Portfolio startup case: INCYMO.AI

Anna Zdorenko, the founder of?INCYMO.AI, talks about their AI-powered solution that enhances ad performance and audience acquisition for mobile games, boosting the number of high-performing ad creatives tenfold and skyrocketing the revenues. The startup addresses the common challenge of improving game ads on social media and in paid search. Despite launching recently, INCYMO.AI has generated significant interest from potential clients in gaming and is planning to scale to the general video ad industry.

Anna Zdorenko, INCYMO.AI CEO:

The bottleneck of mobile games launches

Game companies lose millions of dollars on ineffective ads. I spoke with user acquisition leads, CMOs, owners from 40 companies in the gaming industry, some having marketing budgets exceeding $20M, and they all expressed this pain point.

The success of a game launch is determined by its customer acquisition, no matter if the revenue is generated through in-app purchases (IAP) or in-app adds (IAA). In mobile gaming, the primary user acquisition channel is paid traffic from social media and search. The worldwide ad expenditure on mobile gaming is projected to reach $80B in 2023 and $130B in 2025. Gamedev studios often spend?more than 50% of a game's budget on user acquisition.

But what can a company do to optimize this budget? Google’s and Meta’s ad platforms offer limited transparency and control over campaigns, making the quality of ads the critical lever for success. The difference between good and bad ads can make or break a game launch.

The ads are short videos with sound, often interactive. Game design studios face the challenge of producing the best, most engaging and highest-performing ads possible. But the ads’ performance burns out quickly, what was working yesterday won’t attract anyone tomorrow. You constantly make new ads that you want to stand out.

Producing and renewing multiple versions of ads for each target audience, consistently experimenting with hundreds and thousands of creatives requires a specific skill set in design, fine-tuning digital ad campaigns, analytics and management. As a result, there is a whole market of dedicated agencies producing and testing visual ads. These agencies are an outsource version of in-house design teams. Outsourcers and in-house teams dive deep into each game, learn from their mistakes and keep track of all their experiments. Companies that do not engage high-level ad teams end up creating random creatives, hoping that they will work, instead of treating them as experiments that will help their creatives evolve.

Yet, even working with a top agency doesn’t guarantee success. The assembly line approach to ad creation is a significant challenge. Creative teams must think outside the box and create visuals that stand out and engage each specific audience. But how to scale this approach and not have your design team burn out, when they run out of ideas? How to effectively use your previous experience and not overwhelm your team with analytical work? How to keep producing engaging and performing creatives in a consistent ongoing manner?

This is why user acquisition expenditures lack transparency and predictability. It is common for companies to develop a thousand visuals to test and pay for advertising them, only to find that none of them work.

As a team with ML expertise of more than 10 years, we know that ML can analyze and provide solutions on a large scale. But I wasn’t even surprised to learn that ad agencies do not use any ML-based algorithms to assist them in this process. You need a combo of experience in video creative production, the knowledge of game design, psychology and analytics, and also to be good at creating ML models. The world needed someone to save the day.

Developing smartUA

I teamed up with a partner experienced in game ad creation automation. We became co-founders and produced smartUA.

smartUA helps game studios and agencies create better-performing video ad creatives. It's ideal for casual and mid-core free-to-play mobile games and we plan to expand to hyper-casual, hardcore or even general marketing in the coming future. The algorithm analyzes the game type, target audience and previous ads used for user acquisition earlier by the game’s team. These previous ads are a starting point of improvement.

Here's how it works:

• Data preparation. We describe the game for the algorithm by tagging (specify its type and many other parameters that describe the game for the algorithm). We also describe by tagging the previous creatives that the user acquisition team had used.
• We feed these data into the algorithm. We also feed in the audience’s profile (age, gender etc), geolocation, mobile platform and other information.
• For each audience, the algorithm suggests several ad scenarios. By an ad scenario we mean a text description of a video with a sequence of events, visual elements, narratives, game elements, mechanics and modes to incorporate into the ad that the algorithm “believes” will be perfect for presenting your game.
• You give these scenarios to the team working on video production: it’s either us, your own team or an agency that produces videos for you.
• Once videos are ready, the testing starts. In our recent cases, when we were running the proof-of-concept stage, our clients had to run tests themselves. But now we are launching the new version of smartUA that manages the testing cycle on Facebook 24/7, automatically. It starts the ads and then it stops the underperforming ones once their underperformance gains statistical proof. This lets you save the budget on testing without manual supervision.
• For the ads performing best against your ad KPIs, the algorithm extracts the scenario elements that provide the top performance. It updates the recommendations for the ad scenarios with more details to test.
• In the next round, the algorithm suggests better versions of scenarios based on the knowledge accumulated in previous steps. It suggests variations based on the best creatives of the previous step. We repeat as many rounds as you need.

Note that you already have ad scenarios augmented with the AI even in the first generation, before any actual “selection of the fittest” took any place.

We collect our own database, enriched with public data on what visuals, colors, narratives, sequences, game elements, mechanics and modes shown in the ad work for each kind of audience, industry, genre, geography, platform, performance KPI etc. Our database is increased with each client, which allows our neural network to be trained at a higher level.

We don’t need to produce creatives ourselves. Many agencies and studios are happy that we just suggest scenarios, and they make creatives on their own. But we do have partners ready to make creatives in line with the algorithm’s suggestions, so we can provide a turnkey solution.

The ultimate value of our product is that it lets studios focus on production and takes away the burden of maintaining a consistent system of logs, experiments and analysis. You see results very quickly, in a matter of weeks after the start. We guarantee that in just a few iterations you will find at least one creative that performs well according to your KPI. And after this, in the next iterations, about each 10th creative will also perform well.

Sometimes the creatives our clients had before testing with us, the ones against which smartUA was competing, were really good. But we know that it took many hundreds or even thousands of other creatives for them to throw away before they found those good performers. Our model learns in just a hundred creatives that you need to produce, and then it provides a stable ever-improving result. We are proud to say that in some of our test cases, even our first-generation creatives performed better than the ones the clients previously had.

Highlights

• We have been developing the algorithm for 5 years. From 2020 to 2022 we have validated it with 4 clients. While they had one good-performing ad creative in a 100, with us, they now have 10 out of 100 good performers. One of the clients raised its revenue from $100k-150k to $600k-900k.
• We launched smartUA, the automatized product, in December 2022.
• Since then, we have attracted our first 8 clients, including a major game producer of more than 20 games.
• The conversion to trial, the share of studios we talked to that want to test our solution, is ~70%.
• Revenue growth: $3.3K (Dec 2022), $8.4K (Jan 2023), $11K (Feb 2023).

Future development

We're currently attracting investments as we aim to scale our client base and further train our ML models with new high-quality data for even better performance. Our team of 16 is fully equipped to handle this growth.

We've managed to attract 8 clients within three months without any marketing efforts, which is a promising start. Our goal is to onboard 500 game companies in the next three-five years, which we believe will help us achieve our ambitious target of becoming a unicorn in the industry.

But there’s more to do for us. Gaming was a good market to pilot our AI tailored for video ads. But we have the ambition to adapt our technology to the vast and expanding market of general video advertisement. Imagine ads powered by our AI on the iconic Times Square screen. This is not just a dream, but a tangible benchmark that we are working towards.