AI monetization starts with good data governance

Communications infrastructure provider Zayo launched a data strategy two years ago to unlock its data so that its teams can make better decisions and create business value. But it first had to stop unfettered data access and establish rules for data democratization. Chief Data Officer David Sedlock talked to TM Forum Inform about the critical role of data governance in democratizing data and the operator’s progress in making reliable data available across the organization.  

When Sedlock joined Zayo from AT&T in January 2023, he said the company’s data was democratized entirely “but not in a good way.” Everyone could access everything, but there were no standards and “no single version of truth,” which, if left unchecked, could lead to inconsistent interpretations or even unproductive outcomes.

For its data democratization plan, Zayo started from scratch with a strategy for the architecture, platform, processes, and governance. The operator chose a hybrid centralized/decentralized model, whereby a center of data excellence makes curated data sets available to designated data stewards across the business. 

“If you don't have high-quality data in a centralized, single place … you can't have democratized data. You can't have high-quality AI or use data science or machine learning. None of this works. AI and Generative AI do not matter unless you're good at the fundamentals, which means [having] high-quality data in a single, accessible place,” he said.

Pillars of data democratization

Over the last two years, Zayo has laid the foundations for its data governance architecture. It has built a single data platform and established standard data ingestion patterns. It created data pipelines from nearly 30 system data sources that encompass most of the data it would need for almost any business decision. 

The data goes through a disciplined process within the operator’s center of excellence, working with data stewards, before it is deemed ready for access by other parts of the organization. 

One of the first steps was to identify 819 so-called “critical data elements,”, which are “the most important things that drive more than 95% of decisions in the business,” explained Sedlock.

“We're not trying to solve the world. We're going to solve what is the most important,” he said.

The operator then creates “business-ready data sets,” combining critical data elements, source data from the pipelines, and business logic rules. Driven by input from business data stewards, this generates relevant data for the business, such as a full view of customer data. 

These data sets go through further checks and controls to expose them to the business as what Sedlock calls “gold data sets.” Only certified employees, known as “data citizens” or stewards, can access this data. 

“When [the center of excellence] publishes the gold data sets, that is when we democratize the data. They have gone through our controls. We know they are highly reliable, accurate, and complete. That is when we give authorized access to our internal customers, and they can do anything they want with it,” he said. 

Data sparks business creativity 

Zayo has published more than 70 gold data sets and continues to add more regularly within its Snowflake environment. Analytics and business intelligence platform Sigma enables access to the gold data sets.

With access to curated data sets and the Sigma tools to analyze them, people can create novel data sets that Sedlock’s center of excellence might not have spotted. 

“We’re not the experts of everything, but the people in the business are. They will create more business-relevant analytics due to their hands-on business knowledge. The key is to get those gold data sets into the hands of the people running or interacting with the daily parts of the business and let creativity happen,” said Sedlock.

The availability of gold data sets also paves the way for more use of AI and GenAI. The operator is adamant about not inputting “poor quality data” into AI applications, but now it has more curated data to tap into. 

“We're not allowing an ounce of data to go into AI unless it's coming from the gold data sets. That's the rule, period,” he said.

Putting data and tools to work

Zayo has multiple suppliers for its data architecture strategy. The operator’s data platform and tools run on Microsoft Azure. Snowflake provides its cloud-based data warehouse. In addition to Sigma, it also uses Tableau and PowerBI business intelligence tools.

Another important component is dbt Labs (data build tool), which Zayo uses to store its proprietary business rules and logic in cloud-based Python libraries, making them agnostic to any tool or platform. This gives the operator flexibility to change tools and platform architecture in future without impacting the business, explained Sedlock.  

“The absence of business rule and logic abstraction is a common limitation of data and technology architecture that prevents companies from migrating to best-in-class technology platforms,” he said.

Having this flexibility is crucial, and gives Zayo an advantage, especially given that the ever-changing world of AI is at an early stage and there are “no clear-cut market leaders” yet, he added. 

For data science, machine learning, and AI, Zayo relies on Dataiku, especially for identifying opportunities for cost savings and increased revenue, Sedlock explained. The operator has also started using Fivetran to create data pipelines to move data. 

Together, these tools help support the next stage of the operator’s data journey. 

“Now that we have all the foundational capabilities built, we’re looking at how we can optimize, make things more efficient, and decrease our costs,” he said.