Recommendation System.

In most iGaming products, the recommended games section is generated in one of two ways: either it’s a list of the most popular slots on the platform, or it’s a selection curated manually by the operator’s team. Sometimes it’s a hybrid approach: some games are added manually, while others are selected algorithmically.

This approach is understandable and familiar. But it creates three key problems:

1. All players see the same list, regardless of their preferences.
2. Curated collections require constant manual updates, which is labor-intensive and does not scale well.
3. A static list of games becomes the "default" option rather than a driver of growth.

As a result, the set of recommendations does not consistently help increase retention, LTV, and revenue.

Objective: To increase the operator's revenue and improve the user gaming experience through personalization

To achieve this, MICo (developers of AI and ML solutions for iGaming) integrated the MICo recommendation engine into the product of its client—a gaming operator. The ML model* analyzed each player’s behavior based on a variety of signals and generated personalized game selections.

‍Region:
South Asia

‍Audience:
250,000+ MAU

‍Point of entry:
The “Recommended Games” section, located in the operator’s lobby on the top level; this is one of the key navigation elements that receives a large number of views.
‍*ML (machine learning) — models that analyze data, identify patterns, and make decisions without manual intervention.

The effect was evaluated by comparing two operational scenarios:

1. Manual curation: The list of recommended games is compiled by the operator team.
2. MICo ML Recommendations: The composition and order of games within the same block are determined by an algorithm.

Importantly, the design and layout of the block remained unchanged. No new mechanics were added, and traffic levels remained the same. The only difference was in the logic behind matchmaking.

But to understand exactly what the differences between the two approaches were, it’s worth taking a closer look at how the MICo solution works.

The ML recommender generates a list of personalized recommendations on its own. Here's what happens

• It collects data. The player selects titles, clicks, returns to games, and changes the pace and style—the model analyzes over 200 behavioral parameters for each player.
• It creates a pool of games. Based on the signals, several independent algorithms select relevant games.
• Creates a personalized list. A selection is compiled that reflects the likelihood that the player will be interested in specific games.
• Ranks with a focus on value. The list is reordered to align with the operator’s business objectives—maximizing GGR and ARPU.

Unlike manual curation, which relies on general statistics and past results, ML analyzes the behavior of a specific user and dynamically selects search results, eliminating the need to constantly manually update the list.

Within the first 28 days, the ML-recommendations scenario showed significant growth across key metrics

All metrics were calculated solely for the “Recommended Games” section; in other words, this represents the net effect of the change in the selection logic.

Behavioral metrics:

• Conversion rate from views to clicks: +8 percentage points
• Day-7 retention: +10 p.p.

Financial metrics:

• Average deposit: +40%
• GGR per user in this section: +200%

The results confirm that a recommendation section can be significantly more effective if the selection process is dynamic and based on players' actual behavior and preferences.

What does this mean for operators?

This case study illustrates a simple pattern: recommendations can be more than just a “convenient feature”—they can be a direct driver of revenue if they are adapted to the player’s behavior in real time.

Personalization acts as a real financial lever:

Sustainable growth without increasing your budget. You don’t spend your budget on marketing or hire additional staff—yet your metrics grow because players are getting a more relevant experience.

Reduced manual labor. There’s no need to constantly update lists or allocate resources to do so. The ML solution does this automatically based on real-time data.

Scalability. The more players and events there are, the more accurately the model works. Recommendations can be used everywhere: in the lobby, in search results, in the game list, and even in communications.

The more touchpoints there are, the more noticeable the business impact.

How to Repeat the Result

Test the ML recommender on your own data:

• Duration: 2–4 weeks
• Format: Launch a pilot and evaluate the results
• Result: Measurable uplift and projected growth in metrics

Personalizing recommendations isn’t just a cosmetic improvement. It’s a sustainable growth channel that starts delivering results immediately after launch, without any additional marketing costs. The ML model replaces a static list with dynamic results that adapt to each player, and this has a tangible impact on revenue-related metrics.