#172 - Yali Sassoon - Using LLMs to Support the Analytics Workflow
May 2, 2024
auto_awesome
Yali Sassoon, CTO and Co-founder of Snowplow, discusses the use of LLMs in analytics workflows, behavioral analytics, and the evolution of language models in data science. They explore how LLMs can enhance communication, decision-making, and problem-solving, while also addressing potential risks and ethical concerns. The conversation touches on industry trends, book writing challenges, and upcoming events in data analytics.
Behavioral data analysis reveals unexpected user behaviors, challenging traditional assumptions.
Complexity in data modeling poses a barrier to deriving actionable insights from event-level data.
Balancing AI assistance with human judgment is crucial to prevent errors and ethical dilemmas in analytics processes.
Deep dives
Initial Success of Snow Cloud in Uncovering Unusual Behavioral Data
In the early days of behavioral data analysis, the Snow Cloud technology implemented by Snowplow uncovered some bizarre and unexpected insights for its customers. For instance, a partner company was perplexed by a user having 50 logins from different organizations, which turned out to be a normal practice. Another case involved a bot changing IP addresses and user agents rapidly, fooling the system. These discoveries were not the expected outcome of the data analysis but provided valuable insights into unfamiliar behaviors.
Challenges Faced in Behavioral Data Modeling
A significant challenge in behavioral analytics is the complexity of data modeling. While event-level data can offer detailed insights into individual actions, aggregating this data to derive meaningful conclusions presents a considerable hurdle. Data engineering teams often spend extensive time transforming event data into structured units for analysis, complicating the process of utilizing AI tools for predictions or insights. Addressing these modeling challenges is a key focus for the future of behavioral analytics.
Future Developments in Behavioral Analytics
The future of behavioral analytics lies in streamlining the data modeling processes to unlock deeper insights and predictive capabilities. By simplifying the transformation of event-level data into actionable formats, organizations can enhance their customer understanding and AI integration. This advancement will enable more sophisticated customer analytics and informed decision-making based on behavior patterns and predictions.
Implications of Outsourcing Thinking to AI in Analytics
While leveraging AI models like LLMs can enhance analytics capabilities, concerns arise regarding the extent of reliance on these technologies. Handing over decision-making to AI systems without appropriate oversight can lead to errors, misinformation, or ethically questionable outcomes. Ensuring a balance between human judgment and AI assistance is crucial to prevent detrimental consequences or misguided outcomes in analytical processes.
Workshop Engagements and Book Writing Endeavors
Upcoming workshops in Malaga, Spain, and London, as well as ongoing book writing efforts, showcase a commitment to knowledge sharing and industry engagement. The evolving landscape of data engineering and analytics calls for continuous learning and exploration, exemplified by these interactive workshops and real-time book development. By fostering collaborative learning environments and innovative content creation, professionals in the field can stay abreast of evolving trends and best practices.
It seems like LLMs are taking the analytics world by storm. But how do you use them to support the analytics workflow? Yali Sassoon (CTO, Co-founder of Snowplow) joins us to chat about this and much more.
We'll also likely dive into behavioral analytics and more.