How simple data analytics can put your data to work before you are ‘ML Ready’

Data has become the new holy grail for enterprises. From young startups to decades-old giants, companies across sectors are collecting (or hoping to collect) large volumes of structured, semi-structured and unstructured information to improve their core offerings as well as to drive operational efficiencies.

The idea that comes right away is implementing machine learning, but not every organization has the plan or resources to mobile data right away.

“We live in a time where companies are just collecting data, no matter what the use case or what they’re going to do with it. And that’s exciting, but also a little nerve-wracking because the volume of data that’s being collected, and the way it’s being collected, is not necessarily always being done with a use case in mind,” Ameen Kazerouni, chief data and analytics officer at Orangetheory Fitness, said during a session at VentureBeat’s Transform 2022 conference.

Starting small

The problem makes a major roadblock to data-driven growth, but according to Kazerouni, companies do not always have to swim at the deep end and make heavy investments in AI and ML right from the word go. Instead, they can just start small with basic data practices and then accelerate. 

The executive, who previously led AI efforts at Zappos, said one of the first initiatives when dealing with massive volumes of data should be creating a standardized, shared language to discuss the information being collected. This is important to ensure that the value derived from the data means the same to every stakeholder. 

“I think a lot of CEOs, chief operating officers and CFOs with companies that have collected large volumes of data run into this issue, where everyone uses the same name for metrics, but the value is different depending on which data source they got it from. And that should almost never be the case,” he noted.

Once the shared language is ready, the next step has to be connecting with executives to identify repetitive, time-consuming processes that are being handled by domain experts who could otherwise be assisting on more pressing data matters. According to Kazerouni, these processes should be simplified or automated, which will democratize data, making it available to stakeholders for more informed decision-making. 

“As this happens, you will start seeing the benefits of your data immediately (and look at bigger problems), without having to make large technological investments upfront or going, hey, let’s find something that we can swing machine learning at and work backward from that,” the executive said.

Centralized hub and spoke approach

For best results, Kazerouni emphasized that young companies that are not technology-native should focus on a hub-and-spoke approach instead of trying to build everything in-house. They should just focus on a differentiator and use market solutions to get the piece of technology needed to get the job done.

“However, I also believe in taking the data from that vendor and bringing it in-house to a central hub or data lake, which is effectively using the data at the point of generation for the purpose that [it] was generated for. And if you need to leverage that data elsewhere or connect it to a different data asset, bring it to the centralized hub, connect the data there, and then redistribute it as needed,” he added.

Patience is key

While these methods will drive results from data without requiring heavy investment in machine learning, enterprises should note that the outcome will come in due course, not immediately.

“I would give the data leader the space and the permission to take two or even three quarters to get the foundations down. A good data leader will use those three quarters to identify a really high-value automation or analytics use case that allows for critical building blocks to get invested in along the way while providing some ROI at the end of it,” Kazerouni said, while noting that each use case will increase the velocity of results, bringing down the timeline to two, maybe even one quarter.

Watch the entire discussion on how companies can put their data to work before being ML-ready.