AI's unstoppable march continues to gather pace. Analyst Gartner recently predicts that half of all business decisions will be fully automated or at least partially enhanced by AI agents within the next two years.
Also: 4 ways to turn AI into your business advantage
Some organizations experiment more than others. Four business leaders investigated the shared AI lessons they learned at the recent media roundtable event at Snowflake Summit 2025 in San Francisco. This is what they had to say.
1. What is my cloud strategy?
Wayne Filin-Matthews, Chief Enterprise Architect at Astrazeneca, explained how his organization has pioneered the implementation of AI in several areas.
Pharma Giant has developed an AI-enabled research assistant that will increase productivity for scientific researchers by focusing on reproducibility of scientific methods and the development of new drugs.
Astrazeneca partners with major academic institutions such as Stanford University to run agent AI experiments.
“We're thinking about how we can have a team of agents that can support the traditional scientists who do their research,” Filin Matteuse said.
Also: 4 ways your organization can adapt and thrive in the age of AI
The company is also looking for ways to apply AI in commercial areas. Astrazeneca operates in 126 markets, serving different locations with content is a complex challenge. That's where AI comes into play.
“We used technology from an AI perspective to automate the creation of marketing materials and the creation of information on drug development,” he said.
These experiments highlight the benefits of AI, but also show the importance of the foundations of solid-state data.
Additionally, AI integration starts with the foundation of robust data. There are three strategies for executives to hire
Filin-Matthews said that AI problems can only be solved if a company has built a strong underlying cloud infrastructure.
“When we go on this journey, there are so many use cases where profits are apparent,” he said.
“We are definitely in an age of AI-enabled decision-making. But the key to me is that we can't forget other fundamental elements. Without cloud-first, we can't become AI-First.”
2. Have you addressed data governance concerns?
Amit Patel, Chief Data Officer of Truist's Wholesale Bank, said he learned two important lessons from deploying AI use cases.
Number one was the importance of the underlying data foundation.
“As a bank, we need to prove. 'Where did the data come from? Is it correct? Is it governed? Is it lineage? Is there any metadata? Is there data quality checks?” I have to prove those points to an external regulator,” he said.
“Can't you release a large-scale language model (LLM) into the wild? And you can't just point out the sources I have internally. It must be a governed source. It must be an authorized provisioning point.”
Also: This free Google tool turns AI into a research assistant
Patel said it will help to unravel common problem points in CDOS by focusing on regulated sources.
“Through the process, I discovered that I didn't have as many reliable sources as I would like to point out,” he said. “I have to enable that foundation first. Then I can build on top.”
Patel said the second thing he learned is to assume that anyone using AI at home can easily deploy LLM in an enterprise environment.
“It's not that simple,” he said. “We need to define guardrails for what the model can see. We need to define metadata to derive interpretations of the model. The process takes time.”
Also: Is your business ready? Five ways to keep up
Patel said his team addressed staff misconceptions about how long it takes to use AI through the expected exercise.
“When we started enabling use cases, people began to realize that it wasn't as easy as the point-and-click process,” he said.
“The implementation of technology is faster than before, but it's still challenging and requires time and thought about how to put governance and structure around AI before enabling it for work.”
3.What is the quality of my output?
Anahita Tafvizi, Chief Data and Analytics Director at Snowflake, said her team will help customers develop AI-enabled products for use.
But Tafvizi not only allows her company to sell these products, but organizations can also experiment with these technologies.
“The interesting thing about being a CDO for a data company is that it gives you the privilege of being the first customer of many of our products,” she said.
Tafvizi turned his attention to Snowflake Intelligence, a technology launched at the summit that allows business users to create data agents.
Also: Top 20 AI Tools of 2025 – And the first thing to remember when using them
Her team has partnered closely with the product team to develop an AI-enabled assistant for internal sales organizations.
She recognized that implementing new AI tools poses challenges, especially when it comes to balancing the speed of innovation and governance requirements.
One important consideration is quality. When her team pushed the tool to the sales team, they pondered important questions such as “Is 95% quality enough?”
Tafvizi advised other business leaders to think carefully about these challenges, as staff must trust the results of their AI experiments.
“The focus on quality was important to us,” she said. “Adequate governance structures, access control, lineage, metadata, and semantic models are also important. We've always thought of these as part of the tensions of innovation and speed.”
4. Have you considered unexpected benefits?
Thomas Bodenski, chief data and analytics director at TS Imagine, a financial technology specialist, said his company has been using AI to reduce employee workloads since October 2023.
However, although the focus of AI is often on the automation of manual processes, his experience suggests that business leaders should recognize that technology also produces other benefits.
“Using AI is not just about reducing effort,” he said. “You can make things faster, better and incredible coverage improvements.”
He explained how TS can purchase data from specialist vendors who will send emails about upcoming product changes.
Also: 10 strategies Openai creates powerful AI agents using – you should also use
The company receives 100,000 of these emails per year. Each email needs to be read and its meaning is understood. Traditionally, the work-intensive process has on average consumed 2.5 full-time equivalents per year.
“It's stressful because you can't make mistakes,” he said. “If you miss an information in an email, the system goes down. It's potentially catastrophic as thousands of traders can't trade and thousands of risk managers can't assess their exposure.”
To avoid this scenario, Bodenski said the company has completed the work of buying this time using Snowflake's AI model.
“Now we don't miss out on the outcome,” he said. “These 2.5 full-time equivalents can do knowledge work rather than manual data curation or entry.”
Bodenski said AI can manage what was previously a weakness. Ensure that customer requests are handled on Saturday.
“No one was working back then. Now there's AI and she responds to customer inquiries and assigns tickets to the right people,” he said.
Want more stories about AI? Sign up for Innovationour weekly newsletter.
