What exactly does it mean to get rid of data silos?

Sponsored by

Don’t go chasin’ AI trends. Applying AI to the business challenges that matter most to your organization is where the real potential lies. From productivity and workforce transformation to growth and operational efficiency, learn more with PwC.

As some CFOs scramble to implement AI across their enterprises, the question of data quality quickly comes into focus. Without accurate or consistent data, AI deployment can fall flat on its face.

Among the obstacles enterprises encounter when they try to support AI agents with existing data platforms, according to an MIT Technology Review Insights survey, entrenched data silos were the top problem for half of the 300 IT execs surveyed in February and March.

“Data could be pocketed away in departmental or other silos, with no integration layer available to connect them with business context. Legacy infrastructure is also not designed for the scale and power agents require. And data may be in outdated formats or stored in obsolete devices,” according to the MIT report.

Cleaning up internal data with the goal of desilo-ing it better positions enterprises to deploy AI capabilities, Shahzad Bashir, CEO of legal industry AI platform Morae, told CFO Brew.

Analogous data. The best way to explain clean data, Bashir said, is “if you think about the number of containers I have information in. Right now, I can see my iPad, my iPhone, my laptop, and the list goes on. Then I look at the number of sources, and it drives me crazy in the morning…In the good old days, I would look at email as my source of business communication. Today, I’ve got Teams, I’ve got Zoom, I’ve got WhatsApp, I’ve got SMS,” he said.

Having to check a variety of different channels like what Bashir described is an example of siloed data in our everyday lives; data is more useful when you can access it at ease, like through a phone app or shortcut.

To avoid abandoning data in a far-off container where it isn’t useful, Bashir said that CFOs have to learn how to “separate…the wheat from the chaff” to locate, organize, and move the siloed data (complying with appropriate information governance) to a “shared state” where it’s safe, reliable, and usable. “Therein lies the science and art of info governance, creation, storage, orderly disposition, and then reuse,” he said.

Warehouse specials. How’s a CFO to prevent data from sitting uselessly in individual silos? AI scenario planning and analysis platform Anaplan CFO Hemant Kapadia explained using his own analogy, derived from his 14 years at General Electric as a finance and FP&A manager.

GE had several different data warehouses—one each for sales, operations, and finance, for example—that weren’t able to communicate with each other independently, Kapadia said.

Some Anaplan customers, in contrast, have “connected all four of the major buying centers—call it sales, supply chain, workforce, and finance” and “have been able to make much more elevated, much more sophisticated decisions.”

“The benefits that you’re going to have really scale dramatically as you start to bring multiple data points from other functions together and optimize them in real time. That is the real power of that connected decision-making across a business,” he said.

Don’t aim for a “god agent.” This desilo-ing of AI isn’t just a good thing to have, either. “It’s absolutely essential,” according to Kapadia. But among the several different ways to skin the cat of AI deployment, he has a preferred method.

“We’ve got a series of agents that we’re building across each of the sub-functions within finance, so FP&A being one area, controllership, tax, treasury, you name it,” he said.

“I think if you try and build a god agent, you’re always going to suboptimize,” he said. “The way we like to think about it is, you’ll have agents which will have certain skills that they can do.”

“But then the more important part is being able to stitch together a number of these agents…and different organizations will take four different agents and build a persona out of those agents, if you will,” Kapadia said.

“That variance analysis agent could also be used within the controllership team to do flex analysis; it could be in treasury, looking at changes in bank accounts, etc.”