Elevating the Importance of Proprietary Data

LLMs display impressive general knowledge and, when paired with information derived from web search, can assist with a variety of tasks, from drafting emails to summarizing documents. Additional value emerges when users incorporate proprietary data into a model’s reasoning and responses. This additional context can improve output accuracy, reduce hallucinations and enable agentic workflows for both employee productivity tools and customer-facing applications.

While Retrieval Augmented Generation (RAG) and the increasing length of model context windows allow internal data to be incorporated into LLM prompts, the value of these techniques is proportional to the strength of the underlying data. High-quality data that’s accurate, complete and deduplicated improves the model’s ability to deliver relevant and trustworthy results.

That’s why we believe investments in data—from ERP and CRM software to cloud data warehouses—are foundational to generative AI strategy. In addition to supporting traditional business workflows and analytics, they serve as stores of proprietary intelligence that complement the strengths of AI models. As the capabilities of LLMs evolve, so too must the data foundations that support them. Continued investment in high-quality, well-governed and accessible data will likely be critical for organizations looking to unlock long-term value. In short, we believe the data assets you've built—and the ones you build next—are strategic levers in your company’s ability to compete in an AI-driven landscape.

Accelerating Time to Insight

Many executives can relate to the experience of asking a seemingly simple question, only to find that it may take their team days, weeks or more to get the answer. Data analysis often involves a series of complex, manual steps: locating disparate data sources, stitching them together, validating quality and then, finally, performing the analysis. Technology has improved the process considerably in recent years, from cloud data warehouses and data pipelines to drag-and-drop business intelligence software. Generative AI can help compress the time to insight further. Rather than replacing investments in data infrastructure, we believe AI tools will allow data teams to move faster and operate at greater scale. Two high-impact examples stand out:

1. Code generation: A wave of AI coding assistants has hit the market, and while most are not specific to data engineering or data analysis, they can accelerate the process of writing the code needed to move, clean and analyze data.
2. Unlocking unstructured data: Businesses often have valuable information stored in PDFs, presentations, images and other file formats. Previously, significant manual effort was required to parse these file formats, and in our experience, many teams avoid this work due to time and cost. Today, LLMs can be incorporated into data pipelines to extract discrete data points from these file formats and other, unstructured sources and store them for further processing, querying and analysis. This expands the scope of usable data and reduces the time and cost to answer important questions.

With faster access to a broader range of data and data-driven insights, executives are empowered to ask more questions and expect more timely answers. We believe companies that proactively modernize their data strategies will be better positioned to meet these rising expectations and turn curiosity into competitive advantage.

Shaping the Future of Self-Service

Self-service has long been a goal of data strategies, but for many organizations, it can remain frustratingly out of reach. Despite recent investments in data warehouses and visualization tools, leaders at many growth stage companies still struggle to access and interpret the data they need to make effective decisions.

It’s no wonder. The reality is that data are complex and many people in decision-making roles don’t have the technical background required to create effective queries and make statistically sound conclusions. Meanwhile, data scientists are in high demand and short supply. While good questions and making decisions based on the information available will remain a human responsibility for some time, AI models can assist in several areas of the analytical process. Examples include:

• Translating questions into queries: Models can turn natural-language prompts into SQL or Python code that queries a data source, performs calculations and summarizes results in a narrative. Some of these models are nascent, but we expect the abilities to improve meaningfully over time.
• Providing in-context explanations: Chat interfaces powered by LLMs can teach users key concepts such as how to identify drivers of customer churn, recognize seasonality patterns in revenue or distinguish between metrics like mean and median—in context and just-in-time.

It’s important to note that LLMs are not a substitute for data fluency in your organization. Their impact is greatest when paired with a workforce that understands how to ask smart questions, interpret results and apply insights effectively. As generative AI lowers the barrier to engaging with data, we believe companies should invest not only in AI tools, but also in cultivating a culture of data literacy that enables broader, more strategic use of insights across the organization.

That said, while potential improvements in data self-service are significant, we recommend that leaders tread carefully and keep expectations realistic. Models cannot reliably distinguish between fact and fictional data, they often produce inaccurate calculations and—as has been well documented—LLMs tend to “hallucinate”, confidently making up answers. Responsible use supported by human oversight remains essential.