Data needs to work the way humans already work.
Zenlytic was built after two data scientists felt there was a better way to answer the many daily ad hoc questions they received. Both legacy and new analytics vendors promise self-service, but the reality is users still need to learn complicated tooling, python, or SQL to answer questions beyond a curated dashboard. People just want to ask questions the way they ask their data analysts: in a conversation.
The cognitive layer enables self-serve business intelligence
Zenlytic is the world’s first self-serve business intelligence platform. It combines Zoë, a generative AI data analyst, with a cognitive layer for ad hoc questions and report building so people can make decisions faster.
The cognitive layer is the architecture that allows Zenlytic to work with large language models (LLMs) for self-serve. Data modeling has evolved so data teams could provide curated ways for people to view and understand data. From OLAP cubes, to ungoverned BI structures, to semantic models, data teams have been structuring data to serve those needing insights into the business. The issue is data teams spend a lot of time answering ad hoc questions.
Why the traditional semantic layer doesn't work with LLMs
Looker led the way in semantic layers with the creation of LookML. Modeling languages like LookML are fantastic at modeling data into projects or explores so you can define metrics and reuse them across data in each project. Business users can explore from a dashboard and slice and dice data.
The issue arises when anyone has a new question about data. With a semantic model structure, a user has to know which project or explore in the model to start with. Search based analytic tools are not any better. You can go deep into the data, but the data team needs to curate what data is in each project. Semantic models are a significant advancement from in-memory data structures like Tableau, but data teams are still a bottleneck in guiding and curating data for the business teams.
“No one bothers to log into Looker and find the right explore. They just email or message the data team and wait.”
- VP Data Engineering at Fintech Company
The question is do you want to chat with a dashboard or chat with all of your data? A cognitive layer combines a centralized semantic layer with a large language model. The semantic layer needs to be flat to be able to work across an entire data warehouse. As most semantic layers are deep and multi-layered into explores, a centralized and flat semantic layer is a critical feature of Zenlytic a key feature of the cognitive layer.
The Zenlytic cognitive layer
A cognitive layer combines all of the traditional semantic layer functionality that makes BI great - the same metrics definition syntax in YAML, the same run-time SQL generation.
But then we've added more functionality designed around how people use data in the LLM era. For instance:
Understands natural language instructions: Semantic layers always have some tribal knowledge that's emailed between the data analysts because it can't be encoded in a YAML. The Cognitive Layer combines those natural language instructions with strongly typed YAML definitions to ensure accuracy across all of the data.
Fully centralized: Self-serve users get blocked when they need to pick an Explore, or a dataset. With the Cognitive layer, instead of an LLM asking you which Explore to use, you just get an answer. This is really hard because data warehouses themselves aren't centralized. We had to invent new ways of mapping multiple tables into a single logical concept that an end-user can understand.
Built-in feedback for AI data agents: The key to a reliable AI Agent is providing an environment with strong feedback loops. The Cognitive Layer is constantly advising Zoë, our AI agent, which metrics can be used (and how they can be used together). So Zoë can iterate and find a solution correctly and consistently.
The best part of the Cognitive Layer: it's a superset of a semantic layer, so it's fully backwards compatible. Typical YAML data modeling works the same way in a cognitive layer for delivering BI; the extra functionality is what delivers an amazing LLM experience.
Cognitive layers allow for data agents
With a flat semantic layer, a cognitive layer allows for large language models to interact with the data in a way that is extremely user friendly. Anyone at the company can now interact with the data without knowing anything about the structure or modeling of the data.
Zoë is the AI Data Analyst of Zenlytic. More than a chatbot, text-to-SQL or a data copilot, Zoë is a data agent with access to the full suite of data tooling in the Zenlytic platform. Zenlytic is a conversation first data platform that enables people to make decisions faster.
