“Hey, you got your chocolate in my peanut butter!” With this famous line, 1980’s actress Diane Franklin helped The Hershey Company boost Reese’s Peanut Butter Cups sales by portraying the invention as a lucky accident. In reality, its success is anything but accidental. The 126-year-old company’s disciplined approach to its business now relies on an advanced data analytics program.

Companies like Hershey’s have learned that the difference between success and failure is enabling people across the enterprise, from engineers to marketing executives to IT pros, to access and use meaningful data.

A recent IBM survey found that 70% of executives say their biggest challenge is a lack of visibility into everyday operations. But connecting assets and equipment to data analytics algorithms is not enough. According to research conducted by CrowdFlower, data scientists spend less than half their time actually analyzing data. The rest is the prep work, or data management: harvesting, organizing and consolidating data. This housekeeping is essential to success; the same research found that most data scientists see lack of quality data as the biggest barrier to successful artificial intelligence projects.

Image: OSISoft

Data management can be even more challenging than data science.

Consider International Paper, which conceived a “Mill of the Future” data pilot project with the goal of transitioning from analysis of historical data to predictive maintenance based on recommendations made by a machine-learning engine, which requires large volumes of historical and real-time data.

The company was faced with getting three years’ worth of data to the machine learning engine – 5,000 -10,000 data tags.

“The good news is: we got there,” said International Paper’s Rick Smith, the chemical engineer who oversees the Mill of the Future project. “Our engineers are very happy about this because they actually have to do the data pulls and we have been able to reduce their time significantly,” in combination with good old-fashioned human intelligence. Engineers use “process knowledge” to decide which data is important enough to retrieve and analyze.

Harvesting the right data is the first step but success also requires a data strategy that dictates how it will be categorized, managed and ultimately analyzed.

This is true across a range of industries.

Vermont Electric Power Company for example, knew it needed a data strategy to leverage data from its smart meters, but once it got started, real data analysis didn’t begin for almost a year. The company spent 10 months getting a handle on the data in order to categorize and analyze sensor data and weather forecasts. The project was ultimately successful enough to become a standalone energy analytics company.

Data management is at least as important as data science, if not more so.

Caterpillar for example, undertook the arduous task of connecting heavy equipment around the world, working in some places with distributors as old as Caterpillar itself. Despite disparate geographies, accessibility and conditions of its many machines, Caterpillar adopted a “no asset left behind” strategy.

Caterpillar knew that by enabling data-driven predictive maintenance it would ultimately sell fewer pieces of equipment; and if it didn’t start tracking its machines, another company might instead. But they didn’t know its data management and analytics would be successful enough to create a new revenue stream: The CatConnect now sells custom hardware and software that collects and analyzes data from any manufacturer’s equipment.

I’ve learned two key digital transformation lessons in my career.

First, you must consolidate the right mix of people from business, IT, and OT. You want the least amount of people serving the most lines of business possible. Sometimes one may be ahead of the others in terms of data management, and you can build a team there to serve other lines of business. But you need to keep OT and IT in the loop at all times. Once you know you’re getting the right data, you need your IT people to make it available and accessible, and you need your operations team to act on the data.

To properly manage operational data, companies must consolidate IoT data into an operational core, enabling real-time insights and uses. This system grants more visibility and builds a bridge between operational data to enterprise data systems, increasing data availability.

Second, you can’t ignore the skills gap when it comes to data science. Measure your first projects not just by economic returns but also by the knowledge they build in your organization. Establish “learning KPIs” in addition to business KPIs.

Once you begin to collect and analyze data, there will almost always be more potential projects than your team can successfully execute at once. Prioritize based on expected returns, but don’t forget to consider the skills employees will learn. If two projects appear to offer a similar return, choose what builds skills you can leverage going forward.

Data analysis initiatives always start with cost cutting, but they don’t always end there. New revenue can come from careful data management and analysis -– the secret sauce is planning, process and people management.

Reese’s Peanut Butter Cups, by the way, may or may not have resulted from an accidental mixing of chocolate and peanut butter. Creator H.B. Reese had 16 children, so his basement confectionary probably saw its share of experimentation. When World War II limited his access to ingredients, Reese decided to focus on just one product, and he picked peanut butter cups after studying his sales data. Long before machine learning and artificial intelligence, successful business people were turning data into dollars.

Michelle Odajima is a data science program leader at OSIsoft. With over 15 years in science, technology and using math to solve problems, she brings deep industrial knowledge of solving problems, growing data science teams, supporting operations, big data and analytics, and enterprise data strategies.