models to determine the approximate timeframes and buying stages at which customers tend to most frequently churn. By projecting these important points in the future, these businesses are then able to proactively address common complaints before customers churn, driving higher retention and ultimately strengthening their businesses.
Natural Language Processing (NLP) and Semantic Distance Modeling
Another method of using machine learning specifically for digital advertising is to predict accurate bidding models for low-data keywords, such as long-tail keywords with high purchase intent but little to no empirical data.
In these cases, machine learning-based digital advertising solutions can assign new keyword groups based on semantically similar keyword groups and help advertisers ramp up long-tail keyword groups and all-new ad groups with a minimum of expensive testing time.
Machine learning isn’t necessarily a threat to marketers. On the contrary, it’s a powerful ally that’s making marketers’ lives easier while empowering them to predictively engage their customers in a highly relevant way.
Now, more than ever, it’s important to deliver the right message to the right customer at the right time – and with the power of machine learning, marketers are able to more accurately accomplish this goal by relying on actual data, rather than guesswork.