LLMs are poised to make lumbering business intelligence tools easier and faster to use

At the moment, large organizations often employ “business intelligence” (BI) tools to figure out what the heck is going on inside their operations. This has spawned many lumbering leviathans in the software world.

Now, UK startup Fluent has closed a $7.5 million seed investment round led by Hoxton Ventures and Tiferes Ventures to apply AI-based Large Language Models (LLMs) to business databases, making them far easier to interrogate by the average person.

Essentially, BI tools connect to a business database and use SQL to create visualizations and build out BI dashboards. There are huge companies involved in this space: Tableau (owned by Salesforce), Power BI (owned by Microsoft), Looker (owned by Google), and QuickSight (owned by Amazon) to name just a handful.

The market for solutions is massive. According to one report, the global business intelligence market was valued at $27.11 billion in 2022 and is projected to grow from $29.42 billion in 2023 to $54.27 billion by 2030. Gartner thinks it could be even larger if AI and LLMs are more widely applied.

However, data teams spend lots of time building out these dashboards, especially for large organizations. And there is always the challenge getting people to actually use them — a hard task when data teams groan at the thought of fulfilling requests that could take days to build.

Instead, Fluent wants to be a “conversational layer” via natural-language LLMs that sit on top of a company’s data warehouse. It translates those questions into SQL and generates those answers much faster. So anyone, regardless of technical skills or business context, can ask questions in plain English of their data and obtain insights, according to the company.

Of course, this is likely to significantly shorten response times. Robert Van Den Bergh, CEO of Fluent, told me: “Consultants move from waiting 2 weeks for an insight to 30 seconds. That means they ask lots more questions, use data considerably more in their job. Data becomes something that’s now in their reach.”

Fluent’s clients already include Bain & Company.

Although he admits Fluent is “primarily using Azure OpenAI’s GPT4 model,” he stressed this is not a startup with an “OpenAI wrapper.”

That simplistic approach doesn’t work for generating accurate SQL and, therefore, correct answers to data questions in the BI tools context, he claimed. “Through 18 months of work, we’ve been able to build a method to achieve the accuracy of answers that organisations like Bain & Company can trust and leverage across their organisations.”

Ian Weber, a partner at Bain & Company, said in a supporting statement, “Fluent’s platform has helped us leverage LLMs to interrogate and deliver insights from large complex datasets. Fluent allows our consultants to quickly get the answers they need efficiently and accurately, especially for questions too complex or specific for pre-built data dashboards.”

Van Den Bergh said, “All business users want is answers to questions. They don’t want to do modelling. They want to know how this client was performing versus this client. Or how [they are] doing here. And how is this marketing campaign performing.” He said other players in the market target data users, whereas Fluent targets the business market, not data.

The space of natural language querying has only recently become possible, so it’s not yet a crowded market.

For example, Metabase is an open-source analytics and business intelligence application that allows users to create dashboards more easily. The SF-based company has raised $51 million to date.

Einblick, is a U.S. company that was recently acquired by Databricks (which is positioning to go public), appears to be the closest player to Fluent in the market. However, Fluent claims Einblick’s offering tends towards more technical users within data teams.

Thoughtspot, which has claimed a $4 billion valuation, now also has a natural language querying system.


Leave a Reply

Your email address will not be published. Required fields are marked *