Early enterprise adopters of generative AI have made it clear that a robust data strategy is the cornerstone of any successful AI initiative. To truly unlock AI’s potential as a value multiplier and catalyst for reimagined customer experiences, an easy-to-use and trusted data platform is indispensable.

Our recent report “The Radical ROI of Gen AI” proves gen AI is a profit engine, with more than nine in 10 surveyed early adopters saying that their gen AI investment is in the black. Survey respondents who quantified the ROI of their gen AI initiatives saw an average 41% ROI. Unlocking gen AI’s full potential hinges on a robust, unified data strategy. Eighty-eight percent of early adopters affirm that they need data strategies and tools spanning all generative AI use cases, meaning enterprises need a modern data platform that’s effortless to build and deploy, reliable by design and seamlessly connected across teams, tools and clouds.

Ninety-six percent of early adopters say they’re training, tuning or augmenting their commercial and open source LLMs, and 80% are fine-tuning models with their proprietary data. These essential steps introduce some real challenges. We’re talking about potential headaches around data quality, the amount of data your systems can handle, the risk of bias getting amplified and those privacy concerns — the ones about proprietary business information or customer personal data possibly leaking out — in the outputs. 

That’s why an organization’s approach to generative AI should be built on a strong data platform — to minimize those risks, to reduce surprise costs and to make it easier to adopt the right tools, scale and replicate AI successes, and make sure all of an organization’s data is securely and appropriately leveraged.

It’s easy to get lost in the sheer scale of it — 71% of organizations found that effective model augmentation requires multi-terabytes of data, or several million documents. But the breadth of data isn’t their only issue: Early adopters cite data quality (45%) and quantity (38%) most often among various issues. So your data hygiene and how you manage data becomes a mandatory focus of AI. 

To pile onto the challenge, the vast majority of any company’s data is unstructured — think PDFs, videos and images. So to capitalize on AI’s potential, you need a platform that supports structured and unstructured data without compromising accuracy, quality and governance. Only 11% of the early adopters say that more than half their unstructured data is ready to be used in LLM training and tuning. Even these early adopters, the ones who report great overall success, have hit some snags with their data platforms. At the data platform level, we found:

• 55% of organizations are hampered by time-consuming data management tasks such as labeling.

• 52% struggle with data quality — including issues of error, bias, irrelevance and timeliness.

• 51% say data preparation is too hard.

• 50% cite issues with data sensitivity.

• 42% say they lack the needed range or diversity of data.

All of these challenges are most effectively handled in a unified data platform. Bringing your AI technology to a data foundation that is easy, trusted and connected reduces the challenges that can delay a project or lead to unexpected costs.

Gen AI and cloud-based data platform go hand in hand

The majority of early adopters of gen AI are relying on cloud-based data platforms. And 81% of early adopters say they’re aggressively increasing their investments in cloud-based data warehousing solutions over the next 12 months. That’s no surprise — the cloud offers greater scalability, cost control and governance as well as access to the high-performance compute needed for gen AI initiatives.

Asked what they’re looking for from a cloud-based data warehouse, the early adopters largely pointed to three key features:

• Security: 46% rate this critical with a further 38% saying it is important.

• Advanced AI functionality: 39% say it’s critical, and a further 45% say it is important, to have LLM and ML capabilities under the hood.

• Integrated analytics capabilities: 39% rate this critical with an additional 45% saying it’s important.

Snowflake’s platform addresses all the critical needs of a successful AI strategy, which hinges on a secure, governed and modern data platform. To secure emerging AI applications like sentiment analysis and chatbots, data-driven businesses must implement controlled access and protection throughout data transformation (silver layer), which includes the entire security lifecycle, including:  

• automatic data classification of sensitive data sets and automatic object descriptions for data understanding 

• continuous monitoring and automatic policy enforcement via tag-based access policies and comprehensive lineage graph and access audits 

• ongoing data quality maintenance at scale 

In addition to these out-of-the-box governance capabilities, we have integrated a wide range of LLMs and general ML capabilities directly into the platform, eliminating complex data movement and the need for infrastructure management. Because it’s natively integrated into the platform, this approach preserves the organizations’ strict security boundaries and allows them to apply and enforce consistent governance and access policies across all data applications. 

Gen AI is not another technology isolated from the rest of your modern data stack. It’s a major and transformative initiative that will radiate throughout a data-driven organization. In the race to AI success, these organizations require an underlying data platform that allows them to rapidly scale the new generation of AI-powered applications with simplicity while building and deploying those in a secure and compliant manner at the same time. 

To understand how organizations worldwide are beating the challenges of implementing generative AI and scoring significant returns on their investments, read “The Radical ROI of Gen AI.”