Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
Deploying generative AI and LLMs has become a non-negotiable priority for most enterprises. Competitive pressure to “get in the game” is stress-inducingly high, whether that means shipping new conversational AI solutions for customer service, harnessing coding copilots, generating ad creative, improving cash flow, predicting trends and new profit potentials, or any number of the myriad and rapidly maturing, use cases now available.
But as we’ve seen before, when enterprises are racing to sea-change technology– see cloud transformation, Kubernetes and so on – there is usually no shortage of stumbling blocks to a smooth transition at scale. Even though generative AI and LLMs demand considerable investments just as table stakes, enterprises nevertheless view those investments as essential to their futures. According to hundreds of respondents in IDC’s March 2024 Future Enterprise Resiliency & Spending Survey, they plan to maintain them regardless of economic conditions.
While committed to long-term AI initiatives, enterprises are also becoming increasingly aware of specific hurdles to making their goals a reality. That same IDC survey underscored the challenge of operating AI models and applications at scale, particularly in cloud environments. As they navigate this complex landscape, enterprises are actively seeking best practices and necessary strategy pivots to successfully implement generative AI workloads. This growing recognition of operational complexities stresses the need for a more careful approach to AI adoption, one that can balance the urgency to innovate with the realities of large-scale deployment.
Let’s examine four key factors that enterprises must keep top of mind as they implement – and scale – generative AI workloads.
Organizations struggling with generative AI at scale often find that a lack of developers with the correct skills is a major obstacle standing between them and success. Because this is still an emerging technology, developers, data scientists and data engineers with the experience and talent to shepherd AI projects to completion and earn positive results are still in short supply. Enterprises commonly face difficulty in recruiting and retaining these experts.
At the same time, these enterprises often saddle their AI/ML and big data teams with cumbersome cloud operational duties necessary to support their AI applications and models, stealing away focus from the work that improves those end products. In a virtuous cycle, it’s enterprises that streamline cloud operations and free AI and data experts to do what they do best that have the easiest time recruiting and retaining top talent.
AI isn’t cheap and, perhaps more than any other variable, those runaway bills will prevent many enterprises from achieving sustainability and a healthy return on generative AI projects. As organizations scale their AI/ML and related data-heavy applications, the cloud costs associated with their experimental, model-building, testing and production environments will scale hand-in-hand unless enterprises introduce more efficient practices.
For example, successful teams will utilize automated provisioning, instance size and price optimization and real-time cost insights to keep AI expenses under control. Enterprises will want to be sure to align FinOps practitioners with those working on AI projects; accurate cost visibility and analytics need to be well-understood to ensure resource efficiency.
Infrastructure incapable of supporting the reliability and performance of AI workload demand has hampered many enterprises’ AI ambitions.
Breaking down common infrastructure issues, enterprises can struggle to determine the correct infrastructure for their needs—whether it’s on-prem, hybrid cloud, public cloud, or a combination. Large AI models also call for major computing power, including costly GPU resources, which can easily lead to huge cost and performance inefficiencies if used or configured incorrectly.
Cloud technologies such as Kubernetes similarly require careful management, configuration and monitoring by experts, to overcome their complexity and scale without ballooning expenses. Inexperienced DevOps teams will often rely on manual processes for supporting AI/ML and big data projects, embracing practices that cannot scale while leaving major optimization opportunities on the table. Automated cloud infrastructure optimization techniques and tooling can address these cost and performance issues, especially when paired with AI/ML-specific approaches such as MLOps that automate away the complexity of these deployments.
It’s not uncommon for enterprise leaders to offer generative AI teams a blank check to achieve their goals. They’ll give the go-ahead to blow through budget limits and ignore established business processes, in the hope they’ll expedite bringing working AI solutions to market. In practice, that approach creates huge headaches for every other stakeholder and function AI touches upon, from DevOps to FinOps, product managers, finance and more. Building internal coordination is accomplished via thoughtful processes, not a lack of one.
Studying the key factors limiting enterprises’ generative AI efforts, the need for more effective cloud spending and utilization emerges as a through line. By using established tools and practices for introducing cloud automation, efficient management and cost optimization and adding equivalent controls relevant to AI/ML and big data, enterprises can reduce their growing pains when successfully deploying generative AI workloads at scale.
You May Also Like