ThoughtSpot expands AI offering - but keeps ‘humans in the loop’ and highlights importance of accuracy

Derek du Preez Profile picture for user ddpreez January 31, 2024
Summary:
ThoughtSpot has long used natural language processing to bring data analytics closer to all users. But with the advancements in Generative AI and Large Language Models (LLMs), it has been able to take its product further.

An image of ThoughtSpot being used to identify data held in different states in the US - with a map shown on the screen
(Image sourced via ThoughtSpot )

ThoughtSpot - the analytics vendor that in the past has described itself as ‘the Google for numbers’ - is not wasting time in recognizing the combined potential of its own technology with that of Large Language Models (LLMs), announcing today the expansion of its AI capabilities to help bring data insights closer to all users. 

ThoughtSpot is building on the release of ThoughtSpot Sage early last year, its enterprise data search capability that combined foundational language models with its own proprietary search technology. At the time the vendor had only announced integration with OpenAI’s ChatGPT, but this has since expanded to include Google’s Gemini - with the possibility of others in the future, if there is demand there from the install base. 

The founding objective of ThoughtSpot has been to get data analytics and insights into as many hands as possible across the organization. Its belief is that dashboards and super-coded tools limit knowledge to a select few and don’t allow users to flexibly react to new information as it becomes available. It wants knowledge workers to have just as much access to data as analysts and data teams - by simply being able to ‘ask questions’ of the system. Obviously, the rise in popularity of LLMs, which allow for follow up questions and a seemingly ‘human like interaction’ with a system fall right in ThoughtSpot’s wheelhouse - and it has been quick to capitalize on the opportunity. 

Not only did it announce the launch of Sage, which you can read diginomica’s full write up here, but it also acquired Mode Analytics for $200 million, which was aimed at bringing data teams and business users closer together. 

Today, ThoughtSpot has a number of new features that build on its AI capabilities, with a strong focus on keeping humans in the loop to enable accuracy and trust for enterprise users. We got the chance to sit down with Sumeet Arora, ThoughtSpot’s Chief Development Officer, to get a better understanding of the thinking behind the vendor’s strategy. Arora said: 

There is a renaissance in the data and analytics space. And finally we can move people from dashboard drudgery, which everybody acknowledges, to a world where every workflow and every person is able to derive business outcomes - leveraging insights from data. 

That’s the overall vision for ThoughtSpot: driving business outcomes for everyone and making powerful analytics kind of invisible in the back. It's there, it's helping you, but what's visible are the outcomes. 

The key updates

As highlighted above, ThoughtSpot’s proposition has always relied upon users asking questions of its platform and being delivered insights. However, instead of shying away from LLMs, ThoughtSpot believes the combination of its data mapping and natural language processing proprietary technology alongside LLMs is where really powerful outcomes can be delivered. Arora said: 

So we believe that is happening and obviously the reason it's happening is because the way humans interact with technology is changing. The way of dashboards and people building dashboards - that’s not natural to humans.

Generative AI and the ability for these foundational models to interact with users in a ‘human-like’ way, whilst maintaining the accuracy of the results with ThoughtSpot’s own technology is where it is hoping it can deliver real results for customers. One of the key announcements today includes: 

  • Advanced human-in-the-loop feedback - This incorporates a patent-pending feedback loop to teach and train ThoughtSpot Sage by providing associations between business terms and phrases and an organization's data model. Customers can view and edit all queries, as well as word and phrase-specific user- generated feedback, delivering a high degree of accuracy and trust to results.

According to Arora, one of the key pieces of feedback from customers using ThoughtSpot Sage was that whilst the technology was proficient in data mapping and providing helpful insights, data teams wanted the ability to nudge the training in a way that could reflect the needs of each individual organization. He said: 

Every organization has their own language. The feedback we got over the 200+ customer interactions that we had on Sage was: ‘can you make sure that there is human input into how you are learning the language of the organization and mapping it to the data of the organization? So that mapping we automatically create in ThoughtSpot. But the feedback we got was ‘we want to be able to give feedback to that mapping’ and that's what we have built.

[So], we have greatly advanced ‘human in the loop’. We have basically released a control system between AI and humans, so that you can deploy sage with a lot more accuracy. Sage automatically learns the mapping between business language and data taxonomy, but by adding the human in the loop allows data teams, analysts and admins to make sure that the learning is correct. 

They can actually provide human feedback to uphold things that are correct and also remove things that may not be correct.  incorrectly. This makes sage real in terms of enterprise deployments because that's what enterprises want. They want trust, accuracy, increased accuracy and control. We're giving that.

