Here is an excerpt from an article written by Ganes Kesari  for the MIT Sloan Management Review. To read the complete article, check out others,  sign up for email alerts, and obtain subscription information, please click here.

Illustration Credit:  Carolyn Geason-Beissel/MIT SMR | Getty Images

* * *

When people resist changing the way they make decisions, well-intentioned data science projects are doomed. Here’s how to overcome the key challenges.

When one of the largest global telecom companies grappled with high customer attrition, the marketing team used a heuristics-driven approach to retain customers. For example, if a customer hadn’t made any outgoing calls in three weeks, the company would roll out a promotion.

However, this approach didn’t deliver results, and customer retention was at its lowest in years. Based on the weekly business performance reviews, the CEO knew it was time to try something different. He turned to data science tools and a cross-functional effort to solve the retention challenge.

The marketing team deployed machine learning algorithms to study customer usage patterns and predict churn. Simple techniques like decision trees helped spot factors such as billing amounts and outgoing call patterns — both good predictors of a customer’s propensity to leave.

Tests run on historical data indicated that this approach could improve customer retention by 39% — a hopeful sign. Then the data science team brought in more advanced AI firepower, adding techniques such as neural networks for deeper pattern-spotting.

This approach turned out to be far more accurate and effective, with the potential to improve customer retention by 66% — a considerable uplift. A four-week pilot run with high-value customers confirmed the results.

The data science solution looked ready for a full rollout. Then things turned south.

The marketing product managers refused to use the solution. They found it hard to trust an algorithm that spat out a list of at-risk customers with little explanation. What’s more, many of the data-backed recommendations were counterintuitive. For example, even some customers with long tenures and steady usage were flagged by the algorithm as high risk.

The entire process felt wrong to the managers. Despite the strong pilot results, the users gave the data science solution the cold shoulder.

If you’re wondering whether this is unusual, I can assure you that it is not. The graveyard of data science initiatives is filled with solutions that are advanced, accurate, and well-meaning yet unused.

A company’s investment in multiple data and analytics projects won’t, on its own, result in employees using data insights to make their own decisions. This end goal requires leaders to take different types of interventions. And it’s hard work: In a recent survey of Fortune 1000 CIOs and data executives by NewVantage Partners, about 6 in 10 leaders acknowledged that they haven’t been able to establish a data- and analytics-driven culture.

Why is it so hard? How can leaders foster an environment where decision-making with data insights becomes a habit? Let’s uncover the root causes of failure and examine three practical ways for leaders to nurture a data-driven culture.

Three Factors That Lead to Failure

I’ve seen common failure patterns emerge among doomed data analytics initiatives during my past decade of global consulting work.

Despite a heavy focus on analytics tools and algorithm accuracy, most initiatives struggle to deliver actionable insights. Even when the insights shared are actionable, people are often not excited about using them. Finally, projects that do see meaningful adoption often lose steam after road shows end — and then die slowly.

Chances are you’ve encountered such failure stories. However, addressing these head-on won’t yield much improvement: These are just symptoms of the problem, not the root causes.

Three core factors cause those symptoms:

Most initiatives are run as technology projects. Often, the entire journey starts as a data science initiative or carries a technical name such as predictive analytics project. This alienates business teams. Given the weak link to business users, the project may not address the most significant problems — or real challenges.

Users resist changes to ways of working. Most humans are wired to seek comfort in the familiar. This inertia can be dangerous for analytics initiatives, which almost always involve changes to business processes or decision-making methods. This change resistance is amplified by fear of AI-driven job loss, an emotion that runs high today.

It’s difficult to demonstrate analytics ROI. Despite data’s immense effectiveness, it is often tough to attribute business outcomes to analytics efforts. Since several factors could drive revenue increases or cost savings, linking them to data-driven decisions may not always be possible. Consequently, the inability to quantify ROI can push data initiatives into a downward spiral, with project leaders squandering momentum and losing funding.

* * *

Here is a direct link to the complete article.

Ganes Kesari is the cofounder and chief decision scientist at Gramener.