Real-time analytics presents challenges to unlock benefits

Lost time and money due to misaligned expectations

Costly mistakes can occur if management and employees have different ideas of what real-time means.

When there is potential for losing sales -- or losing a large customer -- the definition of real-time analytics must be well understood, said Venkat Venkataramani, CEO of real-time analytics software company Rockset.

Venkataramani offers an example of a real-time analytics rollout gone wrong. A buy-now-pay-later provider ran into problems because management thought real-time meant up-to-the-minute, while the team implementing the analytics project thought a delay of six hours was real-time enough.

The data team started out with a data warehouse, Venkataramani said. "As they were scaling, the amount of potential loss that can happen in that six-hour window was getting larger and larger. They'd already lost a lot of payments and a lot of top-line growth."

The stakeholders in the project didn't agree on requirements, so didn't choose the correct platform for implementation.

"Using the wrong tools that were not built for real-time will either give you data with a lot of latency like old data that is six hours behind, or it will just give you a very expensive solution where you don't get the ROI," Venkataramani said.

This isn't an uncommon mistake. In fact, 44% of organizations still lack an enterprise-wide data strategy, reported the AI, Data & Analytics Network survey.

Bad business decisions come from bad data

Garbage in, garbage out has long been a truism in the computer field.

When processes move slowly, there's time to catch mistakes and fix them before they get too costly. With real-time analytics, that window of opportunity to correct problems shrinks to almost nothing.

If you're not able to deal with the noise correctly and you move too fast, you can end up making bad decisions.

Companies must make sure data is clean from the start, which is not easy. Out of 500 IT and data professionals, 91% said it impacts their company's performance, according to a May 2022 survey by data pipeline management tool provider Great Expectations.

Gartner estimates the cost of bad data at $12.9 million per year for an average company. The potential losses are exacerbated in real-time environments.

"If you're not able to deal with the noise correctly and you move too fast, you can end up making bad decisions," said Sanjay Srivastava, chief digital officer at Genpact, a professional services firm.

Fixing data problems in real time is not easy, either. In fact, sometimes the data is incorrect at the source. For example, Genpact tracks a number of IoT parameters coming off heavy machinery engines and uses the data to predict maintenance requirements. That data is jittery. Data that comes from physical sensors can be uneven or erratic. Random fluctuations that are just background noise average out over time to provide meaningful information.

But in real time, there's no opportunity.

"If you look at it bit by bit, it may give you inconsistent results of when a repair is needed," Srivastava said. Organizations must be thoughtful about what the data is, what the conclusion is and if it makes sense for the context in which you're using it, he added.

Traditional data quality procedures and tools don't always work, said Roy Schulte, distinguished VP analyst at Gartner. "You don't have time for it," he said.

There are new tools emerging that can help companies deal with real-time data issues, including other software solutions and open source options like Apache Flink that can compute answers multiple times in a short window. For example, someone might want to know how a particular piece of machinery is performing at the moment.

"With Flink, you can compute the answer twice," Schulte said. "You may compute the answer right now and you can come back five minutes later and recompute the same time period." By taking into account late-arriving data, the end result is more accurate, he said.

Lost faith, lost morale, and interdepartmental conflicts

Switching to real-time analytics potentially upends the way a business operates, stressing employees and business units.

In 2023, 94% of organizations plan to increase data investments, according to a NewVantage Partners survey released in January 2023. About 80% say organizational receptivity and alignment, changes to processes, and people and skills are their greatest obstacles to getting value from data.

The first challenge to overcome when rolling out real-time analytics isn't always about data or tools. It's about the organization's mindset, said Jerod Johnson, senior technology evangelist at CData Software, a data integration software provider.

"Everyone is familiar with the practices of analyzing and finding meaning from data collected over weeks, months or even years," he said. "If your organization is analyzing data in real time, you need to have the processes in place to rapidly shift directions if what you're doing is proving ineffective."

This includes adopting new technologies and educating staff about how to use them. When rolling out real-time analytics, companies need to follow change management best practices, including regular communication, building a strong case for change that involves employees in the process, training and support for employees, and plans for how to manage resistance -- and celebrate success. But real-time analytics also require attention to the risks of moving too quickly, including effects of any potential mistakes.

"You need to have the type of atmosphere that accepts the occasional misanalysis," Johnson said. "When learning to make decisions rapidly, organizations need to be comfortable with the concept of failing early and recovering fast."