Why the Marzetti Company’s master data project is paying off in the AI era

Being in the right place at the right time can feel like you’ve struck gold. That’s how Liam Durbin, CIO at the Marzetti Company, must have felt when its initiative to clean corporate data happened to wrap up just as the AI boom started.

Durbin told IT Brew that the Westerville, Ohio-headquartered company, best known for its dips and salad dressings, set off to transform its enterprise resource planning (ERP) software shortly after he joined in 2018. The previous version of its platform had siloed processes across its 16 manufacturing facilities.

“You couldn’t take an employee from one location and put them in another location expecting them to be able to thrive,” Durbin said. “It was just very, very kludgy.”

The next four years were busy ones for the company, which concurrently embarked on a project to unify its data into a master dataset, because a “well-run ERP” requires clean data. As part of the process, Marzetti established a set of common terms, measurements, and definitions across all of its data.

“We want to get the data layer and the master data cleanse because it was necessary to do the ERP project. You can’t have sugar measured one way at one facility and measure sugar a different way at a different facility,” Durbin said. “You want to have all your financials look at sugar the same way.”

Those efforts unexpectedly paid off for the company when the AI boom hit and data became increasingly valuable for analyzing, training models, and more. “Now, as AI has exploded, it has given us the opportunity to sort of leapfrog our competition,” Durbin said.

Clean slate. Durbin said master data is an area of focus that businesses tend to “neglect” because it can be hard to fund data-related initiatives.

“Master data just isn’t a very glamorous project to take on because when you’re done, you kind of have the same thing you had before, but it’s just a little bit better,” Durbin said, adding that it’s hard to explain to executives the value of having clean data.

However, clean data is invaluable for AI-related projects. Earlier this year, an MIT study found 95% of AI projects earn no return on investment. During a Dec. 8 virtual press briefing, Informatica Chief Product Officer Krish Vitaldevara said this is because data supporting models are often fragmented or of poor quality. (Informatica was recently acquired by Salesforce.)

“Different systems store different fragments of meaning, each with its own language, its own rules, and its own truth,” Vitaldevara said. “AI cannot see the full picture because context is scattered, stale, or inconsistent.”

He added that the “biggest bottleneck” businesses encounter when transitioning into agentic enterprises is this lack of “trusted context.”

Clean living. With its freshly unified data environment, Durbin said Marzetti is now well-positioned to use AI for different use cases and has already begun experimenting with the technology. About a month ago, the company rolled out accounts payable automation using optical character recognition, which is technology that changes images of text into machine-readable text. While the company does have more AI projects in the pipeline, Durbin said the company is taking a “patient” approach.

“Instead of it being a hammer searching for a nail, we’re looking at our existing demand and saying, ‘How does AI help us with this? How does AI help us solve this business problem?’” he said.

For companies looking to undertake their own master data project, Durbin recommends they fully fund and staff teams in charge of it and “have a separate arm of the systems integrator dedicated to master data.”

He added: “For a lot of folks, this probably will come across as common sense, but it is an easy area to try and cut corners.”