The generative AI revolution is upon us, and it’s having an impact across all of IT, including the business intelligence and analytics tools used by analysts and executives to glean insights out data. But BI and analytic tool vendors may not be using ChatGPT and other large language models in the way you imagine.
Among BI and analytics vendors, Microsoft arguably is ahead of the curve when it comes to AI, thanks to its close partnership with OpenAI, the creator of ChatGPT. Last month, Microsoft announced that it has added Copilot, its AI application based on OpenAI’s GPT-3.5 (and now GPT-4), to various Microsoft 365 offerings, including Excel and Dynamics 365 Customer Insights.
Microsoft doesn’t appear to have adopted Copilot for PowerBI, its flagship BI and analytics offering, at this point. However, that hasn’t stopped PowerBI customers from using ChatGPT with the analytics app. One user shared on the open Microsoft PowerBI forum how he used ChatGPT to create a date table query in PowerBI.
ThoughtSpot is one BI and analytics vendor that has done the work to integrate OpenAI’s GPT service directly into its product. Last month, the company announced ThoughtSpot Sage, a new GPT-3.5 powered option designed to allow users to “chat with their data,” among other things.
The new offering, which is currently in preview, incorporates GPT-3 in three ways. First, ChatGPT is now an option when using natural language query (NLQ) functionality for searching the ThoughtSpot catalog for pre-existing queries and content. Secondly, it can serve insights from customer data through LLM-powered SQL generation, and create narratives around those generated insights. Lastly, GPT 3.-5 can be used during the data modeling process by generating “synonyms” for data tables, lowering the barrier of entry for working with data in ThoughtSpot.
ThoughtSpot has been at the forefront of using NLQ and getting users comfortable with search-based interactions in its BI offering. The work it’s done to incorporate ChatGPT takes that NLQ work to the next level, says Amit Prakash, the CTO and co-founder of ThoughtSpot.
“What we’re announcing today is a complete rebuild of the natural language processing engine that we had, integrating GPT-3 into it,” Prakash told Datanami last month. “So everything that we’ve been working on all this while to capture the institutional knowledge and apply that in the context of those questions, gets merged with what GPT can do, and then the combined product is looking very promising.”
ChatGPT and all LLMs are prone to hallucinations, which has to be a big concern for BI and analytics vendors. ThoughtSpot addresses ChatGPT and GPT-3’s tendency to make things up by surrounding the API calls to the LLM with lots of careful prompting and pre- and post-processing, Prakash said.
“What you’re doing is injecting the knowledge we have about the data model, about how people like to ask questions, and how they like answers, through sort of fine tuning and prompt engineering to be able to get the reliable answers, as well as sort of preprocessing and postprocessing the before and after, to simplify the task for GPT so that we can get much higher reliability and interactivity,” he said.
Sisense, which specializes in providing embedded analytics for existing applications, is also working with generative AI models. The company ran a blog post in January that describes how its customers can use various Sisense components and GPT-3 to extend their data analytics environment with a generative AI-powered application that can automatically source data from the Internet for use in analytics.
“Making use of GPT-3 is pretty straightforward on a one-off basis,” Sisense Chief Product Officer Scott Castle writes in the blog. “However, leveraging the power of large-language models for business intelligence at scale requires a highly extensible analytics platform in order to make the source data easily accessible, and to integrate the augmented results back into the flow of analysis.”
Domo customers are also starting to play around with ChatGPT integration with the Domo BI product. According to a February presentation, the company is working with partner Pramana Labs and its NLQ offering, dubbed SHIFT, to incorporate ChatGPT functionality into Domo.
“ChatGPT has created a huge amount of buzz since its release in November 2022. At Domo we have been busy exploring how it could improve the experience for Domo content creators and users alike,” writes Domo Global Community Manager Anna Yardley. “The underlying technology is always fun to explore, but it comes to life when we can add business context and leverage the power of the Domo platform behind it.”
Pyramid Analytics has delivered an integration with GPT and its Decision Intelligence Platform. According to its March 20 press release announcing the integration with OpenAI, Pyramid is using the AI in several ways, including generating SQL, DAX, and MDX code for data extraction queries; generating Python and R code to power machine learning logic; to build spreadsheet formulas; and to generate designs for contents and graphics for storyboarding and publication of reports.
Another BI and analytics vendor that’s adopted ChatGPT is AnswerRocket, a 10-year-old Georgia company that counts customers like Kimberly Clark, Pabst, and EMC Insurance as customers. AnswerRocket adopted ChatGPT with its latest release to enable conversational analytics on customers’ own data.
Tellius is also adopting ChatGPT with Copilot, a product it launched on March 17. The new offering leverages GPT to automate several tasks for the BI user, including generating natural langauge synonyms for data, generating SQL and Python code for data transformation routines, and generating narratives that summarize insights surfaced by the query engine. Tellius Copilot is currently in limited preview; the company is hosting a webinar about it on Monday April 24.
GoodData, another provider of analytic tools for embedded use cases, doesn’t appear to have a formal integration with ChatGPT or GPT-3 at this point. But that doesn’t mean that users aren’t playing around with it and getting something to work. Jan Kadlec, a GoodData engineer, wrote in January about how ChatGPT can be utilized in the BI tool via GoodData’s Python SDK.
Salesforce has gone all in on generative AI with its CRM application via Einstein, the in-house AI system used to generate predictions within Salesforce platforms. Last month, the company unveiled Einstein GPT, which can be used to automatically respond to customers as a chatbot, generate emails, schedule meetings, summarize sales opportunities, and create knowledgebase articles and backgrounders.
Salesforce plans to use Einstein GPT across other Salesforce properties, including MuleSoft, Slack, and Tableau, the BI and analytics vendor that Salesforce bought for $15.7 billion in 2019. Tableau is promoting Einstein GPT on its website. It also has a session on Einstein GPT scheduiled for the Tableau Conference being held in three weeks.
Qlik doesn’t officially support GPT, ChatGPT, or another LLM. But that hasn’t stopped customers from playing around with the two products. Qlik power user Göran Sander last month posted a blog about his use of ChatGPT with Qlik Sense. The early returns are mixed. “Can a language model like ChatGPT be used to create code for Qlik Sense?” Sander wrote. “The answer is yes–and no. It’s very impressive indeed but doesn’t quite replace us humans yet.”
MicroStrategy has ChatGPT on the drawing board. At its MicroStrategy World user conference next month, one session is dedicated to a discussion about that integration. “The Product Management team will take you behind the scenes for an exclusive on the most anticipated roadmap item to date: future integration with ChatGPT, the AI chatbot rocking the world,” the session description reads. “Learn more on how we plan to integrate the exciting language-based model to answer questions, deliver dynamic data stories, and make data insights more accessible for all users.”
This is still very early days for integration of LLM models like ChatGPT and GPT-3.5 (and the just-released GPT-4) into BI and analytics tools. Sisense and AnswerRocket garnered the only GPT nods from Gartner in its recently released Magic Quadrant for analytics and BI tools. Data science and ML platforms may be slightly ahead of the game when it comes to using LLMs.
In lieu of the current void when it comes to the integration of LLMs with traditional BI and analytics tools, it’s possible that a new class of analytics product could swoop in and occupy this potentially fertile ground.
GPT-4 Has Arrived: Here’s What to Know
Like ChatGPT? You Haven’t Seen Anything Yet
Hallucinations, Plagiarism, and ChatGPT