Humanteq has learned to divide players by psychotypes. Thanks to this, the CPI decreased by 60%

Alexander Odainik, co-founder of a marketing startup, told us about how the new tool works.

Alexander Odainik

The essence of the solution

Developers of free-play games, when buying traffic, tend to acquire users who are most likely to make a purchase. The problem is that with point targeting, the cost increases significantly.

For example, we know that women pay more often in our application, then when creating campaigns we will limit the gender so as not to show ads to men. Then there is additional targeting, but already by interests, geography and some technical characteristics. All this narrows the funnel and leads to a rise in the cost of the user.

To solve the problem of reducing coverage when segmenting by behavior within the application, we at Humanteq decided to use psychology. We were guided by the fact that psychological characteristics are universal for all people, they do not depend on gender, age, geo.

We also considered that psychological characteristics can indirectly predict behavior, since they determine people’s propensities to commit certain actions. If this is the case, then we can:

a) identify people who are psychologically inclined to make purchases in our application;

b) use psychological lists rather than behavioral lists to buy a new audience.

For example, we can determine that extroverts are most likely to make purchases in a particular product (i.e., all extroverts are more likely to make a payment on average than the entire audience).

Next, we will not look for users who are similar to the players who paid in our application, but extroverts in general. This audience will include both paying and non-paying, but at the same time inclined to pay. Such an audience will be much larger. This will reduce the cost of users, but at the same time preserve their quality.

To determine the psychology of users and train models, we use our Digital Freud application. In it, users take psychological tests, such as OCEAN (Big5), the Schwartz test for life orientations, an intelligence test, and others, and get their psychotype, which they can compare with others, or share on social networks. In parallel, we collect technical data of users’ smartphones, such as the device model, OS versions, lists of installed applications, etc. and we are building a model that, according to technical data, restores the psychological portrait of the user.

Nuances of Work and Facebook Ads Look-alike

An important caveat: Humanteq only allows you to find out the psychotype of the paying audience. Next, you need to use third-party marketing tools to acquire users.

As part of the case below, we used Facebook Ads Look-alike to search for and purchase players of interest to us. This is one of the most popular solutions to attract target users. Its principle is to find the most similar to the desired audience.

But we could use any other advertising solution that supports GAID identifiers of Android devices (such analytics are not available on iOS yet).

When working with Facebook Ads Look-alike, the scenario usually looks like this:

• the developer creates a list of users who have already made a payment in the game;
• unloads them as GAID lists;
• uploads GAID to Facebook Ads Audiences;
• creates an audience based on Look-Alike lists.

Then, through internal ML algorithms, Facebook finds an audience similar to your list. After that, you can launch an advertising campaign.

Case

The effectiveness of work on psychotypes is well demonstrated by our joint case with Beresnev Games.

The Prague studio has a game Gallery: Coloring Book & Decor. This is a coloring book by numbers with a meta in which the user arranges his house.

We integrated the Humanteq SDK into the game to get psychological profiles of players and select a target psychosegment, i.e. a set of psychological characteristics that maximizes the probability of conversion into a purchase.

After we identified the psychological profiles of more than a million users, we analyzed them for conversion to purchase and identified the two most significant segments that are more likely to be converted into paying ones.

The first audience was people with a pronounced value of achievements and high impulsivity. Such people are 150% more likely to make a purchase than the rest of the audience.

On the one hand, these are achievement-oriented people. They want to prove themselves and succeed. They like to demonstrate their worth both intellectually and materially. They are motivated to achieve goals, including game goals.

On the other hand, they differ in their willingness to consume impulsively, are characterized by a refusal to plan their resources (time and finances), spontaneous actions and a tendency to procrastination.

The second dominant psychosegment in the game were people with a high IQ and a pronounced value of hedonism. In this psychosegment, the probability of conversion to purchase is 211% higher compared to the rest of the audience.

High intelligence is, firstly, general giftedness, characterized by the speed of solving intellectual problems and the ability to find patterns, and, secondly, the ability to enjoy solving problems.

Hedonism is the pursuit of sensual pleasures, the desire to pamper and please yourself. It is associated with the willingness to buy something beautiful, even if it does not have a utilitarian function.

After we separated the lists of users by psychosegments separately from each other, we also uploaded standard data on the paying audience. The latter became our control group for the test.

In total, we got three audiences:

• with a pronounced value of achievements and high impulsivity;
• with a high IQ and a pronounced value of hedonism;
• players who have previously made payments to Gallery: Coloring Book & Decor.

Then we built Facebook Ads based on their Look-alike. It turned out that the audiences overlap significantly (the advertising tool of the social network allows you to visualize this):

Their similarity spoke of two things:

• when conducting separate advertising campaigns for each of the audiences, we can expect similar behavior;
• we can’t run a simple test comparing the target audience: with a strong intersection of audiences, problems may arise with the auto-competition of advertising campaigns with each other for the same users.

As a result, we had to exclude audiences from each other in each Adset and run tests sequentially.

Thus, we have the following plan of experiments:

• we compare the psychosegment with the value of achievements and high impulsivity (group B) with the audience that made payments (group A);
• we compare a psychosegment with a high IQ and pronounced hedonism (group C) with an audience that has made payments (group A);
• we compare two psychosegments with each other (groups B and C).

All adsets and campaigns differed only in their audiences. The other settings were identical. Each campaign was optimized for payments and excluded audiences from each other (psychosegments from paying and vice versa). All campaigns lasted the same amount of time, their budgets did not differ.

Results:

It turned out that in comparison with Group A (Look-alike for paying users) both psychosegments showed:

• reduced cost per impression, click and installation;
• saving the price of one paying user (CPA);
• reduction of conversion monetization, ARPPU and ARPU;
• the payback of a segment with a psychosegment that has a high IQ and pronounced hedonism has increased, since its CPI has decreased by 60%. That is, we were able to attract more audience, which for the most part brought more revenue.

By the way, the CTR of both psychosegments turned out to be higher than that of the audience of paying players. This can be explained by the fact that the developer has repeatedly launched campaigns targeting Look-alike to paying users. In other words, most of the paying audience has already had ad impressions. Thanks to the use of psychosegments and the exclusion of the audience from each other, we found a new audience that we had not previously shown advertising to.

We conducted this case on Android. The fact is that the Humanteq SDK so far allows you to work only with this platform.

But we assumed that if the audiences of the application on different platforms are similar, then Look-alike should work equally (or in a similar way) on iOS. To do this, we launched a campaign for a psychosegment built on an Android app on an iOS app. We used the same Look-alike in the iOS campaign that were used in the Android test.

As a result of this campaign, we have reduced CPI by 47%, also with an increase in CTR by 24%. Monetization metrics also turned out to be lower than in the Look-alike campaign for paying users. But we are also seeing a 14% increase in ROAS.

General conclusions

The main thing:

• thanks to psychosegmentation and its application in Look-alike, it is possible to reduce CPI (in our case by 60%);
• CTR is higher on average on psychosegments, which allows you to show creatives to a new audience;
• despite the drop in conversion, ARPPU and ARPU, the volume increases significantly, thanks to which it turns out to save ROAS.

So we believe that psychosegmentation effectively allows you to search for large volumes of new loyal audience, which will provide monetization indicators comparable to the already acquired paying audience.

---