I’m sure many of you, like me, are overwhelmed with the amount of automation entering digital marketing over the past few years. Smart bidding, match types, responsive ad copy…the list goes on. We see the changes Google is making right in front of our eyes and it can be scary to think about our jobs in three years, five years, or even next year! I’ve heard a lot of discussions around whether or not there will be a need for PPC analysts as technology continues to become smarter and more advanced. Can we be replaced by machine learning? Will businesses opt for automation over customization? While I can’t predict the future, I can share my perspective—and I don’t think humans are going anywhere anytime soon.
I recently read Frederick Vallaeys’ newest book Digital Marketing in an AI World, which discusses Google’s advancements in artificial intelligence and machine learning from the perspective of the first AdWords evangelist, and one of the first 500 employees of Google. One of the opening quotes helped open my eyes to some of the conversations our PPC team has been having.
“People are afraid of what they don’t understand and the fear is what makes them angry.”
As PPC analysts, we have a lot to consider when we make optimization decisions. When Google and other platforms release features that remove our visibility and control, we typically get frustrated, angry, and annoyed about our need to now adapt. We don’t know how these features will change account performance or what we’ll need to do differently, and the unknown is what sparks our frustration.
So, the question is raised: Can machines be better at analyzing data than humans?
Machines Are Better
One facet of the argument that Vallaeys presents is that machines are better in terms of efficiency, accuracy, and timeliness. There are quite a few instances where machines can trump humans.
Let’s think about bid adjustments, for example. Analysts consider time of day, day of week, demographic profiling, locations, etc. Imagine all of the combinations that would have to be tested to find the most efficient or effective combination. Machines can do this infinitely faster than humans. We’ve seen this first and foremost with automated bidding taking all of these signals into consideration and, in theory, delivering the perfect combination.
In recent history, users had to do several searches to come across what they are looking for. Nowadays, your audiences see you on Instagram, search for you online, look for your products on Amazon, consult Twitter for peer feedback, etc. Machines can track these users down way faster and more accurately than we can. In addition, computers can do the math to support the value of each of these interactions. Here, we are presented with data-driven attribution—one of the most valuable attribution models used in the PPC industry.
Before Google Analytics, we knew there was a click, but that was it. Now we have gained insight as to what keyword brought users to our landing page and what happened after the click (Did they fill in a form? Buy something?). It’s not just about access to data —tools like Google Analytics allow us to understand what the numbers mean. We now have access to so many useful reports as a result of machines helping us determine what is or is not performing well, or if there are anomalies.
Speaking of all of those signals—the customer’s search, location, time of day, etc.)— machines can almost instantly evaluate these characteristics across any given searcher and simultaneously determine who, among millions of advertisers, would be the most relevant or best fit. This entails the instantaneous analysis of every advertiser’s Max CPC and Quality Score (which is already an analysis in itself), and assignment of an Ad Rank, all while working to achieve the set Target ROAS or Target CPA.
Moore’s Law states that computing power doubles roughly every eighteen months. This means computing power has doubled three and a half times since Workshop Digital was born, almost seven times since the year 2000, and thirteen and a half times since the start of the last decade. At this rate, humans can’t possibly compete with computers.
Given all of this information, it’s obvious that machines are better, right? Not so fast.
One perspective presented in the book that I can’t say I agree with —machines are always better at analysis and prediction. Yes, my Spotify Discover Weekly playlists are awesome thanks to machine learning and its ability to analyze and predict. Nobody can deny that. However, the book provides an example of a raccoon found on Google’s campus that security has been summoned to shoo away as a result of requests submitted through their automated system. Vallaeys argues the machine learning and delegation of tickets is the best and most efficient way to handle requests. Think about this in a more relatable sense.
Let’s say I go outside for my afternoon walk and there is a rabid opossum outside waiting for lunch. If I’m presented with two options—one, walking back inside and notifying our building manager (direct communication); or two, submitting a help ticket to WorkshopDigital.help.com and waiting for the machine to delegate the ticket properly (hopefully) based on previous submission learnings—I’d feel more confident choosing the direct communication. What if I relied on the machine’s knowledge and the ticket was delegated to the one person terrified of opossums because the machine has been trained to otherwise think this person is appropriate for administrative requests? The human element, in some cases, just can’t be replaced.
Humans Are Better
Humans are able to understand and incorporate social cues, creativity, and an emotional perspective that machines are not yet capable of providing. There is so much we overlook when we think about whether or not machines are capable of doing our jobs.
Recognizing What’s Important
Another example given in the book is from a ski resort whose sales are correlated to the amount of snowfall received in any given season. If the machine isn’t first taught to understand that snowfall is important to this particular ski resort, it could start optimizing for something else it deems important. Can machines be useful without the initial human input? Humans are necessary to inform the initial learning phases and guide the intelligence in the right direction.
At the end of the day, humans know what might resonate with other humans better than any machine. We have the emotional tact and understanding to catch something that might be offensive or ask questions in ad copy that just might earn an extra click. For example, PPC Analysts incorporate creative best practices into ad copy, landing pages, and call-to-action buttons every day. Users typically enjoy something more engaging or relatable than a default of “Click Here”. At the core of what we do, we have to know how to communicate—something not yet mastered by machines.
Setting Business Goals
Does it make sense to maximize sales? Earn leads? What’s most important to your client? It’s our job as analysts to determine what a client needs or what their goals should be and have those initial conversations. It is also our job to translate these goals to realistic key performance indicators and input the precise outcomes we are working towards into our systems, so the machines can continue to optimize for the right goals.
Navigating Uncertain Waters
Most members of the Workshop Digital PPC team have been in a situation or two where they’re optimistic about trying auto-bidding on a campaign or account, and it just tanks. We’re confident this bid strategy could help performance, but we just don’t have the volume. So, we defer to micro conversions. We use quality indicators and smaller actions along a conversion path to help the machine recognize that someone who signs up for a newsletter or downloads a whitepaper is “X%” more likely to achieve the goals we were initially aiming for. We have to help the machines recognize these patterns and get back on track when they shut down in the absence of enough data.
Measuring & Reporting Results
We have to know what is valuable for our clients. What do they want to see the most? How can we make it digestible? Could you imagine if your clients only had access to spreadsheets or machine outputs with no real narration or context? Humans can translate large amounts of data into understandable summaries to share with clients. Machines don’t quite have that down.
So, are humans or machines a better fit for our jobs?
The answer is both. For now, both are necessary to get the work done right. The synergy we gain as a result of working with computers, adapting to the changing times, understanding the implications of Moore’s Law, and being prepared for the next wave of technology, will only make us better, faster, and stronger as PPC Analysts.
Machines are better informing human decisions—but humans are learning how to interact with machines to achieve their desired outputs.
- Humans are necessary to create and improve algorithms, as most machine learning is completed under supervision.
- Humans QA and course-correct when things go wrong.
- Humans recognize opportunity using creativity and intuition, something machines can’t yet do.
- Machines can handle data processing in a fraction of the time we could even try.
- Machines recognize correlations, trends, and signals just as fast.
As we focus more on the ways we can leverage machine learning, as opposed to being frightened by its impact, we will learn its nuances and find a way to incorporate these advancements into our jobs, just like we did with cars or the internet. However, it’s important that we recognize that this is happening, and it’s happening quickly. We need to work together and learn how we can be best prepared. Knowledge-sharing is going to be extremely impactful as Google and other platforms continue to advance their machine learning capabilities.
Have you or your agency found a process that works well when you’re adapting to new technologies? If so, we’d love to chat! We’re constantly iterating and looking to find efficiencies in our processes, so feel free to reach out with some ideas.