Self serve: the (elusive) holy grail of data teams from Coalesce 2023

“There's lots of complications with defining metrics…you can't tell the CEO a wrong number for revenue and expect her to come back and look at this tool again.”

Paul Blankley, CTO and co-founder of Zenlytic, discusses self-serve analytics, its challenges, and how it can be achieved. He also highlights the importance of the semantic layer and language learning models (LLMs) in achieving self-serve analytics.

Self-service is the goal of all BI tools but has not been fully realized due to the complexity and evolving nature of data

Paul explains the challenges of self-service in data analytics. He states, "Every BI tool that's ever existed has claimed to be self-serving, and they're not really lying. It's just…the goalposts change as the tools change for self-serve." According to Paul, the definition of self-serve has evolved from simple static dashboards to being able to ask questions and get direct answers without needing to know where to start.

Paul identifies three key challenges in realizing self-serve. "Data discovery is hard,” he states, explaining that the abundance of dashboards and data sources often leads to users reaching out to the data team for assistance. He also highlights the difficulty in defining metrics, due to the complexities and nuances in the data. Finally, he adds that the complexity of data interfaces can be overwhelming for business users who just need specific data.

The combination of language model technology and the semantic layer could be the key to achieving self-service in data analytics

Paul presents a potential solution for achieving self-serve, involving the marriage of language model technology (LLMs) and the semantic layer. "With LLMs and the semantic layer together, we kind of get the best of both worlds,” he says. He explains that combining the comprehension capabilities of LLMs with the correctness and governance provided by the semantic layer could provide business users with a simple interface to ask questions and get the right answers–without needing undue technical expertise or data knowledge.

The ideal self-service data analytics solution should work the way humans already do

Paul emphasizes the importance of creating data analytics solutions that mirror natural human interactions. "Self-serve needs to work the way that humans already work, and humans already work in conversations,” he says. According to Paul, this conversational approach would allow for a more iterative process in answering data queries, without the need for constant involvement from the data team.

The future of data analytics is moving towards a fully self-service model

Paul expresses optimism about the future of data analytics, particularly with the rapid advancements in technology. "This technology is moving so fast. I really think in a year or two, using data is going to feel like having a little army…" He believes that with the right governance and guardrails, the dream of self-serve analytics can become a reality.

Paul’s key insights

• Self-serve analytics is a spectrum that has evolved from static dashboards to more flexible tools, but it still has not reached its full potential
• Challenges in realizing self-serve analytics include difficulty in data discovery, complexity of metrics, and non-intuitive data interfaces
• The combination of LLMs and the semantic layer can pave the way for self-serve analytics. LLMs provide comprehension of user intent, while the semantic layer ensures the correctness of data
• The semantic layer is crucial for defining and governing metrics in self-serve analytics
• Conversational interfaces, which mimic human interaction, are the most effective for self-serve analytics