Effective Strategies for Scaling Generative AI in 2025

This article is part of VentureBeat’s special issue, “AI at Scale: From Vision to Viability.”

As technology advances rapidly, scaling generative AI tools offers both hurdles and possibilities for businesses. By 2025, organizations face the crucial task of balancing their ambitions while remaining practical. The challenge in deploying large language models (LLMs) now goes beyond just adopting advanced tools—it focuses on meaningfully integrating AI to elevate operations, empower teams, and manage costs. Achieving effective scaling requires a significant cultural shift to align AI with overall business goals.

The Shifting Landscape of Scaling Generative AI in 2025

As generative AI transitions from early exploration to large-scale implementation, businesses encounter a pivotal moment. The enthusiasm of initial adoption gives way to real-world challenges such as maintaining efficiency, controlling costs, and staying competitive. Effectively scaling generative AI involves addressing critical questions:

• How can organizations ensure that generative tools are impactful across various departments?
• What infrastructure will support sustainable AI growth without creating roadblocks?
• How can teams adapt to AI-driven workflows?

Success in this environment hinges on three key principles: identifying valuable use cases, ensuring technological adaptability, and nurturing a workforce that is ready to evolve. Companies achieving their AI ambitions do so by crafting strategies that effectively align technology with business needs, while constantly reassessing costs, performance, and necessary cultural adjustments.

Firms such as Wayfair and Expedia exemplify how to merge agility and precision in their generative AI strategies, successfully transforming operations through innovative adoption of LLMs.

Finding the Right Balance Between Customization and Flexibility

The choice between building or buying generative AI solutions often appears straightforward, but companies like Wayfair and Expedia advocate for a more nuanced approach. According to Fiona Tan, Wayfair’s CTO, achieving the right balance between flexibility and specificity is vital. While for general tasks, Wayfair often utilizes Google’s Vertex AI, they design customized solutions for unique requirements. Tan showcases how smaller, cost-effective models can outperform larger, more expensive ones in certain contexts, such as tagging product attributes.

Expedia mirrors this strategy, employing a multi-vendor LLM proxy layer that allows for seamless incorporation of different models. Rajesh Naidu, Expedia’s Senior Vice President, emphasizes their opportunistic method, seeking the best available tools for their needs, while being ready to adapt or create custom solutions when necessary. This kind of flexibility enables the organization to swiftly respond to evolving business demands.

This hybrid strategy is reminiscent of the evolution of enterprise resource planning (ERP) systems in the 1990s. Companies back then navigated a similar dilemma: settle for rigid, out-of-the-box solutions or customize systems extensively. Bold enterprises recognized the benefit of combining external tools with tailored innovations to overcome specific operational challenges, just as organizations are doing today.

Boosting Operational Efficiency for Key Functions

Wayfair and Expedia illustrate how LLMs reach their full potential with focused applications that produce measurable results. Wayfair effectively utilizes generative AI to enhance its product catalog, boosting metadata accuracy, which streamlines operations and improves search capabilities and customer recommendations. Tan notes another significant application: generative AI analyzes outdated database structures, helping to manage technical debts and uncover efficiencies when legacy systems are involved.

For Expedia, the deployment of generative AI spans various areas, including customer service and development operations. Naidu mentions a custom AI tool developed for call summarization, ensuring that “90% of travelers can connect with an agent within 30 seconds,” significantly enhancing customer satisfaction. Furthermore, GitHub Copilot accelerates code generation and debugging throughout the company. These operational improvements highlight the importance of aligning generative AI functions with well-defined, high-value business objectives.

Hardware Considerations for Effective AI Scaling

When discussing scaling generative AI, the significance of hardware is often underestimated, yet it is crucial for long-term sustainability. Both Wayfair and Expedia depend heavily on cloud infrastructure to manage their generative AI workloads. Wayfair continually assesses various cloud solutions, like Google, while also recognizing the need for localized resources to deal with real-time applications effectively.

Expedia emphasizes flexibility; mainly running on AWS, the company implements a proxy layer that optimally directs tasks to the most efficient computing environment. This approach guarantees both performance and cost-effectiveness, preventing expenses associated with inference from escalating unnecessarily. Naidu asserts that this kind of adaptability becomes essential as enterprise AI applications grow in complexity and require more processing power.

This focus on infrastructure reflects wider trends in enterprise computing, similar to the transition from monolithic data centers to microservices. As Wayfair and Expedia enhance their LLM capabilities, they exemplify the critical need to balance cloud scalability with emerging technologies, like edge computing and tailored hardware solutions.

Cultural Shifts, Training, and Governance

Implementing LLMs isn’t just a technological interaction; it’s a shift in culture too. Both Wayfair and Expedia stress the importance of creating a favorable environment to adopt and incorporate generative AI tools. At Wayfair, thorough training programs enable employees from all areas to adapt to evolving workflows—especially in customer service, where AI-generated responses require human supervision to maintain the company’s voice.

Expedia enhances governance with its Responsible AI Council, which oversees significant generative AI initiatives. This council ensures that projects align with ethical standards and business goals, thereby nurturing trust within the organization. Naidu points out the need for updating metrics to evaluate the effectiveness of generative AI. Standard key performance indicators often miss the mark, prompting Expedia to adopt precision and recall metrics more representative of their goals.

Valuable Insights for Successful AI Scaling

The experiences of Wayfair and Expedia provide insightful lessons for organizations aiming to effectively scale generative AI. Their success revolves around clearly identifying business use cases, fostering technological flexibility, and cultivating a culture of adaptability. Their hybrid strategies present a balanced model of innovation and efficiency, ensuring that investments in AI produce real, measurable results.

The challenge of scaling AI in 2025 is heightened by the rapid technological and cultural shifts. The hybrid methods, adaptable infrastructures, and strong data-driven cultures exemplified in successful AI practices today will lay the groundwork for future innovations. Businesses that establish these foundations now will not only enhance their AI abilities but will also cultivate resilience, adaptability, and lasting competitive advantages.

As we progress, ongoing challenges surrounding inference costs, real-time capabilities, and evolving infrastructure needs will continue to shape the generative AI landscape for enterprises.