So far this week, we’ve seen generative AI coming to CRM from Salesforce and customer service chatbots from Forethought, and those are just the ones I’ve personally covered. Today, we’re looking at ThoughtSpot’s generative AI input, which allows you to query your data using natural language to get text or a graph, as appropriate, with the correct answer.
This is an approach Thoughtspot has been working on for years. In 2019, when I spoke to the company about its $248 million Series E (the company was valued at $1.95 billion at the time), it was already using AI to translate plain language queries like “what’s best-selling shoe in Portland?” ‘ in SQL behind the scenes and delivering a response.
That’s not that different from what you’re announcing today, but it now relies on GPT-3 to allow users to enter a query and get a similar result. It just took some time for the technology to catch up with the vision.
“We always wanted to build an interface based on pure natural language intent. In fact, I can tell you that four years ago we had an internal project to build our own large language model. We paused because we knew that when the capabilities of the public large language model became available, we could put it on top. [of our our products] and offer the best, most flexible and highly accurate platform, and that’s what we’ve done,” ThoughtSpot CEO Sudeesh Nair told TechCrunch.
Perhaps the biggest criticism of ChatGPT is that it sometimes gives the wrong answer, but it’s essential that Thoughtspot provide an accurate answer when using technology to query data. In this sense, the company takes advantage of the GPT-3 API to help translate natural language into SQL, but also adds its own layer to ensure that it returns the only correct answer because with data there is no room for error.
“That’s why while large language models make sense, making them reliable for business computing, for database queries, is a complete game changer and…we’ve actually built the stack differently to provide accuracy and trust at scale in large companies,” Nair said. .
The company understands that no matter how hard you try, you won’t always get it right, so it’s also created a feedback loop to let you know when you’ve made a mistake, either because it’s inaccurate or because the customer presented the data differently. How does the algorithm do it?
The user can change the way it measures something by editing the query, or giving a thumbs up or down, depending on the answer, and the program can use this feedback to adjust answers in the future.
Different types of AI come into play, both when the user asks the question and when Thoughtspot retrieves and generates the answer. Additionally, Thoughtspot AI can help enterprise data experts create custom data models for their source data.
The company was founded in 2012 and has raised more than $660 million, according to Crunchbase. TO private beta of the new integration with GPT-3 opens today.
