The rapid pace of the gaming industry, where overnight hits emerge from the fringe, continually challenges large publishers and non-endemic investors. Traditional business intelligence processes, reliant on time-consuming surveys and manual analysis, struggle to detect meaningful signals before the market moves. According to analyst Joost van Dreunen, co-founder of SuperData Research and Aldora, this is a structural issue requiring a new approach to information. Leveraging small language models (SLMs) and predictive analytics presents a solution, offering publishers a secure, agile method to interrogate market data and forecast trends, thereby revolutionizing how investment decisions are made.
The Structural Lag in Traditional Business Intelligence
The core problem is one of speed versus process. Van Dreunen describes the conventional approach as a ceremonial, multi-week cycle involving proposal reviews, survey iterations, and deck presentations by large intelligence firms like Nielsen. By the time the finalized data reaches decision-makers, the world—and the market—has already shifted. This cumbersome model leaves major companies unable to detect emerging hits like Grow A Garden or Steal A Brainrot, as they lack the organizational agility to process real-time signals from various gaming channels. The reliance on outdated, large-scale survey data creates a critical informational lag in a fast-moving creative industry.
AI as a Distribution Tool: The SLM Advantage
While AI is often viewed as a content generator, its more transformative role is as a distribution tool for information. Van Dreunen emphasizes the distinct advantages of small language models over massive LLMs. SLMs are smaller, cheaper to maintain, and more energy-efficient, operating on edge compute even from a smartphone. Crucially, they are secure, allowing companies to combine public data with internal, proprietary datasets without sharing that sensitive information with a third-party like OpenAI. This enables expert-level access to intelligence in a safe, efficient manner tailored to specific needs.
Specialization Over Generalization
The nature of the games industry demands specialization. A publisher focused on mobile shooters doesn’t require a model trained on every conceivable topic; it needs one proficient in answering a focused subset of questions. SLMs excel at doing one thing really well, mirroring the industry’s own segmented expertise. This allows for precise, cost-effective analysis without the unnecessary ‘firepower’ of a 700-billion parameter model when a 7-billion parameter model suffices.
Practical Functionality and Analyst-Level Insight
On a base level, an SLM system can respond to prompts asking about biggest games, best performers, or next steps. Over time, through reinforcement learning trained on a company’s internal data, it can evolve to take on the role of an analyst. Van Dreunen describes this as anticipating the next question: hearing a query from a manager and identifying the underlying problem—for instance, shifting the focus from user acquisition to gun variety in a shooter game. This capability allows for multiple, rapid iterations on a problem, far surpassing the weeks-long wait of traditional analysis.
Managing AI Hallucination with Human Oversight
Acknowledging the risk of AI being ‘confidently wrong,’ Van Dreunen stresses that human operators remain a mandatory inefficiency in the mix. This human oversight prevents AI illnesses and flattens the layers between information provider and consumer. The onus, however, is on the user—much like a past analyst moving from Excel to SQL—to be trained to drill into the data and verify results. Ensuring the input data is structured appropriately is key to avoiding the equivalent of an AI hallucination.
The Expert Council Concept and Predictive Potential
The system’s power lies in democratizing access to expertise. While an analyst like van Dreunen cannot speak to every manager in a large organization, an SLM trained on aggregated industry knowledge can act as a ‘council of experts.’ This model provides one-to-one access to insights based on case studies, research, and historical interviews, helping teams think through problems with heavyweight counsel. As data accumulates, the predictive capability grows.
Building the Predictive Data Lake
Aldora has spent two years aggregating structured data (earnings reports, executive transcripts) and unstructured data (academic articles, case studies, interviews). This data lake, built on publicly available information supplemented by analytical models, allows for deep insight into industry decision-makers. By tracking a CEO’s career history, their past successes, and stated philosophies, the system can better predict likely future decisions. In an industry directed by only a few thousand key individuals, this consistency offers a significant forecasting advantage.
The Product Rollout: Secure, Localized Implementation
The practical rollout, expected in Q3 of 2024, will involve providing companies access to a localized SLM that feeds off Aldora’s data lake. The output will initially be expressed in conventional metrics via leaderboards showing the biggest IPs across gaming channels and their effective reach. Users can then engage with a prompt-based interface to ask specific questions about their own projects.
A Secure Data Room for Proprietary Experimentation
A key feature is security and localization. Companies can subscribe to the software and run a local version, creating a secure data room. They can feed the model with their proprietary ideas and data, asking ‘what do you think?’ without any information leaving their premises to a cloud server. This allows for safe experimentation on either an incidental basis—for a major launch decision—or a routine weekly check on specific categories.
Catering to Endemic and Non-Endemic Audiences
The application serves two core audiences. For endemic game publishers and studios, it offers accelerated, expert-driven business intelligence. For the growing wave of non-endemic companies—brand managers, private equity firms, venture capitalists—entering gaming as a major marketing channel, it provides essential education and insight. As van Dreunen notes, many financial investors have limited gaming knowledge beyond major titles like FIFA or World of Warcraft, leaving exotic indies and emerging genres outside their scope. The SLM system bridges this knowledge gap for those investing billions with minimal industry familiarity.
Ultimately, the integration of SLM-based predictive analytics represents a necessary evolution for publisher investment strategy. It replaces slow, generalized intelligence with fast, specialized insight, enabling decision-makers to ask iterative questions and receive expert-caliber analysis securely. By combining a rich, aggregated data lake with localized models and mandatory human oversight, this approach offers a path to not only understand the current market but to meaningfully predict its trajectory, empowering both gaming veterans and new entrants to navigate the industry’s volatile landscape with greater confidence and foresight.