But to be clear, this isn’t about hallucinations, which have found to be common in the use of LLMs across broad datasets - where effectively the foundational model makes up an answer that it thinks could possibly, maybe be right (but is oftentimes fictional). ThoughtSpot is using LLMs for the conversational piece, but is relying on its proprietary technology to ensure accuracy. Arora said: 

There are no hallucinations in ThoughtSpot - [that’s been the case] from day one. Remember, we only use LLMs for what they're good at, which is to understand the intent behind the question. 

And then we use ThoughtSpot’s decade long work in natural language interpretation to 100% accurate analytics. There was never a risk of hallucination. But accuracy is a great point. 

Further to the human in the loop updates, ThoughtSpot has also announced: 

  • Conversational BI - Ask Sage expands the ThoughtSpot Sage experience to conversational BI, giving users the ability to ask follow up questions as they explore and interact with data visualizations, all through a familiar conversational, natural language experience. 

  • Embedded natural language search - With ThoughtSpot Sage Embed, customers can bring ThoughtSpot Sage into every product and web app, delivering GenAI experiences to end customers wherever they are working. 

  • AskDocs - With AskDocs, developers can now ask coding questions in natural language and receive contextual GenAI-assisted instructions and code from the Visual Embed SDK in both ThoughtSpot and developer documentation.

Using conversation and language to explore data, whilst integrating these capabilities into apps where users are working - not requiring them to exit and go into ThoughtSpot - is central to the vendor’s overarching ambitions. Arora said: 

What is analysis? Analysis is essentially breaking a complex question down into pieces. When you can do that in a chain, it allows the chain of thought in our head to be replicated in the technology. 

Remember we are trying to make powerful analytics invisible and make outcomes visible. Where are people? People are spending time in other apps, they don't have to come to ThoughtSpot. So everything that we announced here is embeddable into other apps. We’ve made it easier for the app developers to embed ThoughtSpot and have conversational BI in their app. 

Other product updates from ThoughtSpot today include: 

  • AI Assist - Data teams can use AI Assist to help generate SQL to refine, create, understand, troubleshoot, and optimize queries, helping businesses get insights more quickly and scalably. 

  • AI Highlights on Liveboards - This aims to help extract important takeaways from a Liveboard, surfacing both expected and unexpected changes since a user’s last visit, along with the attributes driving the most change.

  • KPI Monitor - Instead of waiting for you to login and check on a metric, KPI Monitor alerts you to what’s changing in your business and why. This is all explained to you in natural language via AI-generated change analysis and delivered wherever you are via the ThoughtSpot mobile app.

It’s all about the ROI

Taking a step back and looking at the bigger picture, I was keen to get an understanding from Arora about the advice he’s giving customers when it comes to the implementation of Generative AI and what buyers are asking of ThoughtSpot when it comes to metrics for measuring success. On the latter point, Arora is clear that businesses see the introduction of these tools as a clear path to efficiency and revenue gains. He said: 

Everything is ROI based. They see Gen AI and natural language as a way to get more value out of data. People want all data to be analyzed and everyone to benefit from it. - that's what data leaders want. Generative AI and natural language interfaces promise to really help achieve that. 

So a lot of the use cases are about: can I dramatically expand the reach of analytics in my organization? And actually not just to knowledge workers, but also even for analysts. It’s about improving their productivity, because they can do a lot more, faster, and get into much much more advanced analytics. 

So a lot of the use cases are around ROI for the data teams and the analysts and also the overall ROI for the organization. Really putting data to business benefit. 

And Arora is clear that 2024 is the year where we should expect more action and less talk when it comes to the use of Generative AI, as we hopefully move beyond the peak of the hype cycle. He said: 

What I would say is, this is the year of real deployment. It's going to happen. Last year people were making announcements and all that, but this is the year when deployments will be real. I want to say that customers should go in with eyes open around trust and accuracy. 

And the only way you have trust and accuracy is when you can actually architecturally guarantee that. And the only way to architecturally guarantee that for everyone is if the outcomes can be explained, right? Bet on a system that is architected for human feedback, that is architected for accuracy, trust and explaining. 

My take

I think it would have been easy for ThoughtSpot’s leadership to look over at the hype being generated by LLMs and say ‘we already have natural language processing, we’ve been doing this for years’. However, it’s a sign of maturity to recognize that it’s not a zero sum game and that taking the best of both worlds could lead to better outcomes for customers. But as Arora notes, 2024 is the year of implementation and outcomes - we look forward to the customer stories and seeing the results of these tools in action. 

Loading
A grey colored placeholder image