Machine Learning – The Future is Now!
I've been wondering how to write this post for a long time. It's difficult to summarize everything that's been happening lately in a few sentences. The world is rapidly moving forward, and many new trends are emerging, but is it worth always following them all? Whether you're a sole proprietorship, a marketing or sales specialist, an SEO/AdWords specialist, or the CEO of a large company (I've had the privilege of meeting all of these positions), you'll certainly find something to interest you in this post. It's about optimization through the lens of machine learning.
The demand for data and information is one of the strongest trends of recent times. It's no surprise, after all, that it's worth knowing whether you're spending your money effectively. The answer to this "data hunger" is a flood of online marketing analysis tools, various monitoring tools, scanners, and so on. They're all beautiful, colorful, and transparent (more or less), but are they useful? Do we really need so many different charts, tables, reports, and so on? After all, each new tool is just another report, a result that should be analyzed if we don't want to feel like we're throwing money at essentially useless tools. How many reports do we find truly unique or eye-opening data? I'd go so far as to say that 80% of these tools are worthless, with no significant impact on our marketing. Once again, the Pareto principle proves true. At Verseo, we focus on a few favorite (because they're useful and convenient) metrics. Of course, we test new things, but most of them suffer the same fate. We disable them because they don't bring anything new. This "minimalism," however, doesn't stop us from being one of the largest agencies in Central and Eastern Europe and a strategic partner of Google. For the purposes of this post, I've made a quick summary: when I worked as an active salesperson, over two years I made around 25,000 phone calls to companies, held over 200 meetings, and participated in over a dozen different training sessions. Thanks to this, I had the opportunity to meet many fantastic people, but also to meet the various problems of the companies they worked for. One of these problems was common to virtually every company... What is it? Lack of time... This isn't a particularly surprising observation, but it leads to a stark conclusion: companies don't need more data analysis tools because they don't have enough time to analyze them anyway. That's why so few companies conduct marketing based on real, up-to-date data.
Data is not everything…
Data analysis and optimizing marketing activities based on it is particularly difficult. Most of us know basic metrics like CPC, CPA, and ROI – it's not secret knowledge. However, when we start talking about IRR, NPV, ROMI or the recently increasingly fashionable CLTV, the first problems begin. However, this is not something bad or shameful. We don't have to (and can't) know everything.
Therefore, we must be aware that no matter what analytical tool we purchase, we will still need an additional person to analyze the data. Of course, it generates additional costs for the company, and few people like them. While a large company can still afford such additional costs, for small companies such analytics is practically unattainable. Let's be honest, purchasing a data collection tool without also arranging an analyst is a waste of money.
The funny thing is that small businesses sometimes find it more profitable to increase their advertising budget directly than to invest in additional data analytics. Practice shows that this is not such a bad or stupid strategy. Dlaczego? Because with budgets of PLN 1000-4000, there is so little data that doing super-detailed analyses doesn't make much sense anyway (sometimes it's better to just allocate an additional budget to the campaign instead of paying an analyst). I know this is a controversial statement for many, but our observations show that sometimes it just happens.
According to various sources, in 2014, over 80% of companies in Poland had Google Analytics installed on their websites, but (attention!) less than 2% logged in. Our “Web Analytics” tab, despite being in the main menu on the verseo.pl website for 3 years, was the least visited tab of all. Let us quote the classic: “The facts speak for themselves”!
The problem of data analysis is further exacerbated by the fact that there is more and more data. The reasons include: a larger number of traffic sources and extensive targeting options (gender, age, device type, time, location and many others). And this is just the beginning, there is more and more talk about the massively growing mobile traffic. Mobile traffic is a separate topic that deserves an article; suffice it to mention that in some industries in Poland, over 50% of queries already come from mobile devices, and in the US we have campaigns where this indicator reaches up to 70%. Many teenagers today no longer remember phones with buttons and no internet access, and my 2-year-old daughter wants to use the TV by touching the screen. “They” – accustomed to touch screens, mobility and constant access to the Internet – are growing in strength (and with it their consumer potential) every day. They do not accept non-mobile websites, they leave websites that display poorly on their smartphones, and they ignore companies that are unable or unwilling to ensure they are mobile friendly. This means that traffic from mobile devices will only increase, more and more new applications will appear, and advertising in augmented reality products is probably a matter of time.
There will be more and more data, and that means more work on the marketing side.
What can we do?
We're at a point where we realize that paying for additional analytics without an analyst doesn't make sense. We don't have the time and/or knowledge to handle it and analyze the data ourselves. The thought of this ever-increasing amount of data is overwhelming, yet the key to our company's success lies in well-analyzed data. So what should we do, how should we live? The solution is at our fingertips and quite simple... Let's use machines for this arduous task! This isn't science fiction; it's a solution available practically "out of the box."
Machine Learning – What is it?
