Getty Images/iStockphoto

ThoughtSpot update adds feedback loop to help train GenAI

The vendor continues to focus on generative AI, updating its conversational interface with more in-depth analysis capabilities and adding new methods of improving LLM accuracy.

Highlighted by a new human-generated feedback loop aimed at improving the accuracy of Sage, the vendor's generative AI-powered natural language interface, ThoughtSpot on Wednesday unveiled its latest analytics platform update.

ThoughtSpot first introduced Sage in March 2023 following an integration with OpenAI's GPT-3. Two months later, the vendor moved Sage into public preview, where it remains as ThoughtSpot works to continue improving Sage's accuracy and trustworthiness.

In addition to the feedback loop, ThoughtSpot introduced Conversational BI, an addition to Sage that enables deep data exploration using natural language.

Meanwhile, beyond Sage, ThoughtSpot's platform update includes an AI assistant to help generate code in Mode, a code-first analytics platform that ThoughtSpot acquired in June 2023. The update also adds generative AI (GenAI) features that surface insights in dashboards and help monitor KPIs.

Together, the new capabilities represent a significant update, given that they aim to augment humans with generative AI but don't remove responsibility for generative AI from humans, according to Kevin Petrie, an analyst at Eckerson Group.

"The right way to frame the GenAI renaissance [is that] creative humans are assisted by GenAI," he said. "Humans must remain in charge to mitigate risk. This is a good step forward for the application of language models to business intelligence."

The right way to frame the GenAI renaissance [is that] creative humans are assisted by GenAI. Humans must remain in charge to mitigate risk. This is a good step forward for the application of language models to business intelligence.
Kevin PetrieAnalyst, Eckerson Group

Based in Mountain View, Calif., ThoughtSpot is an analytics vendor whose search-based platform enables customers to explore and analyze data using natural language.

Like peers including Qlik and Tableau, the vendor has made generative AI a primary focus of product development since OpenAI's launch of ChatGPT in November 2022 was a substantial improvement in large language model (LLM) capabilities.


Unlike many analytics vendors whose platforms required customers to use code to query and analyze data, ThoughtSpot's platform was designed from its start to enable customers to use natural language processing (NLP) to work with data.

Analytics use within enterprises has stagnated at around a quarter of all employees for about two decades. By reducing the need to use code, ThoughtSpot and other vendors that developed NLP capabilities and tools aimed to simplify the use of BI platforms so that more people could work with data.

ThoughtSpot's platform, however, had a limited vocabulary, as did the NLP tools other vendors introduced to complement their code-first capabilities. As a result, customers couldn't use true natural language, and platforms such as ThoughtSpot still required at least some level of data literacy training.

LLMs eliminate the limitations of a finite vocabulary. They are trained on all available public data and have the vocabularies of a dictionary. In addition, they are trained to infer intent, reducing the need for business analysts to phrase queries in highly specific ways.

Sage is an integration of ThoughtSpot's existing search and analytics capabilities with LLM technology to enable true natural language interactions with data. Its new capabilities, meanwhile, are designed to enable deeper data exploration within ThoughtSpot than previously possible using natural language, as well as to help Sage learn a specific business so that it can answer questions unique to a given enterprise.

Conversational BI is a new capability in Sage that lets customers ask follow-up questions as they work with data, enabling deeper analysis. The feedback loop overseen by humans, meanwhile, trains Sage to learn an organization's unique characteristics by enabling users to verify whether the mappings between natural language and ThoughtSpot's technology are correct.

LLMs are good at understanding language and being able to answer questions about anything in the public realm. They are not, however, privy to the proprietary data organizations need to make decisions specific to their business.

Combining LLMs with proprietary data is thus key to helping enterprises derive value from generative AI, according to Sumeet Arora, ThoughtSpot's chief development officer.

"What people can do is teach the system," Arora said. "They can reinforce the correct behaviors and remove the incorrect behaviors."

That means of training Sage with human feedback is significant, according to David Menninger, an analyst at ISG's Ventana Research.

He noted that tools such as Sage are finally making conversational analytics a reality after traditional NLP tools failed to help analytics reach a broad audience. However, users need to be able to trust the results generated by Sage and other LLM-powered analytics tools. Generative AI is subject to hallucinations -- including false, misleading and offensive results -- so making sure the models are properly trained is essential.

Having a feedback loop that includes people to train Sage therefore can help lead to trusted data.

"By employing GenAI, Sage is much more powerful than earlier search-based capabilities ThoughtSpot offered," Menninger said. "They are ... aware of some of the pitfalls [of conversational interfaces] and have addressed issues such as trust and having a human in the loop."

In addition to the new feedback loop, Petrie noted that the improved conversational BI capabilities in Sage are also significant for ThoughtSpot users.

