AI for AI: how to make "chat with your data" attainable within 2025
Since June 2025, Snowflake and Databricks customers across the globe are considering "AI-driven self-service analytics" to be a top priority. There's a ton of excitement, but also hurdles.
Earlier this month, Snowflake’s CDAO, Anahita Tafvizi, sat down with Forbes’ Maggie McGrath and shared her team’s own journey in leveraging Snowflake Intelligence and the new AI capabilities out of Snowflake.
Anahita’s team is Snowflake’s “customer zero”, they get to use new functionality before it gets released to customers, as part of the company’s D&A org supporting Snowflake’s own business. Great example of dogfooding.
Anahita’s team built a Go-to-market-AI capability internally - helping sales people at Snowflake get answers regarding Snowflake’s own products and sales stats. This allowed people to get access to information in minutes, instead of what would have taken weeks. This is an important point to focus on - it’s not about productivity boost. It’s about getting answers 100x faster. They sound like two sides of the same coin, but one side is far more valuable than the other.
Snowflake isn’t the only one to realize this. Databricks, one of Snowflake’s biggest competitors, came to the same realization as well. Both companies touted their AI capabilities in June 2025 in San Francisco, at their key conferences (back to back in the same location, funnily enough). The technology, the foundational models, are finally capable of generating reliable insights out of structure data. This is a trillion-dollar opportunity for these vendors.
Snowflake calls it Cortex Analyst and Intelligence.
Databricks calls it Genie and Databricks One.
Now, thousands of Databricks and Snowflake customers around the globe are looking closely at what AI can do for their self-serve analytics initiatives. “chat with your data” has become center stage.
Case in point: I recently spoke with a 4,000 person company in the US. They said: “When we last met you, in Feb 2025, AI for analytics was not a priority at all. It was something we thought we might consider for 2026. Now, in Sep 2025, it’s a top priority for our D&A team.”
AI for Analytics has arrived.
“Quality is very important to build user trust”
In the same Forbes episode, Anahita talked about how difficult it was to test the GTM AI bot, making sure the results were trustworthy. “The effort we invested was even more than the efficiency gains at times.”
This is not a Snowflake technological issue. This is a broad AI issue. To make data accessible through AI, you need to:
Document and model your data, in a way the AI can understand.
Test the interface - for each model, run dozens or hundreds of questions and assess the accuracy of its responses.
Keep the model up to date, and re-test, as new columns/tables are added or changes are made.
Each of these bullets, takes massive effort. They are also things human hate to do: 1. documentation is something we are terrible at, 2. doing the same thing again and again is demotivating.
So, there’s massive potential here - to accelerate time to insight by 100x - but also an enormous amount of effort required. The effort could outweigh the benefits. One CDO who has gone through this told us it cost her company more than $300,000 to do just the first two bullets above. That’s a lot of money and effort, and doesn’t even include the on-going maintenance of the models.
The solution: use AI to enable AI
A lot of the work required to enable AI for analytics can actually be automated, with AI. The understanding of the data, modeling it, documenting it, automatically testing it and updating it again and again… can be accelerated drastically with AI.
Can AI do 100% of it? No. Not today.
But it can do 80-85%. So instead of spending $300,000 to set up AI for analytics, you can spend $50,000.
It’s not just making this process cheaper. It’s also about making it faster (standing up an AI data chatbot in days, and not months) and more accurate (through rigorous, automated, testing).
Then, you use normal software capabilities to build a lifecycle around this entire thing, and you end up with this:
So, AI for AI.
Solid is one of the vendors doing this - helping Snowflake and Databricks customers accelerate their journey of “chat with your data”. If you’re interested to see more, contact us today.