Utilizing artificial intelligence and machine learning is nothing new in the marketing world. We’ve discussed it in a broader sense with regards to AI marketing trends for 2020. However, as the tools continue to grow, both in their power and popularity, it’s worth it to know when to use machine learning performance marketing tools and when not to.
So where do we draw the line when it comes to automation? And what does that mean for the future of PPC marketing professionals?
How Did Machine Learning Get Here?
Early computer programs were built on simple scientific logic. A defined input gives a defined output. If this, then that. Do this, then do that. If confused, give an error message. These programs solved relatively simple tasks, but it was the beginning of automation.
From there we evolved to mathematical models which allowed us to process large amounts of data much quicker, perform large scale calculations, and easily parse out numerical relationships. Mathematical models were soon followed by statistical models, where this data was used to calculate probabilities and project odds. While these tools were game changers and still are extremely useful, they both require a human to compare the data and provide future iterations.
Where machine learning differs is in its ability to use previous data to figure out answers to future questions. It also has the ability to change its mind as more data comes in, thus removing the human need for future experiments.
Types of Machine Learning
Machine learning is broken down into three segments, defined mostly by the types of questions they’re asked.
- Supervised – when a machine is learning via supervised data, it’s learning to answer a question that we already know the right answer for. Ironically, many of the questions you’re asked to prove you’re not a robot (“Click on all the squares that have a stop sign”) are actually teaching robots how to answer that question. With enough examples provided of labeled data, these bots will eventually be able to answer these questions themselves with increasing accuracy.
- Unsupervised – when a machine is learning via unsupervised data, you’re asking it a question you don’t know the answer to. You provide the unlabeled data, and the machine is able to provide patterns and relationships within it. An example of this would be leveraging Facebook’s audience tool to create a lookalike audience – “Here is my audience. Tell me what they all have in common, and find more people like them.”
- Reinforcement – the last type of learning is when there is no absolute “right” answer. Some answers might be better than others, but there’s no absolute truth. Using another Facebook example, it’s creating your ads and setting a budget to run across multiple audiences and telling Facebook, “Put the right ads in front of the right people at the right time.” Since there’s no solution that’s going to work 100% of the time, the machine performs experiments, learns from the results, and optimizes over time.
To Automate, or Not To Automate
It should be clear by now the benefits of utilizing machine learning performance marketing tactics. Digital media captures more data than we know what to do with, and we have the tools that can solve large scale problems, even if there’s no correct answer.
However, as powerful as these tools are, they still have their limits. While it sounds like you can just push a button and get a perfect marketing campaign, the reality is that despite their ability to learn and improve, machine learning is still just a tool, and marketing professionals need to understand how and when to use them, as well as the best place to focus their energy.
First, let’s identify some of the things that machine learning does best.
- Crunch and organize massive amounts of data
- Identify trends and relationships, and make optimizations
- Execute routine tasks
A bot can pull reports and optimize bids better and faster than a human ever can. They can optimize between ads and audiences in real-time, driving better results at a more efficient cost. Bots don’t get tired, make mistakes, or have an attitude about the job they’re given.
Using Machine Learning Performance Marketing Tactics
Given what we know about how AI works and what it’s good at, below are some examples of machine learning performance marketing tactics that marketers can take advantage of right now.
- In-Market Audiences – Google looks at signals such as search history and content consumed to identify people who are likely in the latter stages of a buying decision. Without machine learning crunching vast amounts of data and behavior in real-time, it would be nearly impossible to find and target these audiences at scale.
- Lookalike Audiences – After you’ve built a core customer base, utilizing lookalike audiences may be the best way to reach people most similar to your existing customers at scale. Using Facebook, for example, you can leverage your customer data by using machine learning to find commonalities among them (location, demographics, affinities) and target people that match those characteristics. Depending on how your data source, you can even segment deeper into products, because the audience of Product A could look very different than the audience of Product B, and you can serve them different messaging accordingly. Without machine learning, this research would be a laborious process, relying on imperfect data and heavy assumptions about who your customers really are.
We run lead generation campaigns for a client of ours on Facebook where we target a variety of audiences, but most can be categorized as Lookalike, Affinity, or Retargeting. We update our Lookalikes on a weekly basis to continue defining our audience as business grows, and they’ve consistently been our top performer. From a Cost per Lead standpoint, Lookalikes have outperformed Affinity audiences by 43% and even outperformed our pre-qualified Retargeting audiences!
It’s worth noting here that while Lookalike audiences are extremely powerful, marketers need to continue testing new audiences to avoid data bias. Lookalike audiences however still require a level of management. Setting and forgetting will result in your ads only reaching a handful of demographics and affinities, ignoring many that would potentially be interested in your product. Continuing to make new assumptions and test new audiences will develop a more complete persona profile.
- Responsive Search Ads – In the pre-machine learning days, trying to find the right headline and description combos for AdWords was a manual process of creating multiple iterations of similar-looking ads to A/B test, then study the data in order to make adjustments and discover which ad(s) performed best. With Google’s new Responsive Search Ads, we now have the option to set up multiple headlines and descriptions within the same ad, allowing machine learning to test and optimize towards the best performing combos. It’s essentially the same process before, but leveraging the available machine learning tools to get results faster and with less manual work.
