Sisense has launched an integration with ChatGPT that easily enables customers to automate time-consuming data preparation tasks. In addition, the integration lets Sisense users augment their own data, with ChatGPT’s vast access to worldwide data.
Founded in 2004 and based in New York City, Sisense is an analytics vendor specializing in embedded BI. Its Extense Framework, unveiled in 2021, enables users to infuse data products — including dashboards, models and reports — driven by augmented intelligence and machine learning into work applications.
ChatGPT, meanwhile, is a chatbot that was unveiled by OpenAI in November. It has already become widely used and the subject of worldwide interest.
It builds on the capabilities of GPT-3, which was first introduced in 2020. The AI model dramatically advances the question-and-response ability of chatbots so that analytics users can query data and receive responses — including detailed narratives — without writing code.
Partly as a result of its launch, many analytics industry insiders expect natural language processing to be a significant trend in 2023. Experts expect vendors to integrate ChatGPT into their platforms to ease the burdens on data scientists and data engineers, as well as enable business users to more deeply engage with data.
Now, Sisense has done just that, finding a good initial application in data preparation with an integration unveiled in a blog post on Jan. 6, according to Donald Farmer, founder and principal of TreeHive Strategy.
“Great work by Sisense to get this integration out into the wild while the excitement about ChatGPT is still hot,” he said.
It’s a very smart move because it gives users a very specific advantage in data preparation rather than some vague, generalized AI capability that is good for demos but not much else.
But Sisense’s integration with ChatGPT isn’t related to much of the excitement about the chatbot, which mainly has to do for now with creative writing, Farmer continued. Instead, it’s about semantic search and data classification and has practical applications to data management and analysis.
“It’s a very smart move because it gives users a very specific advantage in data preparation, rather than some vague, generalized AI capability that is good for demos but not much else,” Farmer said.
The integration
Sisense’s integration with ChatGPT is essentially a bridge, according to Amir Orad, Sisense’s CEO.
It connects customers’ Sisense instance with their ChatGPT accounts — at no cost beyond what they already pay for Sisense — to enable them to apply ChatGPT’s capabilities to their data in any way they see fit.
One of those ways is to take on onerous, time-consuming data preparation tasks like adding new fields to existing data sets.
For example, if a user wants to know how many of their customers are companies in a given industry, the information can be found only if industry is one of the fields in the data set.
If it’s not, adding that field for potentially thousands of customers would typically require data engineers to extract a data set from their organizations’ data warehouses, add industry to each company in the data sets, look up the industry of each company and input it into the industry field, and then reload the data set back into the data warehouse.
Sisense’s integration with ChatGPT enables users to automate that work and execute it in seconds, according to the vendor.
In order to ease its integration, Sisense developed a new user interface using its APIs to provide a question-and-answer environment. In that context, users can simply tell ChatGPT how they want to augment their data set and ChatGPT will do the work.
ChatGPT translates the request into a Python script and executes the task as if a data engineer had written the code.
And because ChatGPT is so simple and takes such little time to execute tasks, the integration not only enables analytics consumers to quickly do what might otherwise take weeks but also to take on tasks that would have been too overwhelming.
“We’ve built a bridge from the analytics product to the AI that takes the data you are working on and the question you want to ask, sends them to the AI, reads the answer, and puts it back with an ad-hoc table with consumable data in the BI tool,” Orad said. “And it’s literally done in a few seconds.”
Also critical is that ChatGPT is able to populate the new fields by providing the information itself from its own database, Orad added.
In the example of adding industry to an existing data set, ChatGPT is programmed to know the industry of a given company, just as it knows the author of a book, all the countries in Asia or who the president of the United States was during the Civil War.
“ChatGPT actually reflects, in a pretty good manner, humanity’s database,” Orad said. “It has data about anything. You can take some data set, and you can augment that data with information you get from ChatGPT.”
There is, however, a trade-off to ChatGPT’s ease-of-use, Orad added: Its data is not 100% accurate.
But the trade-off is likely worth it, Orad said, adding that doing something with 95% accuracy in two seconds is usually more beneficial than doing it with 100% accuracy in two weeks.
In addition, a user can ask ChatGPT what the source is for the given information requested and provided, giving the user the all-important data lineage. And, Orad noted, ChatGPT is in its first iteration and will only get better — including more accurate — with the passage of time. Also, ChatGBT’s data is current to only the end of 2021.
“Most of the data is extremely accurate, but you don’t have guarantees that everything will be perfect,” Orad said. “And in three years, this version will look like a children’s toy compared to what will be around.”
Among other data management and analytics vendors, Grid, a startup based in in Reykjavik, Iceland, has also released an integration with ChatGPT.
Grid, like Kloud.io and Sigma Computing, uses a spreadsheet interface to give users a familiar look to their analytics environment. And with ChatGPT, the vendor now enables users to complete complex spreadsheet formulas.
“This is a very useful scenario, which many users will find enables them to perform more complex analyses than before,” Farmer said in reference to Grid.
He added that he expects many data management and analytics vendors to follow the lead of Sisense and Grid and develop integrations with ChatGPT.
“For sure other vendors will follow,” Farmer said. “The ability to generate [code] from a natural language query is bound to be leveraged by many tools to improve the query generation experience.”
A sample dashboard from Sisense displays an organization’s revenue data.
More use cases
Other potential applications of the integration between Sisense and ChatGPT include some of the natural language query and generation capabilities already offered by some analytics vendors — not only Sisense but also ThoughtSpot and Tableau, among others — and sentiment analysis to identify the tone of a text.
Sentiment analysis can be done with AI tools, but different tools specialize in discovering different sentiments, Orad noted. For example, one AI tool might be able to detect whether something has a certain political bent. But to tell whether the tone is positive or negative, another tool is needed.
“You can do it all with ChatGPT,” Orad said. “You can understand the tone of voice, if it’s accurate or not accurate, if it’s liberal or conservative. You can understand if the style of writing is five years old or 50 years old. It’s a much more rich analysis of text than we had previously.”
He added that there will likely be many more applications that Sisense will learn of from its customers but hasn’t yet imagined given that ChatGPT is so new and the integration between Sisense and ChatGPT just launched.
Farmer, meanwhile, said he expects to see other vendors use ChatGPT to add natural language generation capabilities like data storytelling, which is the automatically generated explanation of data.
He cautioned, however, that ChatGPT likely won’t be able to deliver the same depth of description provided by tools from vendors like Tableau (via its acquisition of Narrative Science), Toucan Toco and Yellowfin that have invested in data storytelling and developed sophisticated data storytelling tools.
“The underlying model of narrative generation they use has a very detailed semantic understanding of the relationships between different data elements, so it can deliver very precise, accurate and meaningful text,” Farmer said. “ChatGPT … creates a pastiche, an imitation of other examples it has seen.”
As a result, some of the integrations between analytics vendors and ChatGPT could be disappointing.
“What Sisense has done is much smarter,” Farmer said. “They have focused on very particular and workable capabilities. And so has Grid.”
