40% of Companies Ready to Abandon AI: How to Avoid Failure

The enthusiasm surrounding artificial intelligence (AI) agents is palpable, but their effectiveness in terms of return on investment remains to be demonstrated. According to a recent analysis by Gartner, by 2027, 40% of companies may abandon their autonomous AI agents. This prediction is driven by governance gaps that are often only discovered after incidents occur when these agents are already in production.

At the Snowflake Summit in San Francisco, three digital leaders shared their experiences implementing AI agents in their respective organizations. They identified three essential strategies for effectively leveraging these technologies: using structured frameworks, engaging expert analysts, and monetizing data.

Focus on Frameworks

Matt Luizzi, Vice President of Analytics at Whoop, explained how his company continuously collects biometric data to provide insights into health and wellness. Whoop uses Snowflake's analytics services to support its internal operations. Luizzi emphasized that AI agents, particularly CoCo, Snowflake's coding agent, play an increasingly crucial role in this process.

"We have been using CoCo for several months now. We started with the analytics team, which is able to quickly assess the relevance of responses to queries and expand this process," he said. "We are now at a stage where we have established more formalized evaluation frameworks and are beginning to deploy agents at scale."

Luizzi clarified that the company has software engineers conducting A/B tests, using CoCo to analyze results, propose new features, test them, and iterate. "This approach accelerates the delivery of business value by automating the experimentation framework while bringing value to customers," he added.

Utilize Expert Analysts

Madeleine Want, Vice President of Data at Fanatics, oversees data engineering, data science, and machine learning in the organization’s betting and gaming division, supported by the Snowflake platform.

"When we started experimenting, we weren't sure what would work, but we found that the better the quality of the underlying data and the more effective the governance, the more the LLM was able to derive meaning and respond to questions effectively," she stated.

Although this focus on data and governance may seem obvious today, Want noted that it was not the case 18 months ago. "We had a lot of experience as an organization building custom machine learning models, so it was hard to believe that the idea of importing a third-party model and overlaying it directly on the data could work for analysis."

Want added that her organization has seen more success over time. The investment required in the context layer is decreasing, as is the degree of oversight an agent needs before it can start answering questions autonomously.

Monetize Your Data

Sriram Sitaraman, Chief Information Officer at Synopsys, stated that his organization has been a long-time customer of Snowflake, using the data platform and its agent services, such as CoCo, to fuel its decision-making processes.

About 18 months ago, Sitaraman indicated that the company recognized the potential of AI agents to handle tasks typically performed by junior employees, such as executing quick queries, creating charts, and deriving insights. "We leveraged this capability and said, 'Okay, if we create a knowledge agent, we can start deploying it in multiple dimensions.'"

Sitaraman specified that the team evaluated the potential of AI across three dimensions: the quality of results, the time taken to achieve results, and the cost of results. They found that AI has a positive impact in all three areas, which he described as a significant breakthrough.

"Start with the data - monetize your data using AI," Sitaraman said, reflecting on his company's agent evolution. "No matter the volume you invest in the initiative, because AI is truly a linear scale. The more data AI has, the better decisions it makes."

However, Sitaraman also issued a warning. "One thing we realized is that there isn't much difference today between automation and autonomy, so you need to be cautious," he said.

Sitaraman encouraged professionals to identify the right use cases, build the right frameworks, and never underestimate what an agent can do. "You can deploy an agent and say, 'This is a business operations agent.' Often, nothing prevents it from also becoming a business analyst agent or another type of agent," he concluded.