As a result, the app managed to raise 500,000 installations within a year at a cost of $23 per purchase.
ROAS 7d was 130%.
The most traffic came from four platforms: Google, Facebook, Unity Ads and TikTok. We continue to promote Pheon on these platforms.
With a two-week frequency, the company releases updates, such as coins that can be used to buy additional services within the app. We help test innovations. Following each update, we take a cut and see how the new product has affected the return on investment of the attracted traffic.
To enable the client to quickly evaluate the testing results, our analysts update the report on advertising placements efficiency on a daily basis. It involves all the key metrics, data analysis and recommendations. In addition, we regularly host video conferences with the advertiser, where we discuss the testing results and make joint decisions on further actions. There is a separate chat in Telegram for urgent questions.

The Pheon AI app promotion was launched on nine platforms at once: Google, Facebook*, TikTok, X (Twitter), Snapchat, Apple Search Ads, in-app services Unity Ads, IronSource and Mintegral.

Each platform had its own characteristics, which we took into consideration during the promotion.

1. To improve promotion in Unity Ads, we allocated a large budget for launching by CPI, identified several geopoints, and directed traffic to installations. Having received a sufficient number of purchase events, we optimized advertising for ROAS. This enabled us to show ads only to those users who were most likely to buy a subscription.
2. At Facebook*, the most vital part of the job involved making creatives. These took up about 80% of the time.
3. Early on, integration with Google was set up through AppsFlyer and advertising campaigns were launched with event optimization. We achieved good results, but then the traffic started to behave unstable and did not always pay off. We discussed with the advertiser the integration of the Firebase platform into the app to open up value optimization in Google. After launching advertising campaigns with optimization for the value of actions, we managed to achieve stable ROAS results and increase the volume on the source by 2.5 times.
4. TikTok encountered a problem with creatives: the source did not accept videos that it considered explicit. For example, if the model had a deep neckline or an exposed stomach. We had to change creatives and adapt to the platform requirements.
5. Apple Search Ads had the option to only run on keywords. Still, the cost per installation was too high, and the advertising did not pay off well. Therefore, we decided to shut down the service.

At the first launch, the advertising budget was allocated equally between the platforms, and after a few days, a cut was made and the funds were redistributed. Further on, we continuously assessed the efficiency of advertising using five parameters: CPI, creative CTR, creative IPM, purchase CPA, ROAS. The budget was increased in campaigns that achieved good results, and those with no prospects were turned off.
On Facebook and Google, geotargeting was set up for both the world and specific locations. For example, we launched a worldwide campaign and, if it paid off, we looked at which countries the system itself would distribute traffic to. Next, we allocated these points into separate campaigns and tested out their performance.
During the campaigns, we tested different creatives, gender, and age settings. We also performed based on specific events:

• installations;
• installations and actions, such as paying for a subscription;
• actions, such as completing onboarding.

On certain platforms, advertising was launched for look-alike audiences. The system tracked users who purchased the app from us, and then, based on this database, it collected similar audiences and broadcasted advertising to them.

We set up and optimized advertising campaigns

Using the Pheon chatbot, you can chat with a digital copy of any person — from Winston Churchill and Spider-Man to a completely realistic girl who could live next door. To reach the maximum number of users, we have prepared several creatives that could potentially be of interest to various people.
For example, for users who are looking for a virtual girlfriend, we have made videos with beautiful girls.

We showed potential buyers that there are different girls in the app
For users who would like to chat with real people, we showed how to create a character based on a photo.

You can upload a photo of your great-grandfather to the Pheon chatbot and talk to him

Since the app key feature is the ability to create any character, we decided to develop animated 3D creatives. Despite the complexity and high cost of 3D development, we completed this task on our own thanks to our in-house production.
This video shows that you can build a certain kind of relationship with the virtual character: communicate with her as a stranger, a friend or a wife
Since the beginning of our cooperation, we have made and tested 1079 video creatives and 816 advertising banners.

We prepared creatives

We tested a number of hypotheses while working with the client. For example, we offered users different subscription periods, discounts on purchases, and in-app surveys. Having analyzed the competitors, we suggested expanding the questionnaire for new users. The ten-question survey resulted in 20% more purchases than the shorter one, which brought in three to four.
When onboarding new users, they were shown different interfaces, offered variable avatars, and demonstrated the constantly changing main page splash screen. This enabled us to figure out which images worked best and at the same time introduce customers to all the features of the chatbot.
This is what one of the start page options looks like. It contains a maximum of useful information to make it easier for users to master the app.
The testing results were taken into consideration when adjusting advertising campaigns. Successful solutions were scaled up.

We put forward hypotheses

We analyzed the app and created a profile of the user who is most likely to purchase a subscription. We found out that 80% of our audience are single men over 30 years old.
Two main groups of people were identified:

• Some people chatting and flirting with women and would like to have a virtual friend or girlfriend.
• Some dream of talking to their idols or famous characters, such as comic book and anime heroes.
We used the collected information to develop advertising creatives.

We created a target audience portrait

Using the Pheon app, you can find a virtual friend or create a digital copy of a real person. For example, you can upload a photo of an ex-girlfriend and chat with her. The bot can write text, record video, and can also be easily customized — given unique character traits and taught how to respond.
The user can select the character category for chat. This can be a legend, an anime hero, or an ordinary person
When the client approached us, they only had an iOS app with minimal promotion using in-house resources. We suggested developing an Android version of the service, which proved more successful later on. The main objective was to attract the maximum number of solvent users to the app.
Since the client receives the main income from in-app purchases, the product payback is especially important for them. Net Revenue was supposed to be $ 50,000-$ 100,000 per month, and the main KPI was ROAS.

In April 2023, Pheon company approached us with a request to promote a new chatbot featuring built-in AI. This was our first time promoting an app like that, but everything worked out thanks to an experienced team and an efficient work plan.
We will guide you through why you need to find out the interests of your target audience, whether it is difficult to develop creatives for an AI-featured app with, and what you should consider when setting up advertising campaigns. The case was written with the help of the head of the media buying department Ilya Arinichev and the director of client service Kristina Ostrovskaya who accompanies this client and is involved in the strategy development.

How to Raise 500,000 Installs to an App for Chatting with Real-People Avatars

500,000 Installs in a Year: How we Promoted an App with Built-In AI