High-quality data key to unlocking value from AI, GenAI: Snowflake AI head

“Businesses often struggle with scattered data across multiple systems, leading many to adopt data platforms like ours to consolidate, govern, and analyse data effectively,” he told Mint in a video interview from his office in San Mateo, California.

Cloud data platforms help organisations integrate data from various departments and sources, enabling them to manage, analyse and run AI models efficiently, thus enhancing governance, security, and productivity. Snowflake, according to Gultekin, offers “seamless data integration without needing complex transfers,” allowing companies to process and share massive datasets.

“As AI becomes critical, organisations prefer running AI models close to their data. Snowflake supports this by offering a secure environment with robust governance, ensuring that sensitive data remains protected,” he said, adding, “With massive datasets—often in petabytes—customers prefer to run computations directly where the data resides, avoiding the cost and complexity of moving it elsewhere.”

The technology needs a “very ready” data foundation to feed on, something the vast majority of businesses today (78%) do not possess, according to a joint report by the Massachusetts Institute of Technology Technology Review Insights and Snowflake, underscoring the need for high-quality data to power GenAI projects.

Many participants said they were more interested in leveraging GenAI’s ability to improve efficiency and productivity (72%), boost market competitiveness (55%), and drive better products and services (47%), rather than just increase revenue (30%) or reduce costs (24%).

No silos

Snowflake, according to Gultekin, eliminates data silos and ensures that the data is prepared for AI applications. He added that Snowflake’s cloud-agnostic platform works across Google Cloud, Microsoft’s Azure, and Amazon Web Services (AWS), enabling companies to operate efficiently in multi-cloud environments. Organisations use Snowflake-powered AI for multiple purposes. Some enhance business intelligence by enabling real-time query responses, while others build chatbots for efficient knowledge management, according to Gultekin.

He cited a few cases in point. TS Imagine, a financial services firm, automated the classification of vendor emails with AI, reducing processing time by 95% from 4,000 hours. Siemens Electronics implemented a chatbot to provide its research team with immediate access to insights from 700,000 pages of documents. Pharmaceutical company Bayer used Snowflake to transform how its teams interact with business intelligence.

… data platforms like ours to consolidate, govern, and analyse data effectively

“Instead of waiting days for analysts to respond to dashboard queries, their AI-powered chatbot provides real-time answers, streamlining decision-making,” Gulketin explained. “Trust is fundamental—customers rely on Snowflake to handle sensitive data securely within its boundaries. By running large language models (LLMs) directly within the platform, Snowflake ensures robust governance and makes AI adoption easy and efficient.”

Other major vendors in the cloud data platform space include Databricks, Oracle, AWS, Microsoft Azure and Google Cloud. With rising demand for data-driven insights, the global decision intelligence industry is forecast to grow to $64 billion by 2034, up from $12.1 billion this year, according to Future Market Insights, Inc.

Using AI agentic systems

The future of AI, according to Gultekin, points toward autonomous agentic systems, which can perform tasks independently with minimal human involvement, unlocking new productivity levels. Snowflake also integrates agentic AI systems that refine queries to ensure accuracy and align answers with user intent. They operate independently, choosing tools and data sources as needed, such as retrieving stock prices or news documents, showcasing early-stage autonomy.

“Our agentic system goes beyond simple translation by reasoning through multiple steps. It generates SQL queries (Structured Query Language queries are instructions that databases can understand), assesses whether they accurately match the user’s intent, and refines the query if necessary. Multiple LLMs work together to perform this reasoning, marking an early stage in the development of agentic systems,” Gultekin explained.

If a customer asks for the latest news about a company, for instance, the system queries recent news documents. On the other hand, if the question is about stock performance, the model accesses structured financial data to provide the current stock price and trends. The ability to reason about which tool to call upon demonstrates the system’s agentic capabilities.

“These systems are also designed to be extensible, meaning they can incorporate additional tools into their workflows. For example, the LLM might call on a specific tool when needed to enhance its reasoning. Importantly, this process is entirely self-supervised, with no human intervention. The system operates autonomously, reflecting the growing sophistication and promise of agentic models,” he added.

Addressing hallucinations

Gultekin, though, acknowledged that addressing AI challenges requires reducing model hallucinations, which occur when GenAI models throw up inaccurate results.

Snowflake’s approach, he explained, involves building AI systems that only respond when verified information is available, ensuring governance and access controls align with user permissions. This ensures, for example, that HR chatbots provide responses based on access rights, preventing unauthorised disclosures.

The less we focus on purely mechanical tasks, and the more we nurture harmony and understanding among each other, the better society will be.

Snowflake balances the use of general-purpose models, or LLMs, and task-specific models, or small language models (SLMs). According to Gultekin, while general-purpose models offer flexibility, task-specific models are favoured for efficiency in areas such as sentiment analysis and classification.

“Cost and speed influence the choice between these models, with companies prioritising efficiency and accuracy based on specific needs,” he said.

Preparing for the future

Gultekin explained that the shift from traditional machine learning (ML) to GenAI is redefining how businesses analyse both structured and unstructured data. Generative AI enables large-scale analysis of documents, images and call logs, empowering business users to access insights without analyst support.

Companies continue to build on traditional AI foundations—like fraud detection—while expanding into new unstructured data applications, democratising data access and improving productivity.

“Governance remains a crucial aspect of AI adoption, with organisations establishing AI oversight boards and rigorously testing models before deploying them in production,” he said.

He added that as businesses explore new models, synthetic data too becomes essential, enabling continuous model improvement. An example of synthetic data use is Google’s AlphaGo, which achieved superhuman abilities by playing against itself and learning from it. But this data, too, he acknowledged, needs to be monitored for biases.

When asked how freshers should prepare for a future where AI and GenAI are automating hundreds of tasks, Gultekin said, “This is one question I have reflected on personally, especially with my middle-school-aged children. I believe life revolves around relationships. Human connections are what drive the world forward. The less we focus on purely mechanical tasks, and the more we nurture harmony and understanding among each other, the better society will be. While this might sound like a soft perspective, I think it’s essential.”

As an example, he pointed out that we typically have been teaching kids to communicate with machines using programming languages.

“However, that’s now shifting. Today, we can interact with computers using natural human language, which is fascinating. This change underscores the importance of creativity—being able to communicate meaningfully with technology. What you say and how you think become more valuable than the mechanics of coding alone,” he concluded.