We can say that these are "self-learning systems", "machine learning" or "self-learning algorithms", some even go so far as to call it artificial intelligence, although I would be more cautious with this.
What is this about?
Machine learning systems collect and process vast amounts of data. By combining fields such as computer science, automation, and statistics, software can learn (and acquire new knowledge) through data collection. After specifying appropriate algorithms and developing advanced statistical models, we can "force" the system to perform appropriate automated processes and improve its own knowledge.
And what does it look like in practice?
Imagine combining a dozen or so online marketing specialists (Google AdWords, Facebook, Bing, etc.), internet analysts, statisticians, and programmers into one super specialist, a robot that doesn't get sick, works 24/7, and gets better and smarter every day! And on top of that, we can constantly add new improvements from above, making it even better every day. Sounds impossible? But it's already here, works well, and is within everyone's reach. Do you know Uber? Yes, it's the company that orders transportation and handles conflicts with taxi drivers. But this isn't about evaluating their performance. Uber uses "machine learning" to continuously monitor real-time data on traffic volume, travel times, demand, and supply for its services in order to optimally adjust transportation fees. In marketing, however, we can connect systems/machines based on self-learning algorithms to analyze our online campaigns. We can create any combination of algorithms that will work towards our goals, for example, maximizing revenue from our campaign by selecting optimal click rates for all keywords in our online campaign, for example, in the Google.pl search engine (of course, this greatly simplifies the entire system). The system collects data about our campaign and makes (ITSELF!) decisions about targeting and bid optimization. You don't have to waste time on this, and the system will also base its decisions/changes on hard statistical data. What's more, it does this every day, even while we sleep. It also keeps an eye on whether everything is okay and notifies us if any anomalies are detected. This sounds like the description of the ideal employee (or at least a very good one). Mechanization in many areas of life is no longer surprising; now it's time for marketing.
But does it really work?
Let's answer a few questions: How many Google AdWords specialists do we have in Poland? Many. How many Google AdWords specialists with extensive experience do we have in Poland? Fewer. How many Google AdWords specialists do we have in Poland with extensive experience, who also have extensive experience in web analytics and are proficient in Excel (macros, etc.)? Very few. And these "very few" can command very high prices for their services – they are simply exceptional.
Time for a little case study
We have a super specialist who's well-versed in optimization and analytics. He manages a single store offering 30,000 products, including a product listing campaign (Google Shopping), remarketing, search engine campaigns, perhaps Gmail, and DSA. Let's say 10 campaigns in total (an optimistic estimate). He's well-versed in AdWords Editor and Excel, so building and monitoring campaigns isn't a problem. Nothing particularly sophisticated or difficult to manage, but... Suddenly, instead of one country, we're running a campaign for 10 countries. It turns out that we need to duplicate activities for each country. 10 campaigns suddenly become 100 campaigns, each of which needs to be reviewed and monitored. Google isn't everything; we want campaigns for Bing users, and we end up with 150 campaigns. We want Facebook, and we end up with 200 campaigns. And what happens? Our super specialist, who has never let us down, is starting to make mistakes, and we're starting to think our campaigns aren't entirely effective. Let's be honest, our specialist can't monitor and analyze data day after day. Let's be honest, humans aren't machines. So let's ask ourselves again: does it work? Can a machine effectively help a specialist?
My own example
I have all the Google AdWords certifications currently available, as well as a Google AdWords Trainer certificate (I train others), and over a decade of experience in online marketing. I've run several hundred Google AdWords campaigns, from seedlings, watches, and perfumes to forex and loans, where CPCs can reach 100 PLN per click. I gained my first experience running campaigns with budgets of 500 PLN/month, and I've graduated to budgets of several hundred thousand PLN. And today... I have my own agency and several hundred clients, campaigns with budgets of several hundred thousand PLN. Fortunately, during my studies, I had a lot of experience with data analysis, doing some multivariate analyses for perfume testing using an electronic nose (those were the days!), so analysis and statistics are no fear for me either. I conducted an experiment. I connected all my online campaigns to the VCM/DoubleClick platform and continued running them with machine support. What was the result? Who won? Unfortunately, I had to bow my head in humility. The machines defeated me in every case. Sometimes the results were slightly better, sometimes unfortunately, I was "massacred," and the results month over month were much better. Where is the main advantage? I can make changes daily, create rules for bid changes. Periodically, I can export the data to Excel, recalculate the data, and based on the results, make larger optimization changes. But I can't do this day after day, several times a day, at the level of every single parameter. While I would normally make a thousand changes in a single campaign, the machine made 30,000 of them, based on advanced statistical models, day after day. At night, on weekends, in the middle of the day. No one can keep up with this, and from this arises the full power of optimization using a machine like VCM. From many thousands of small, planned and calculated micro-changes. This means the machine makes better decisions than a human, so in this case, calling such software "small artificial intelligence" is not such a mistake.
