Atlassian's Secoda acquisition adds data catalog for Rovo AI

Atlassian has acquired Toronto-based data catalog company Secoda for an undisclosed amount to bolster structured data management for its Rovo AI product.

The deal, disclosed this week, echoes similar acquisitions of data management companies by other enterprise service management vendors this year, including ServiceNow's Data.world buy and Salesforce's $8 billion deal for Informatica. In all cases, the goal is to better organize and govern data for use by AI agents by adding a data catalog, which presents a unified view of data assets in an enterprise, including where they are stored, when they were created and accessed, and their transformation and flow between systems.

In Atlassian's case, Rovo AI agents have seen strong adoption this year.But some large enterprise customers have struggled with organizing large amounts of structured corporate data in a way that makes them most useful for AI, according to Tiffany To, executive vice president and general manager of enterprise and platform at Atlassian, in an interview with Informa TechTarget this week.

"They go through this mental model of going, 'I've got to do something in AI, let me get access to a bunch of LLMs [large language models]," To said. "And then they realize, 'I have to do a lot of homework with my data -- it's not about tuning a custom LLM or training it.' If I don't have my data actually organized, if I don't have context, if I don't have secure governance for it, a lot of people get stuck."

Tiffany To

So far, Atlassian's Teamwork Graph has set that context for Rovo in unstructured data -- things like Jira tickets, Confluence documents and Google Drive files -- but to complete the picture and better inform enterprise decisions using AI agents, customers need similar access to structured data, according to To.

"Customers are often asking things like, 'How many Jira issues are in progress right now?' Or, 'Is this team running ahead of schedule or not?'" she said. "They're often querying the metadata that's tracking all of these projects, and today, often that's done quite manually with analytics. … It's quite a time-consuming process."

Naturally, customers want to use Rovo agents to answer these questions, To said. "But that means Rovo needs to deeply understand how that structured data is organized … so you have both relationships between objects in the Teamwork Graph, and Secoda gives you the greater context of what those objects are."

One Atlassian customer that already has Rovo AI agents in production welcomed the addition of Secoda.

If [AI agents] are confused as to the fundamental meaning of the data in a data store, it's going to be a disastrous train wreck.

"We've been using Rovo Enterprise Search for almost nine months now. The next step was the AI agents -- we have almost a dozen agents in production, quite a few in testing and a big backlog," said Kasia Wakarecy, vice president of enterprise apps, data and AI at Pythian, a data and analytics services company based in Ottawa. "We generally find that AI deployments require strong data foundation to be more accurate and precise so any additional ways to govern the data is critical to success."

Taking a page from Palantir's playbook

Secoda will add a corporate ontology specific to each customer to the Teamwork knowledge graph, said Charles Betz, an analyst at Forrester Research.

"We need AI to have a clear semantic model, also called an ontology or a glossary," Betz said. "AI agents are going to be operating at speed and with a lot of power across APIs and all kinds of internal resources, and if they are confused as to the fundamental meaning of the data in a data store, it's going to be a disastrous train wreck."

ServiceNow, Salesforce and Atlassian are tackling this problem in a way that's similar to Palantir, with a strategy the industry has dubbed a 'data flywheel.'

"Palantir continually updates a dynamic data model … constantly examining the meaning of the work a company is doing and the definitions of the language being used, building a dynamic graph and then using that to inform workflows," Betz said.

Charles Betz

What distinguishes a company like Atlassian from Palantir is the data it already has in Jira, which Betz called "an absolute gold mine of semantic information about thousands of enterprises around the world.

“That kind of metadata increases the organization's capability to understand itself and to create new value, because it is in Jira that people start first telling the story of what it is they're working on and what it's supposed to do."

Initially, Secoda's IP will be integrated with Jira Assets, Atlassian's configuration management database, a core feature of its cloud platform that also underpins products such as Jira Service Management. That work will focus on creating insights into structured data separately from the unstructured data covered by Teamwork Graph, but eventually, the plan is to unify those two sides of the data management picture, To said.

Users will have direct access to a training workflow for Secoda's agents under that integration, which, To said, was a selling point for Secoda as Atlassian evaluated various companies in the data catalog market.

"Secoda is sharing with the user the logic the agent was going through to assemble content, and that becomes really important, because it also has a place for the user to respond," she said. "So there's actually an interaction loop between the user and the agent to decide, 'Okay, for this particular object or data term, this is the right way to do it.'"

Beth Pariseau, a senior news writer for Informa TechTarget, is an award-winning veteran of IT journalism covering DevOps. Have a tip? Email her or reach out @PariseauTT.