Tableau, Informatica, ThoughtSpot Tout Generative AI
A number of data and analytics vendors are having their annual conferences this week. And three of the vendors putting on big events — Salesforce/Tableau, ThoughtSpot and Informatica — are using their conferences to announce new generative AI features in their platforms, leveraging OpenAI’s Generative Pre-trained Transformer (GPT) Large Language Model (LLM). They’re making other announcements too, and since there’s thematic unity across them, we’ll enumerate and analyze them together here.
Tableau Conference is serving as the launch vehicle for Tableau GPT, Tableau Pulse and Tableau’s VizQL Service. Informatica World (which, like Tableau’s event, is being held in Las Vegas), is the occasion the data management juggernaut is using to announce CLAIRE GPT and Claire Copilot. And at its Beyond virtual event, ThoughtSpot is ushering in the announcement of ThoughtSpot Sage, which uses GPT to supercharge the natural language search- and AI-driven business intelligence capabilities that ThouhtSpot has long championed.
All three vendors are using GPT to enable natural language interfaces to their core functionality. In Tableau’s and ThoughtSpot’s cases, this includes querying, explanatory analysis, insights, and suggested further inquiries. For Informatica, it includes generated data classifications; inferred data lineage; multicolumn completeness analysis; as well as data pipeline generation, automated data masking, data cleansing and standardization rules for what the company calls “self-integrating systems.”
The vendors are also implementing user feedback loops to refine their models’ training on an iterative basis. For example, Informatica says CLAIRE GPT can “learn from feedback from users and continuously adapt to generate more accurate and relevant responses” and ThoughtSpot includes “human-in-the-loop feedback capabilities.”
It’s important to understand that rather than using OpenAI’s ChatGPT service, Tableau, ThoughtSpot and ostensibly Informatica are using OpenAI’s base GPT model and integrating that into their own platforms. ThoughtSpot’s co-founder and CTO, Amit Prakash, explained in a briefing with The New Stack that GPT is being used to enhance the search, data modeling and ontology capabilities the company’s platform already had. Rather than just generating a base SQL query, GPT is being used to build search tokens as well, and then subsequently generate the more refined SQL that is actually executed. GPT can also be used to identify synonyms, allowing more flexible natural language search, without requiring the synonyms to be manually enumerated into the model.
Tableau is integrating GPT in a similar fashion and, like ThoughtSpot, says it’s doing so in such a way as to enable the use of other LLMs in the future, rather than being tightly coupled with OpenAI’s technology. Compare all this to how GPT technology might be used at the database level, as with the SQL GPT technology that analytic database provider Kinetica announced last week. Kinetica is using ChatGPT to generate SQL but it’s adding its own Web service above it to anonymize queries for the customer and is leveraging its own use of GPUs and vector operations to execute the ChatGPT-generated SQL (the complexity of which is indeterminate) in an efficient manner.
Tableau’s LLM capabilities are surfacing particularly prominently in a product it’s calling Tableau Pulse. Pulse allows users to query and follow metrics, and share them through Slack (also part of the Salesforce family), email, mobile and dashboards. It’s also integrating with Salesforce Data Cloud, which “harmonizes” a company’s data in Salesforce and external sources, aiding customer 360 and similar scenarios. Salesforce says its Data Cloud will support “zero-ETL” integration with Google BigQuery and, through Tableau, will support “hyper-accelerated queries” and “instant analytics.”
ThoughtSpot is integrating with Slack too, as well as Excel, Google Sheets and Google Slides. The integrations are made possible, respectively, by the Spot Slack AI assistant, ThoughtSpot Analytics for Excel, ThoughtSpot Analytics Sheets/ThoughtSpot Connected Sheets and integrations with Google Connected Slides. And in the realm of following metrics, ThoughtSpot Monitor for Mobile will let users track specific KPIs by subscribing to them in the ThoughtSpot Mobile app and receiving notifications on their iOS or Android devices when the underlying metrics change.
Want more? ThoughtSpot is releasing new live query database connectors for Amazon RDS for PostgreSQL, Amazon Aurora for PostgreSQL, Microsoft SQL Server and a “universal” JDBC connector. There are also data catalog integrations with Alation, Atlan and Collibra, Metaphor and data.world. The company is also launching a SaaS version of its platform on Google Cloud, allowing customers to purchase ThoughtSpot with Google credits on Google Marketplace.
And back in the data integration arena, Informatica is adding support for Azure Private Link to complement the support it added for AWS PrivateLink last year. It’s also adding support for customer-managed keys across the respective key stores on AWS, Azure and Google Cloud, as well as a new Secure Agent Key Vault feature that will work with AWS Secrets Manager, Azure Key Vault and HashiCorp Vault.
Modeling and Embedding
In the world of data modeling, ThoughtSpot is adding a new Data Modeling Studio for its native platform; integration with Looker LookML models; and model sync for dbt, allowing for direct connections to models already built in dbt. Tableau, meanwhile, will be launching a new VizQL Data Service, letting customers leverage centralized data models and allowing Tableau to work as a “headless” BI back-end service, supporting numerous embedded BI scenarios.
Also on the embedded BI front, Tableau Embedded Analytics is launching a developer preview of a new “Embedding Playground” and will GA support for ephemeral users this year. Meanwhile, the ThoughtSpot Everywhere embedded BI platform is adding customization capabilities for Liveboards, a new CSS-based styling framework, and support for Git-based version control.
Good Things, if You Wait
If all of this sounds futuristic, much of it is. While a few things are hitting general availability (GA) today, many are only in private preview, and a few aren’t even that far along yet. ThoughtSpot Sage is in private preview, while Tableau Pulse and Salesforce Data Cloud’s zero-ETL for Google BigQuery will both “pilot” in H2 2023 (i.e. the second half of this calendar year), while hyper-accelerated queries on Data Cloud for Tableau, as well as Ephemeral users on Tableau Embedded Analytics, will GA in that timeframe. Tableau’s VizQL Data Service will enter a developer preview in 2024.
ThoughtSpot likes the H2 2023 timeframe too, making its Slack integration available within it. Right now, Slack integration is in private preview, as is ThoughtSpot Sage, Monitor for Mobile and integration with Google Connected Slides. Monitor for Mobile will be available “in the coming months,” but there is no mention of Sage’s GA timeframe at all. ThoughtSpot Analytics Sheets and ThoughtSpot Connected Sheets are available today, however.
Informatica’s CLAIRE GPT will become GA in H2 2023 and will enter preview in July. Informatica’s customer-managed key support is coming this Spring (which should mean real soon) while Azure Private Link support is coming this summer and Secure Agent Key Vault is coming this fall.
Look a Few Moves Ahead
It’s important to note that H2 2023 releases could come as late as the very end of the year, which is almost eight months away, leaving more than enough time for a new version of GPT, with new capabilities, to emerge. That means these features could be even more impactful than their private preview incarnations might suggest, covering additional use cases, and making analytics even more accessible to less-technical users.
Perhaps that potential innovation bonus is why Tableau, Informatica and ThoughtSpot are holding their GAs until H2, but they may need to hasten their release cadence to keep up with the pace of change in the LLM arena. It’s hard to know, though: ironically, while generative AI is based on predictive linguistic computation, it’s pretty hard to predict what the technology will do next and how soon it will do it.