In a recent webinar, Starburst CEO Justin Borgman and Co-Founder and VP of AI/ML Products Matt Fuller walked through how AIDA enables natural language access to governed enterprise data, generating queries, visualizations, and insights in real time. AIDA is the first AI assistant built to reason across all of your enterprise data. It connects directly to distributed data across data lakes, data warehouses, and other data sources without requiring it to be moved or centralized first.
For technical teams evaluating AI-driven analytics, the real value of AIDA doesn’t come from the demo alone, but also from understanding how this works in practice. Teams need to know how Starburst’s AIDA results are validated, how data is governed, and what it takes to operate at scale.
Below are the six questions that came up most in the webinar, along with answers for teams doing serious AI evaluations.
1. How do you know that AIDA’s answers are actually correct?
AIDA is built to be accurate and to improve the accuracy of its results compared to its competitors. This accuracy is driven by three things working together.
- First, internal benchmarking against Spider and Bird ensures that AIDA generates quantitatively and qualitatively accurate results against standards.
- Second, feedback mechanisms will also be introduced, allowing users to provide ratings and comments. This enables data owners to review, test against, and update context and data of data products to ensure continuous updates to accuracy.
- Most importantly, AIDA relies on well-defined data products and business context, such as descriptions, mappings, and semantics, to interpret questions within the context of your organization.
Since Starburst is built around universal access through data federation, that commitment extends to accessing the context needed to ensure accuracy. The better the context layer, the more reliable the answers.
Together, all three of these things impact accuracy, and we are confident that AIDA improves accuracy within probabilistic systems of this type.
2. Can data products be governed or restricted before AIDA uses them?
Yes, data governance is a key principle underlying AIDA’s design and the design of the data products AIDA draws on. Just like any data product, data governance is enforced through Starburst’s existing access control layer, which includes RBAC and ABAC policies, column-level and row-level controls, and the ability to mask or deny sensitive fields entirely.
Additionally, as an AI agent acting on behalf of a user, AIDA’s role is always clearly defined by the logged-in user. AIDA always operates under a defined role just like a human user, meaning what it can see or query is identical to the user – no more, no less. AIDA relies on the same robust and comprehensive access control you’ve come to expect from Starburst.
3. What does it cost to run AIDA?
AIDA is built to allow you to control costs in a modular way. Its cost model has two components, each of which varies depending on the deployment model. This means that the cost structure is different for Starburst Enterprise and Starburst Galaxy in the following way.
AIDA on Starburst Enterprise
For users deploying Starburst Enterprise, there is a tiered license cost, which is higher than the cost for Galaxy. Enterprise customers bring their own model, and model costs are managed directly by the customer.
AIDA on Starburst Galaxy
For users deploying Starburst Galaxy, there is a different, tiered structure. Galaxy customers using higher tiers (e.g. Enterprise and Mission Critical tiers), costs consist of the tier cost plus the token usage for the underlying models. Notably, these tiers do not operate under a bring-your-own-model approach. This contrasts with other features on Galaxy, like AI SQL functions, where a bring-your-own-model approach does apply.
In each case, AIDA allows organizations to manage enterprise costs in the way that works best for them.
4. Can AIDA query across multiple data platforms at scale?
Yes. Data federation is a core capability of Starburst’s underlying data foundation, and that is just as true of AIDA as it is any other Starburst-powered workflow. Notably, AIDA builds on this federation by allowing users to ask questions across distributed data sources (e.g., Snowflake, Redshift, Amazon Athena) without moving data. This universal access to data provides universal access to context, which is AIDA’s true superpower.
5. How does Starburst handle data quality in a federated environment?
Although data quality is not directly handled by AIDA, there are multiple ways that data quality may be managed through the underlying data foundation. For example, Starburst Galaxy includes built-in data quality features that are designed to work within federated environments. These same features are leveraged by AIDA indirectly.
For more information, you can read more about data quality and Starburst here. Starburst also coexists interoperably with a number of leading data quality technologies. This also provides the option of fitting around any existing data quality options within your current data stack.
6. Will AIDA replace traditional BI tools?
AI is changing the way people do Business Intelligence, and AIDA is here to support changing paradigms around BI intelligence. In the long term, we think that BI tools will be replaced by AI. In the short term, a full replacement is less likely than a change in how current tools are used.
At Starburst, we believe in optionality and choice, and that extends to AIDA. For organizations that are ready to make the full leap to AI from BI, AIDA can facilitate that. For others who are looking to change how they work and leverage AI in some use cases, AIDA is also a great choice for that use case. Over time, if you change and evolve your priorities, AIDA can help you get there. The choice is always yours.
What does all of this mean?
AIDA is a powerful tool for generating answers across your whole business. Powered by the Starburst Enterprise Intelligence Platform (SEIP), it is built for context, delivering governed, natural-language access to distributed data, without forcing you to move or centralize this data first. For teams spending more time wrangling BI dashboards than making decisions, that’s a meaningful shift.
If you want to see how it works in practice, the full webinar covers the architecture and live demos in depth.
Watch the full session: From Dashboards to AI Decisions: Rethinking Enterprise Intelligence.
