When to use Google Ads Smart Bidding, and when not to.

Discover why it’s important to have a toolbox to meet all challenges that arise.

Arben Kqiku, acount manager at comtogether

By ARBEN KQIKU

What is Google Ads Smart Bidding?

Machine learning bidding strategies are being sold to companies as a panacea for all their problems. In practice, however, there are good and bad ways to apply this new and powerful technology. Far from being the universal solution it is made out to be, Google Ads Smart Bidding, for instance, requires careful monitoring and selective use in order to yield results. If you don’t know what you’re doing with it, machine learning may lead you to flush much of your precious advertising budget down the drain. The bidding system at work in the Google Ads ecosystem holds automated auctions for keyword-associated advertising space. Control of these spaces are the subject of fierce competition among advertisers, and a bidding strategy can make—or break—the overall success of a campaign. The amounts that are bid can be fixed manually or this decision can be delegated to an automatic strategy known as Smart Bidding, Google Ads’ machine learning algorithms.

Garbage in, garbage out

There is no doubt that AI-powered machine learning algorithms are among the great game-changers in online marketing today. They can yield truly remarkable results and every marketer should consider them as a viable option. The success of these algorithms, however, is highly data-dependent. They need the right data in order to make good bids. It’s the old garbage in/garbage out principle, where bad quality input predictably corrupts the output and leads to faulty results. Most of us will have already experienced the effects of this ourselves. For example, let’s suppose that someone you know uses your computer to research the price of baby strollers. From that moment onwards, you begin to notice that many of the ads being displayed to you have something to do with pregnancy, maternity, and babies. You have no use for these products because, unlike the person who used your computer, you don’t have a child on the way. Your data profile now contains inaccurate information. To get around the garbage-data problem, these small inaccuracies need to be filtered out. This can only be done at scale. The more data, the better. Machine learning, therefore, not only needs quality data, but lots of it too.

The initial short-comings of machine learning in digital marketing

As illustrated by the following graph, machine learning will not provide optimal ROI until it gathers enough quality data, at which point its efficiency will take off as it begins to make truly effective and data-driven bids. This initial data-gathering phase tends to be expensive for relatively few results. For this reason, we call it the depression zone. Nobody wants to go through there, especially when you need reliable results under the constraint of a restricted budget.
cumulative revenue evolution using a machine learning approach

The obvious solution to avoid this depression zone would be to create your own bidding strategy that does not rely on machine learning, but that nevertheless collects data which can later be fed into Google’s Smart Bidding algorithm. Easier said than done. And yet, we have a solution that does just that—a simple solution that delivers during the early-stage slump and provides the data from which the machine can learn and thrive later on.

Human heuristics while the machine learns

As an agency, we work very closely with all of Google’s tools on a daily basis, and we have done extensive testing on machine learning algorithms alongside a number of our algorithms developed in-house. Based on the results of these experiences, we have developed and extensively tested a proprietary tool which allows us to reliably get cost-efficient bidding results even when data is scarce. It is a simple, elegant, and efficient solution based on the science of heuristics.

cumulative evolution of revenue by using a heuristic approach and cost

A heuristic is a computational method designed to solve complex decision-making problems giving preference to speed and approximation, rather than absolute accuracy. While it does not have the same long-term maximization potential as machine learning, a heuristic provides imperfect, but immediate and robust results. Think of a heuristic as a rule of thumb or an educated guess, like guesstimating someone’s weight based on their height. As illustrated in the graph above, our heuristic starts earning as soon as we put it out there, and it does not need to go through the same costly learning period to generate ROI as Google’s machine learning does.

Finding the sweet spot

Let’s say running a marketing campaign can be compared with a child learning to ride a bicycle. Using machine learning from the get-go is like starting out the child with a full size road bike, whereas relying on a heuristic is more like putting training wheels on its sides. Without the training wheels, the child will fall over a lot until he or she adapts and figures out how to stay stable. Using the training wheels avoids the falls and optimizes the learning curve. The question then becomes: when can we take off the training wheels?

Heuristics are beneficial in the early stages of a campaign, and machine learning becomes better later on, provided it has quality data to rely on. The trick, therefore, is to switch to machine learning driven decision-making once enough data is there to guarantee solid outcomes. The vertical line in the graph below illustrates the sweet spot, or inflection point, at which the training wheels should come off and you should switch from the heuristic to machine learning.

évolution cumulative du revenu en combinant une approche heuristique et une approche machine learning

Where exactly this sweet spot is will depend on the situation, of course. Finding it with accuracy and without incurring too much cost is an art more than a science. That said, there is a simple technique we use to monitor for this inflection point. Once we think we have generated sufficient data to feed into Smart Bidding, we run A/B testing campaigns, where the one is set up to use the machine learning, and the other remains on our heuristic algorithm. We then compare both performances and move forward with the most efficient.

Smart Bidding and Beyond

Machine learning is one the most powerful tools out there for marketers, but that does not mean it’s the perfect tool for every job. Every new product is its own case study and every new campaign has a learning curve. It’s important to have a toolbox to meet all challenges that arise.

The point here is not to talk down the effectiveness of machine learning, but rather to understand how and when it can be used most efficiently, and what other solutions may be useful to apply alongside it. Heuristics are one of these solutions, and our proprietary heuristic is a prime example.

Have you ever experienced difficulties with machine learning algorithms or been disappointed with Google Ads? Do you know which of your channels drive the most ROI?

Whatever your questions are, we at comtogether are happy to help. Don’t hesitate to get in touch!

Spread the word

About the author

Arben
Arben graduated with a Master in Psychology. Apart from keeping us in good mental shape, he brings his passion & creativity to data analytics and programs the automations that save our customers valuable time.

You want to learn more about Google Ads ? Discover our Whitepaper ! 👇