That "quick question" over Slack has been the bane of data analysts forever. Imagine those are now handled by ChatGPT, giving quick and reliable answers to business users. Stakeholders are happy, and data analysts can focus on deeper, more impactful work. Are we about to finally see this happening?
In this episode I talk with David Jayatillake (Co-Founder & CEO at Delphi Labs) about how large language models like GPT-4 are changing the way we work with data. What does this mean for data analysts or analytics engineers, and where do these new tools fit into the modern data stack?
Key takeaways:
- A lot of tools already offer a text-to-SQL approach. While this can be very useful to increase productivity for data analysts or analytics engineers, it's problematic as an interface for business users. When the semantic layer is effectively generated on the fly with every new query, results are unpredictable and can lead to a loss of trust.
- With a semantic layer, analytics engineers and data analysts can implement business logic and and expose data and metrics to business users in a safe and reliable way. (For example, dbt offers a semantic layer, but a lot of BI tools like Looker or Metabase have their own as well.)
- Delphi builds on top of these existing semantic layers, offering a natural language interface for business users. Instead of digging through a BI tool, stakeholders can simply ask their question in Slack. The answers will be limited to what is defined in the semantic layer, therefore avoiding the risk of wrong results.
- When data analysts are freed from answering "simple" requests, they can focus on deeper, more complex work to generate insights and recommendations to the business.
- While AI might eventually be able to take over most operational tasks, David believes that strategic decision making will still require human oversight in the future.
- Besides building data tools, David is also very active in the data community. He hosts a Mastodon server for data folks, and you can find him on dbt Slack and Locally Optimistic. You should also check out his Substack where he's written a lot about semantic layers recently.
Find David and Valentin on LinkedIn.