What's the catch?
Even though the machine outperformed me, I was the architect of its success. I set it up, told the machine what to do, for what purpose, and monitored it daily. Without my knowledge and supervision, it couldn't have been better. Here's the secret: the machine won't replace a specialist, but it provides them with tremendous support. It will perform the most difficult, tedious analytical work daily, but it must be configured correctly. This is an incredible solution: the human first teaches the machine, then observes what the machine does and learns from it. This allows not only the machine but also the human to improve.
Are there any threats?
These machines require highly informed users; simply connecting them to existing campaigns can do more harm than good. This must be done very skillfully and properly. It may turn out that a given campaign will only be ready for integration after a thorough overhaul.
Is it expensive?
Yes, it's expensive – we're not saying otherwise. The good news is that when an agency implements such a system comprehensively, it doesn't significantly impact the final costs for clients. For small budgets, the fee for using a system like VCM is included in the price, so a small client with a budget of 1000 PLN receives a super specialist at no extra cost, while for large budgets, we're talking about an additional fee of several percent. This is key information; you can gain a competitive advantage today and receive additional support for your company without incurring significant additional costs. This solution is close to Verseo's philosophy: effective assistance for companies in online marketing while maintaining reasonable service prices. Speaking of systems based on "machine learning," let's take a look at what we can do now and the current situation. We'll be focusing on Google AdWords, as it's one of the main products whose performance can be improved by machines.
We will divide the individual solutions according to the degree of automation and optimization: 1. At the beginning we have
Google AdWords panel, through which we can create campaigns and "manually" set bids in our campaigns. 2. Larger campaigns are run using
Google AdWords Editor and Excel, using filters we can change rates en masse (i.e. also "manually"). 3. This category is not on the chart, but we can in
Google Ads using rules or scripts to create
automatic rate changes For example, if the position drops below 4, increase the bid by PLN X, if the position is higher than 2, decrease the bid by PLN Y. You can create really advanced scripts here, but they are still just rules; they don't "learn." 4.
DoubleClick Search – Google's official campaign optimization product with pre-built bid optimization strategies that makes changes to our campaigns up to four times a day. This system uses self-learning algorithms. It's a very good product that works efficiently (DoubleClick is a whole family of tools). 5. VCM –
Verseo Campaign Manager – a comprehensive platform for automating and optimizing online campaigns, including ready-made bid optimization strategies and the ability to manage a large number of campaigns.
Google Smart Bidding Google's algorithms for real-time bid optimization ("auction-time bidding"), a combination of self-learning algorithms with bid optimization for each individual auction and user. This category could be placed alongside VCM and DoubleClick, as these systems can work together (and they can be used with VCM and DC in parallel). Google also considers many signals unavailable to either platform. What we've discussed in this article applies to points 4-6 from the list above. DoubleClick Search and VCM are, on the one hand, very similar products, but on the other hand, they differ significantly. They can't be used simultaneously (in terms of bid optimization strategies), as they operate similarly. Why did I prioritize VCM? Because DoubleClick isn't suitable for mass campaign management. It's suitable for large, individual campaigns, but I can't imagine an agency handling 1000 clients using DoubleClick. Therefore, VCM, as a platform, has an advantage over DoubleClick Search. DC was also designed with large clients in mind. VCM handles smaller budgets significantly better than DoubleClick Search, and for large campaigns, it's just as effective as DC. Our algorithms are better suited to the specifics of a smaller campaign. Why is this? Few agencies connect small budgets of PLN 2-6 to DoubleClick Search. This is VCM's main advantage, as Verseo has acquired the knowledge and experience in managing campaigns with both small and large budgets. Another advantage is the knowledge we gain from daily collaboration with clients, which DoubleClick/Google simply can't. By running online campaigns using VCM or DoubleClick, we're currently discovering entirely new optimization fields and gaining access to areas that were previously unavailable to us, as machines provide us with new knowledge. People used to say that 2015 was the year of "analytics," 2016 was the year of "mobile." The very names suggest they're no longer up-to-date. It's finally 2017 – the era (not just the year) of machines. This is the era that will change everything. We are pleased to present to you the first machine manufactured by Verseo (the first implementation of this scale in Poland).
VCM – Verseo Campaign Manager – detailed descriptionSummary
The above article covers the following topics:
- Machine learning is a revolution in online marketing – it allows companies to effectively optimize advertising campaigns without having to manually analyze huge amounts of data.
- Analytical tools alone are not enough – without a dedicated analyst, most companies do not have the time or knowledge to use the collected data to improve campaign results.
- The solution is automation using machines – self-learning algorithms can perform tedious analytical work 24/7, introducing thousands of micro-changes per day.
- Machine learning is a system that collects data, learns from it, and independently makes optimization decisions – it does not replace a specialist, but provides them with powerful support.
