Rubrik co-founder Soham Mazumdar, who left in 2023, has a new data startup called WisdomAI. The company offers AI data analytics that can deliver business insights with structured, unstructured, and even “dirty” data, meaning data not cleaned of typos or errors.

Working with data where and how it is, that’s essentially the holy grail for enterprise business intelligence software and why Coatue led the giant seed round of $23 million. Madrona, GTM Capital, The Anthology Fund, and others also participated. 

Rather than asking a data analytics team to run reports, business managers can ask WisdomAI questions and drill into the details. 

Mazumdar gives an example of a chief of revenue wanting to know, “How am I going to close my quarter?” WisdomAI’s answer would offer a list of pending deals the team should focus on, along with the information on what’s delaying each one, such as the list of the questions each customer is waiting on. 

“You can get the CRO to literally see all the way down to this last level of detail through our platform with, like, five key strokes, as opposed to a process which involves five individuals, including some analysts, and a whole lot of time,” Mazumdar told TechCrunch.

That’s just one example of the type of questions WisdomAI hopes to answer.

Another early customer is an oil and gas company that has thousands of workers in the field using WisdomAI to ask questions about production, tapping into data from everything from stored documents to telemetry.

Obviously, every business analytics tool already available — and a host of startups — are also offering AI-powered natural language prompts.

WisdomAI stands out for the pedigree of the founders — all previously worked with Mazumdar at Rubrik. But the platform’s superpower is its accuracy, even against messy data, Mazumdar says. It can find answers in structured data like databases as well as unstructured data stored in files.

Equally importantly, WidsomAI won’t deliver hallucinations.

Most enterprises are pursuing AI app accuracy by focusing on the data used to train their AI models, as well as model size, prompt engineering, and, perhaps, real-time retrieval techniques like retrieval-augmented generation (RAG). Yet they still run the risk of fabricated answers.

WisdomAI uses GenAI in the query formation — not in the creation of answers. “Ultimately, GenAI can hallucinate. What we use GenAI to do is to write small little programs … that can query these different systems,” Mazumdar says. 

So if WidsomAI’s model hallucinates, all it will do is write a fake query that fails to retrieve data. The data itself – the answer to the question – won’t be fabricated.

WisdomAI claims ConocoPhillips, Cisco, and Descope as early customers and has customers who are working with major cloud data storage services like Snowflake, Google’s BigQuery, Amazon’s Redshift, Databricks, and Postgres. It can be trained on any data storage system by studying the query language through query logs and other sources, Mazumdar says.