by Thor Olavsrud

Convergint builds data science team from the ground up

Mar 01, 2021
Data ScienceDigital TransformationIT Leadership

The installer of integrated building systems’ first-ever CIO, Bhuvana Badrinathan, launched data capabilities to predict sales revenue using automated data ingestion and feature engineering.

Bhuvana Badrinathan, CIO, Convergint
Credit: Convergint

Leveraging data, analytics, and machine learning can give companies a competitive advantage. But a successful machine learning program can seem out of reach for many companies: They may not have enough data scientists, data engineers, or other specific talent; or the data science process may be too time-consuming or require too many resources.

That’s the quandary that Convergint Technologies faced a few years ago. Leadership at the global services-based systems integrator, which designs, installs, and services integrated building systems, such as electronic security, fire alarm, and life safety systems, wanted to leverage AI for sales forecasting. The only problem? Convergint didn’t have a data science function. It didn’t even have a CIO.

That all started to change in May 2019 when the Schaumburg, Ill.-based Convergint hired Bhuvana Badrinathan to serve as its first CIO.

“When I joined, we didn’t even have a formal data team. We didn’t have data scientists at all,” Badrinathan says. “Data is absolutely something that we needed to create a robust team around and ensure that we’ve got a budget around. We’re establishing data governance and a data steering committee.”

Convergint found itself in need of a new ERP system. It also needed to support a host of customer-facing applications.

“We needed to increase our data capabilities,” she says. “When you don’t have data, or when you have a lot of data all over the place but you’re not able to actually use that data for decision-making and strategy, it’s almost like you’re blindfolded. You don’t know exactly what direction to head in.”

Greenfield data science

When Badrinathan signed on, Convergint did not have adequate resources for the upfront data analysis the company needed, and their existing process required a lot of integrations to get useful insights. As a result, the company struggled to create a scalable, repeatable process for executable models. The models the company did create often took months to build and deploy, all resulting in inaccurate forecasting and slow time to market.

“One of the first things that appealed to me about Convergint is the fact that I would be able to take on the data side of things, because that’s the differentiator,” Badrinathan says. “Getting Convergint to be a business that makes data-driven decisions is one of my passions, and simplifying our footprint is another big passion.”

For Badrinathan, a successful digital transformation, like the one she was hired to deliver at Convergint, depends on three things:

  1. A solid foundation, by which she means identifying two or three use cases that are extremely valuable to the company.
  2. Wise investments that maximize competitive advantage in the form of profits, goals, and value.
  3. Discipline, to stay relentlessly focused on the use cases and the competitive advantage they aim to deliver.

To leverage data in decision-making, everyone at Convergint needed easy access to verified and validated data. “It goes back to getting trusted data sources that are verified and validated, ensuring that we’ve got such things as Power BI dashboards that not only present data, but serve as a common source of information which people can use to make decisions,” says Badrinathan, whose team then needed to create a dependable development pipeline with trusted intake and delivery processes.

Of course, data democratization is easy to say, but can prove challenging in practice. Badrinathan has started by focusing on data movement — how to extract data from one source, transform it, and then load it into another. Adopting a master data management (MDM) approach was part of that process to ensure the data was properly cleansed and enriched. She was adamant the company prioritize hiring a data scientist to guide it through that process, though that proved a challenging task as well.

“We needed a data scientist immediately, but what I found after hiring and getting data scientists is that everything’s taking an enormous amount of time,” she says. “Once we asked a good set of questions of the data and of the business, then put it together in a POC, didn’t take a long amount of time. What took time was taking that and operationalizing it. That took forever.”

She had started with three data scientists, but that soon was reduced to one.

To help, Convergint turned to machine learning company dotData and brought in its AutoML 2.0 platform for data science automation. Badrinathan explains the platform has helped Convergint automate its entire data science process, from data ingestion to feature engineering, machine learning selection, and integration of machine learning models into production environments. Automation has helped the company’s one data scientist build more accurate forecasting models and a much more rapid clip.

“When we had a few data scientists on board and were running through this, it would take months to be able to answer a question like, “what are the trends? Why is this like this? What can we expect in future?” Now it takes days or maybe a couple of weeks,” Badrinathan says. “We now have a monthly data steering committee where 90% of the time is taken with looking at what are our business use cases? What are the priorities?”

Although the transformation still has a long way to go, Badrinathan says it has already started to pay dividends. One of the first data projects was a machine learning algorithm to help Convergint predict revenues. That algorithm and the ability to adjust the model rapidly has helped the company navigate the Black Swan event of the global COVID-19 pandemic.

“The machine doesn’t know a pandemic,” she says. “All it knows is the prior historical data that you’ve fed it. But over the months, as the pandemic was in place, trends changed. As the trends changed the machine learned and it adjusted. The first month, it didn’t understand it. It overpredicted the second month. By the time the third and fourth month came around, it started normalizing. It was able to learn from that data.”