Tibco targets data model deployment with ModelOps launch

Tibco targets data model deployment with ModelOps launch

Tibco on Tuesday launched ModelOps, a tool designed to enable organizations to quickly deploy data science models at scale to any user’s workflow.

Tibco, an analytics vendor founded in 1997 and based in Palo Alto, Calif., first revealed ModelOps in preview in May 2021.

The release marks the general availability of the tool after more than a year of fine-tuning based on customer feedback.

Putting models to work

ModelOps is aimed at bridging the gap between investing in data science and benefitting from that investment by simplifying the operationalization of machine learning models, according to Mark Palmer, Tibco’s general manager and senior vice president of analytics, data science and data products.

According to a 2022 report by NewVantage Partners, more than 90% of organizations are investing in augmented intelligence and machine learning initiatives. Barely a quarter, however, say they have been able to widely deploy AI and ML models.

The big challenge is that companies are making algorithms and not using them. This is the big hurdle. They’re spending all this money, but [the models] are stuck in a lab. They’re not operationalizing models. That’s the problem.

Mark PalmerGeneral manager and SVP, Tibco

“The big challenge is that companies are making algorithms and not using them,” Palmer said. “This is the big hurdle. They’re spending all this money, but [the models] are stuck in a lab. They’re not operationalizing models. That’s the problem.”

Common barriers to the wide deployment of AI and machine learning models include difficulty applying analytics to applications, identifying and mitigating biasand managing an algorithm’s behavior in workplace applications, according to Tibco.

But perhaps the most significant barrier that leads to the gap laid out in NewVantage Partners’ report is access, Palmer said.

Model and machine learning operations — like DevOps and DataOps (data operations) — are growing trends in analytics aimed at making it simple and fast to deploy data science models. Beyond Tibco, other vendors offering model and machine learning operations tools include AWS, Cloudera and Domino Data Lab.

Tibco ModelOps, however, takes a different approach to AI and machine learning model deployment than those other toolsaccording to Palmer.

AI and machine learning models are most often developed by data scientists and IT professionals, and if they’re ever deployed that deployment usually has to be done by those same people.

Tibco designed ModelOps to be an operating system for AI, and it does not require code to move a data science model from development to deployment. That opens up access to AI and machine learning models to new personas within organizations that work with data, rather than just data scientists and IT professionals.

ModelOps enables self-service users such as business analysts, business leaders, application developers and data engineers to deploy models.

“We think everyone else gets it wrong because they focus on data scientists,” Palmer said. “We’ve expressly designed a no-code environment because we fundamentally believe the reason companies self-report that they’re not effectively deploying AI models is because they’re locked inside a chamber in a lab, and the MLOps tools that are out there are designed to facilitate and make that worse.”

Beyond targeting personas other than data scientists and IT professionals, Tibco designed ModelOps to enable easy deployment by making it format-agnostic and easy for organizations to manage and deploy model pipelines and production environments, according to Tibco.

Among the formats supported by ModelOps are API-based models in any cloud service or on-premises system. It can be used not only with other Tibco tools such as Spotfire and Data Virtualization, but also with non-Tibco BI platforms like Tableau and Microsoft Power BI.

The next step for Tibco

The addition of ModelOps to the Tibco platform, meanwhile, is a logical and necessary step for the vendor, according to David Menninger, an analyst at Ventana Research.

“If you are — or aspire to be — a platform vendor, you need to offer these capabilities,” he said. “As Tibco has acquired and built out a broader portfolio — including data science — it’s natural for them to include ModelOps.”

Model operations are critical to data science, Menninger continued. And a tool like ModelOps can not only enable self-service model deployment but also ease some of the burden on data scientists.

“If a vendor wants to truly be in the AI/ML market, they need to provide this functionality or partner with others that provide it,” he said. “There are several challenges that ModelOps attempts to address: training models, deploying models, detecting bias, detecting drift and then retraining and redeploying models as necessary. Many of these tasks are still done with scripts and manual processes.”

Like Menninger, Andy Thurai, an analyst at Constellation Research, said the development of ModelOps is a logical move by Tibco.

Model deployment is an issue for many organizations — as are a host of other management issues related to data science models, including model governance, validation, version control and testing — Thurai noted.

Tibco, meanwhile, has long offered data management and data integration capabilities that could be applied to the management and integration of AI and machine learning models.

“Given the lead position Tibco has in the data management, integration and APIs areas, this is only a natural addition,” Thurai said. “It could help a lot of Tibco customers who have been struggling with their AI model pipeline.”

He added that many organizations have either built their own systems for deploying models or gravitated toward tools developed by other vendors. For those who haven’t, ModelOps has the potential to be a significant addition.

“For existing Tibco customers who haven’t done either and are starting to dabble in ModelOps, this might be a worthy solution to look at,” Thurai said.

Roadmap

Now that ModelOps is generally available, future plans for the new capability include integrations with such Tibco platforms as Spotfire, WebFocus and Jaspersoft — the vendor’s three BI tools — so that users will be able to work with data models without having to leave their BI environment.

And similar integrations with Tibco Data Virtualization, Tibco Cloud EBX and Tibco Streaming are also planned, according to Palmer.

Spotfire 12, the next update of Tibco’s most popular analytics platform, is planned for later this year and was previewed during the vendor’s recent Analytics Foruma virtual user conference.

The integration between ModelOps and Spotfire will be included in that update. Users will be able to work in their Spotfire environment, and work with and launch data science models there more easily than they could previously.

“ModelOps will be built into Spotfire,” Palmer said. “The customer won’t even think about it. They’ll open and connect a model. They won’t even know it’s ModelOps. The product is in there, instead of the products just working well together.”

Similarly, ModelOps will be included in other Tibco products as they’re updated, he continued.

“It will be a feature of Tibco EBX, WebFocus and Streaming, but the first one will be Spotfire,” Palmer said.

LEAVE A REPLY

Please enter your comment!
Please enter your name here