Sergey Nivens - Fotolia

Becoming a data-driven enterprise requires art, science

Business analytics strategy often seeks to squeeze as much subjectivity from the process as possible. But when data is limited, human-algorithm pairings can perform well.

Most business leaders know by now that a data-driven enterprise tends to outperform non-data-driven organizations. But being data-driven means more than just using analytics.

"We have sort of an art and science approach," said Anil Goyal, senior vice president of operations at Black Book, a Gainesville, Ga., company that helps auto dealers, insurers and others involved in the vehicle market to place values on vehicles, whether new or used.

The Black Book team has been busy in recent months. Following the devastating hurricanes Harvey and Irma, somewhere between 500,000 and one million vehicles were damaged, many of which were totaled, according to Black Book. It helped insurance companies evaluate some of the damage.

To assess a vehicle's value, Black Book uses a mix of data platforms and analytics software, including data management systems like SQL Server, Hadoop and Qubole, and analysis tools such as R and Tableau. The organization's analytics team uses the various technologies to develop algorithms and analyze data coming from auctions, wholesale transactions and retail listings.

But it's not all about the data. Once the algorithms produce a valuation, a group of analysts reviews it and tweaks it as needed to make sure it reflects present market realities. Goyal said this fusion of machine and intuition is necessary for producing accurate valuations, due to limitations in available data.

"You have to have checks and balances," he said. "If we had access to data to the detailed level, and it was very transparent and it said exactly how damaged a car was, we would be very confident in spitting out a number through an analytics process alone."

People's role in a data-driven enterprise

Many businesses today assume that being a data-driven enterprise means basing decisions entirely on data. This can remove people's biases from the decision-making process and make strategies more objective.

Claudia Imhoff, president of consultancy Intelligent Solutions, offers insights into the needs of a data-driven enterprise.

But this assumes perfect data and a deep knowledge of markets and business processes that can be transposed into analytical algorithms. This isn't always possible for Black Book, which is why Goyal said it's important to have people reviewing valuations.

For example, after Hurricane Harvey hit Texas in August, purchases of pickup trucks were lower than might be expected for an area where pickups are generally popular. If valuations were based solely on data from sales and auctions, an algorithm might have concluded that pickups weren't that popular and give them a relatively low value.

However, the team knew that sales were low simply because of low supply. Low supply and high demand should actually produce high prices, but Black Book didn't have digital data it could feed into models reflecting this fact. It took the team's knowledge to work this into their valuations.

Jared Kalfus, Black Book's senior vice president of sales, said the group can accomplish a lot with data only. As a data-driven enterprise, the company has invested heavily in analytics, and its tools and processes get Black Book very close to accurate valuations. But adding in the group members' own human knowledge is the secret sauce that helps them stand out.

"We aggregate all that data and have a really strong combination of art and science to produce our valuations," he said. 

Next Steps

Everything you need to know about developing a data-driven culture

How American Express does data-driven decisions

Data-driven organizations are more productive, profitable

Dig Deeper on Customer analytics