In 2026, there’s no more debate about the business value of LiveOps in games. 84% of mobile in-app purchase (IAP) revenues come from games running LiveOps. Games with LiveOps programs deliver 2-3x higher lifetime value (LTV) per player compared to those without LiveOps. Besides, Gartner also projects that 80% of future gaming revenues will come from just 20% of existing players.

Through a recent March 2025 acquisition, a global game publisher and developer (with a LiveOps architecture) surpassed $10 billion in lifetime revenue across its entire portfolio, ranking it among the highest earners in the mobile gaming segment. The earnings included revenues from both self-developed and acquired games. The acquisition involved adding millions of active users (from the acquired game) to their existing portfolio.

While the publisher had no issues managing existing players built on its existing infrastructure, the challenge involved integrating players (from the acquired game) added using a different infrastructure and engineering team.

The requirement was to integrate both sets of players into a single “universe.” While LiveOps works at the title level, the current system does not work at the portfolio level. Here’s our insight into how the Semantic Twin can close the gap at the portfolio level.

LiveOps in games: AI for game player retention

For any game publisher with a single title, LiveOps can easily drive the personalization engine. Overall, most publishers have successfully addressed the need to personalize at the title level. Considering the Gartner projection, publishers have also deployed an effective churn prediction model trained on session frequency, players’ IAP behavior, and event participation.

At the portfolio level (for example, 15 or more titles), publishers need a shared intelligence layer that can understand players across the entire portfolio – and not simply in each game. This notable shift is no longer about targeting (or retaining) the right players but a structural problem that requires gaming infrastructure modernization.

Consider a high-revenue player that works across a portfolio (with multiple titles). Here’s how the current system works in practice:

• The player has a low session frequency for one title, which causes the churn model to flag them for re-engagement and automatically send them a discount offer.
• The player is spending full price on other titles in the same portfolio, but the churn model has zero visibility into this aspect.
• In this scenario, the AI agent for the game (LiveOps) sends multiple interventions to the same player in a short 48-hour window.
• The player receives a discounted offer for a title that they were ready to buy at the full price.
• The publisher ends up losing revenue on a high-value title.

Why companies need the right architectural foundation for AI in gaming

Online gaming companies need an AI-based architectural foundation for modern LiveOps that’s focused around real-time player context. Here are some of its essential pillars:

1. Canonical player entity (CPE)
   This element acts as the single source of truth for any player’s current state across gaming sessions and touchpoints. In place of using disparate databases for storing player records, CPE consolidates the data into a unified system that’s optimized for AI agents.
2. Shared behavioral definitions across models
   This represents a standardized semantic layer that prevents AI agents from interpreting players’ actions differently.
3. Cross-agent intervention memory
   In a complex portfolio, multiple AI agents can operate on the same player base. This requires coordination among agents. Additionally, every player in an active title session must be visible to every agent.
4. Data lineage for personalized decisions
   This ensures that every event trigger and action is traceable to AI-driven player segmentation data and underlying logic. For every AI-driven decision, the architecture must support complete auditability and explainability.

How Onix’s Semantic Twin can deliver the solution

Powered by Agentic AI, Onix’s Wingspan platform and Semantic Twin can help gaming companies build a unified foundation across the entire portfolio. With this semantic layer, each LiveOps agent (across titles) begins from the same player entity, behavioral definition, and view.

Here are some of the capabilities of Onix’s Semantic Twin technology:

Capability	Methodology
Resolves various player entities across titles.	By scanning account systems, device IDs, and behavioral data (across games), it can automatically build the canonical player graph.
Governs the player’s behavioral definitions.	By centralizing the definitions of high-value lapse risks and offering eligibility at the portfolio level.
Connects AI agents to shared context.	As every LiveOps AI agent uses the same knowledge graph, it ensures that cross-portfolio spenders are accurately classified before every model run.
Coordinates agent intervention.	Through cross-agent memory, it can now record every offer and campaign decision. Agents can also check the shared state before firing any intervention.
Traces every AI-enabled decision to its source.	Ensures the full lineage from personalized decision to player’s data and model logic, thus meeting explainability requirements.
Scales with acquisitions.	This ensures that for every new title joining the portfolio, the player’s data is merged into the existing canonical graph (without any new integration).

Conclusion

In the modern gaming domain, publishers with the most AI agents deployed across titles cannot claim to build an intelligent portfolio. Rather, it’s those that can build the shared foundation for LiveOps agents before deployment.

As a Google Cloud partner for seamless cloud migration solutions, Onix has equipped the gaming industry with real-time data analytics solutions to unlock business value. Our Agentic AI platform Wingspan delivers the power of the Semantic Twin to provide contextual intelligence across the entire portfolio.

Are you looking to elevate your LiveOps to a new level of intelligence? Get in touch with us to know more about the capabilities of Wingspan 2.0, or book a product demo today.

Frequently Asked Questions

1. Why is title-level LiveOps insufficient for publishers with multiple games?
   Title-level LiveOps operates in isolation. Without portfolio-wide visibility, an AI model might see a player inactive in one game and offer them a discount, unaware that the same player is actively spending full price in another title—leading to lost revenues and conflicting player messaging.
2. What are the key pillars of a modern, portfolio-level LiveOps AI architecture?
   A scalable portfolio architecture relies on four core elements: a canonical player entity, shared behavioral definitions, cross-agent intervention memory, and full data lineage.
3. How does Onix’s Semantic Twin solve multi-title integration issues?
   Powered by Onix’s Wingspan platform, the Semantic Twin connects account systems, device IDs, and behavioral data into a unified canonical knowledge graph. This gives AI agents a shared context, coordinates interventions, and automatically merges data from newly acquired games without requiring them to rebuild the infrastructure.