In the eight months since first being released in public preview, Sage has been able to respond to queries and searches in the same natural language customers use to ask questions and initiate searches. However, Sage was not able to have a true conversation with follow-up questions and suggestions. That level of analysis required more complex skills on the part of the user.

Now, Sage can carry on a complete conversation, enabling customers to engage in deep analytics to derive insights that can lead to decisions and actions.

"Data teams can explore, organize and interpret data faster than before by conversing with the platform rather than configuring or coding SQL commands from scratch," Petrie said. "This improves productivity, enabling data analysts and developers to complete projects in less time and thereby take on more work."

Screenshot of an AI-generated graph in a ThoughtSpot dashboard.
An AI-generated graph from ThoughtSpot displays the top bakery brands in California.

More new capabilities

Beyond the new features in Sage, ThoughtSpot's platform update includes six more new tools. Among them, Ask Docs is the most significant, according to Menninger.

Ask Docs is a feature that lets developers ask coding questions using natural language and receive generative AI-assisted recommendations -- including code -- so that they can document their work and build data applications faster and easier.

Menninger noted that while natural language search and query represent an important way generative AI can help analytics consumers, code recommendations and code generation represent another.

"Ask Docs is a great way to apply GenAI beyond conversational analytics," he said. "There are so many ways GenAI can enhance the user experience. Having a conversational experience with the documentation is so much more productive than searching for the right information."

In addition to the new features in Sage and Ask Docs, ThoughtSpot's platform update includes the following:

  • AI Assist, a tool in Mode aimed at helping users generate SQL code to develop and refine code-first data queries.
  • AI Highlights on Liveboards, a feature that automatically surfaces changes to interactive dashboards and explains the drivers behind the changes.
  • KPI Monitor, an automated tool that observes metrics for changes and alerts users when changes occur, including AI-generated analyses that explain the reasons behind changing KPIs.
  • ThoughtSpot Sage Embed, a tool that enables customers to embed Sage into all of their data products and applications.
  • AI Center and Model Interoperability, features that are designed for data administrators to set privacy and security controls on all of ThoughtSpot's AI capabilities.

In sum, the new capabilities included in ThoughtSpot's update represent progress for the vendor, according to Menninger. In particular, it demonstrates generative AI being put to actual use.

"To the extent you consider GenAI and conversational computing significant, this release is significant," Menninger said. "ThoughtSpot has put GenAI features in market earlier than many other analytics vendors and has introduced early releases of some of its future capabilities."

Arora, meanwhile, noted that the release represents ThoughtSpot's continuing effort to remove data from dashboards and use AI to deliver analytics to users within their workflows. Sage Embed is one tool that epitomizes that evolution. KPI Monitor is another.

"We are moving people from that world of dashboard drudgery to a world of business outcomes for everyone, delivered by AI-powered insights from data," he said. "The only way it happens is if we leverage AI to find insights that are relevant and powerful, and make them easy to [consume]."

Future plans

Looking ahead, Arora said ThoughtSpot is making a push to help customers expand their deployments with Sage and other generative AI tools that have the potential to help more employees within organizations work with data.

In addition, ongoing efforts include continuing to improve the accuracy of generative AI-powered insights so that users can trust their results and enabling customers to choose the LLMs they want to use with their proprietary data, Arora continued.

"This is the year GenAI goes from all the hype to actual use," he said.

Key to ThoughtSpot's work toward improving generative AI accuracy will be the continued involvement of reinforcement learning from human feedback, according to Petrie.

He noted that the introduction of the feedback loop in the vendor's latest update is just the beginning, and there will be fine-tuning that needs to be done over time.

"Human experts must oversee, inspect and correct both inputs and outputs when using GenAI, given governance risks such as hallucinations, bias and privacy breaches," Petrie said. "ThoughtSpot's advanced feedback loop is a good step toward mitigating [risks] and making language models more trustworthy. It's a long-term process, and we're in early days."

Menninger, meanwhile, is an advocate of planning tools that help organizations explore and prepare for various potential scenarios. He said he'd like Sage and similar generative AI tools from other analytics vendors to do more forward-looking analyses.

In addition, he suggested that the application of generative AI to data pipelines represents an opportunity for ThoughtSpot and its peers to expand.

"It will be interesting to see how far analytics vendors go to address data pipelines with GenAI," Menninger said. "Since that is the most time-consuming part of analytics, it could significantly improve productivity of data and analytics teams."

Eric Avidon is a senior news writer for TechTarget Editorial and a journalist with more than 25 years of experience. He covers analytics and data management.

Next Steps

New embeddable MicroStrategy Auto bot expands AI beyond BI

Dig Deeper on Business intelligence technology

Data Management
Content Management