At NoGood, we tested Responsive Search Ads versus Expanded Text Ads to see which is better at driving traffic to our own site. The Responsive Search Ads won by a considerable margin, with a CTR 40% higher!
- Smart Bidding – Google’s smart bidding features allow marketers to set goals beyond the simple CPC and focus on further funnel metrics with goals of Maximizing Conversions, Target CPA, and Target ROAS. The previous CPC method was a bit of a manual guessing game, where humans had to decide if they were willing to pay extra for bids on certain keywords with the hopes that those users were more likely to convert. Now, machine learning can take all those factors into account to better understand your audience, and adjust bids and budget allocation accordingly.
- Attribution – We all know last-click attribution isn’t the right model, but it’s what so many brands default to just to keep an apples to apples of their different marketing channels. We’ve all heard the rule of thumb that it takes 5-7 touch points before a sale is made, though the reality is that number is often even higher in the digital world. Machine learning tools for multi-touch attribution are truly the only way to understand the full consumer journey, analyze the impact of each touchpoint, and identify which tactics are actually driving conversions and sales.
- Data Visualization – As marketing becomes more and more data-driven, drawing actionable insights from your data becomes even more critical. While tools like Excel or Tableau are great, they still require a ton of manual work to collect and organize your data – time that could be better spent analyzing the results themselves and taking action on the learnings. AI-driven reporting tools not only sort and visualize data in a fraction of the time but can help identify trends and patterns that could otherwise easily be missed.
While all these tools make life easier for many PPC marketers, it certainly doesn’t make the job obsolete – it just changes the nature of the role. Rather than focusing on the minor details, automation frees us up to work on higher-level strategy.
The machine will only do what it’s told. You need to know what to tell it.
As marketers, this means we need to evolve from tacticians to strategists. Our role is no longer to stare at dashboards, making marginal shifts of budgets as new data comes in. We need to grow from campaign managers into strategic advisors.
So first off, congrats on the promotion!
Machine learning performance marketing tools can handle the menial work which will save us time, improve quality, and allow us to scale better and faster than ever before. However, humans are not obsolete yet! We’re still needed to provide common sense, emotion, empathy, and make sense out of everything when the answers aren’t perfect. So what aspects of the process still require a human touch?
- Define what problem to solve. Machine learning can help get you towards the right answer, but the marketer still needs to provide the right question. This requires a deep understanding of an industry and knowing where your business or product fits into that landscape. It requires a complete understanding of business objectives.
- Define the metrics that matter. Machine learning can optimize towards whatever metrics you give it – clicks, video views, conversions, etc. It’s up to marketers to determine what the right goals are. Should your campaign goal be for sales, or does your product need more brand awareness first? Maybe your first campaign should optimize for maximum clicks, which would help your bot collect more post-click data, helping future sales-driven campaigns.
- Find meaning in results. A machine will understand your question, optimize toward your goal, collect and sort through endless data, and give you an answer. However, only a human can interpret if that answer is meaningful. Humans need to provide the “smell test” to see if the results make sense, and if so, what they mean for the business.
- Transfer learnings from one platform to another. With the increasing walled gardens and law for data protection, taking an insight from one campaign still requires humans to relay that to other media. This goes beyond just media, however. Audience insights from one platform can help guide product development as well. Say you’re an apparel brand and your AI built a marketing persona that said your core audience were pet lovers. Product teams can leverage that learning to develop a line of products for pets. Bot learnings result in human actions.
- Know Your Data. Your outputs are only as good as your inputs. A machine does the best it can with the data it has, but it can’t make judgments about the data itself. Marketers still need to ensure that the data is clean and being interpreted properly. How was the data collected? How can it be integrated or combined with other data? What could go wrong and what would that look like? These are still problems that only a human can solve.
- Know the technology landscape and use the best tools. When all you have is a hammer, every problem looks like a nail. However, in the ever-expanding digital landscape, we have a lot more tools and a lot more problems. One size fits all won’t drive the results we need. Marketers need to always be learning about how technology is evolving – how it works, how it can be used, what its potential dangers are – so that when the time comes and a new problem arises, they’ll know to put the hammer down and pick up the screwdriver.
- Grow your skills. With bots taking care of the mindless, time-consuming tasks, marketers now have even more opportunities to add new abilities to their skillsets. Learn more advanced psychology, so you can better position your brand and write better ad copy. Learn a programming language like Python or SQL to help you better analyze your data.
Machine learning performance marketing tools have undoubtedly changed the PPC game, giving marketers tools to execute faster, more accurately, and more efficiently. However, it still takes human power to tell the machines what to do, and interpret the results they give us.
Marketers now need to focus on taking machine learnings and translating them into actionable insights. We need to know everything about the industries we’re in, know the pain points of our audience, and how to position our products within that marketplace. We need to move away from focusing on tactics to focusing on strategy.