What do you first think of when we say liquidity?

When I initially encountered this term on Facebook Blueprint learning, the association was the flow of money and finances. This is not what’s meant here. Namely, in this concrete case, liquidity is linked to advertising and its systems of machine learning. 

These systems have been used in various industries for some time and different purposes. How does this all play out in the context of digital marketing? 

Well, we use this system to help us make important decisions and optimize in real-time, which is related to advertising and ad bidding on the platform. Many different factors determine the price of the ad, considering that the auction system is very much alive and changing fast. In any given moment in time, thousands of advertisers compete in an attempt to reach the desired audience. It’s clear we require a scaled and systemized approach to the algorithm. Also, in order to achieve the best results, there is a need for proper choice of device, platforms, and end destinations for deploying ads. This is, in essence, the definition of machine learning on the Facebook advertising platform. The machine gives us the answer to the question of which ads will fare best and how to allocate the budget and creative in the most efficient manner. With the aid of this system, we’re allowed to create flexible campaigns without restrictions to impede our results. You can set up everything on autopilot, keep the engines running, and give the algorithm the freedom to manage all the options and seek the ones that yield maximum results. Here, we touch on the topic of the advertising platform’s liquidity. 

Liquidity in advertising revolves around letting the machine do as much as possible. That is to say, setting as few resections as we can during the development campaign and giving a large manoeuvring space to the algorithm, which can then start leaning, evaluating results, and harnessing data without any limits. Information flows unhindered on all fronts. 

For instance, let’s say we have a campaign that targets certain people. What will happen if we strictly confine this target group?

We would certainly reach a fewer number of people. If we broaden the target group and are less restrictive, we improve the flow and liquidity. We invest the algorithm with more freedom to come up with the best target audience in real-time. 

To ensure the system of machine learning operates as intended, we first have to know what our goals are. We need to think strategically, knowing what we want to achieve. Do we work on brand awareness or aspire to spur sales with our advertising? Before setting up the system and feeding it information, we must get these aspects clear. The next matter is related to choosing what kind of creative solutions we use to present ourselves. These solutions and goals aren’t something we can create in a machinelike fashion. They are all about us and ideas in our heads. Facebook cites one good example. If you employ navigation to find your way to the store, the navigation device (algorithm) will take you there via the shortest route. But, before all of that happens, you have to tell the system which store you wish to visit. 

Thus, before anything else, we have to pose the questions: where am I and where do I want to go?

The machine is rather useless if we don’t have a developed strategy and power of strategic reasoning. A conclusion to draw is that it’s necessary to feed the algorithm with data and that this data must be quality and relevant. It also has to exist in ample quantity, so that the system can learn based on it. After the learning phase, the system is ready to make various decisions and optimize our campaigns in real-time. 

For example, a lot of data that feeds the machine comes from a website. How do we figure out what activities take place on our website and how they affect the machine?

We can learn this if we have Pixel installed, a piece of code that initiates different kinds of events that measure conversions. Pixel gathers information and is capable of letting the system know about the behavior of users on the site. It helps us understand what actions occurred and when. Also, we have SDK, a program that is similar to Pixel with the distinction that it tracks activities on apps. 

Tools for Liquidity 

There are four tools that narrow or expand the freedom of the algorithm. 

Optimization according to campaign budget— this type of optimization distributes the budget not based on ad sets, but on the campaign level. That means that the system is in control of every ad set. For instance, let’s assume the budget for one set is depleted, and there is some leftover in the other set. In case the people from the first set go online, that’s a chance to gain results. Allocating money to sets prevents us from reaching those people, while campaign-level distribution sends the money to where the new potential buyers are. Basically, the budget is spent where the goal of the campaign is the cheapest to meet. Such a spending model is the official recommendation of Facebook for structuring campaigns, and it surely gives the algorithm more freedom to make valid decisions in terms of optimization. 

Creating an audience— the tool for creating your audience is very powerful. You can capture people in various ways. Take the example of those who added something to the cart, but didn’t purchase anything. They can be later targeted with an offer to buy precisely that product they added. Again, you can broaden the targeting scope and find people who are similar to them, looking at demographics or geographical location. You could discover consumers who exhibit specific behavior or interests, people who are married or engaged, parents, individuals born in a certain month, or their friends. The possibilities abound, but we can’t go overboard and give too much information and restrictions to the system. We have to leave enough room for the machine to look for new opportunities. 

If you input user interests, you can input more info, as you expand the potential target group that way. The algorithm recognizes that info as an aggregate union, where every aggregate represents one interest. If you add people similar to the ones that performed an action (lookalike audience), don’t impose restrictions regarding the city or interests. That’s too much information, and the flow will get thinner. 

Automatic ad deployment— There are four big platforms for deploying ads: Facebook, Instagram, Audience Network, and Messenger. All of these platforms have their subgroups. With the automatic deployment, your ad will display on all destinations within the Facebook system. The liquidity of machine learning comes in very handy then. It prompts the systems to focus on destinations with the lower costs of deployment. If we launch an ad on Facebook alone, shut Instagram down, there is a chance we potentially turned the cheaper destination associated with better results off. Putting more deployments in place increases the likelihood of finding more low-cost conversions, wherever they may be. Another crucial thing is that we don’t have to change the whole campaign when the result is poor or too expensive. All in all, this is a more streamlined process. 

Creative optimization— the next point in the process of automatic distribution of ads we have to ponder is how our creative looks on various channels for distribution. The dimensions for visuals on Instagram and Facebook stories differ from visuals on news feeds. The platform enables us to configure one creative, and the system will automatically adjust it for different deployments. This option is handy, but it’s not always recommended or an overly fortunate solution. There is another better option. We can directly set up visuals for deployments. 

What we have as a possibility here is the dynamic optimization of language. However, speaking from personal experience, I wouldn’t advise it. The system is very clumsy when translating from one language to another. 

A more prescribed option is undoubtedly a dynamic creative setup. We can insert different texts and visuals into one ad, complete with different headlines and calls-to-action (CTAs). Based on this information, the system creates different combinations of ads and makes decisions about what creative to focus on depending on the results. Again we come to the outcome of giving the machine the freedom even in aspects like ad design. There is definitely more than one level of liquidity. 

When advertising on the Facebook platform, we see that a lot of stuff is going on “under the hood”. Different tools for optimizing the budget, audience, deployment, and the creative itself are beneficial and make the system of machine learning more flexible. They also make it more liquid, improves the flow, and ultimately, they let us achieve better results and at a lower cost. 

In reality, we raise the effectiveness of algorithms via quality data. That is the only way to turn the machine to work to our advantage. We are the navigators, and if we don’t navigate well, the machine won’t know the way. 

If you require support in social media advertising, contact us.