The vast amount of data available today has the potential to usher in a new era of evidence-based innovation within companies, supporting novel concepts with concrete proof. Driven by the goal of better meeting customer needs, optimizing operations, and refining strategies, businesses have invested heavily in data, technology, and analytical expertise over the past decade. There continue to be advancements in BI, for example, augmented analytics. Despite these efforts, many companies struggle to develop a robust, data-driven culture where decisions are consistently based on data.

What makes this so challenging?

Research across various industries shows that the primary barriers to becoming data-driven are not technical but cultural. While it’s relatively straightforward to incorporate data into decision-making processes, making data use a routine and automatic part of daily operations for employees is much more difficult. This shift in organizational culture typically represents a significant challenge. To address this, we have outlined 10 key principles to help establish and maintain a data-centric culture.

A data-driven culture must be established at the highest levels of management

Companies with robust data-driven cultures often have top executives who set clear expectations that decisions should be based on data, making it a standard practice rather than an exception. These leaders lead by example. For instance, at one retail bank, top executives guide product launches by reviewing data from controlled market trials. At a prominent tech company, senior leaders spend the first 30 minutes of meetings reviewing detailed summaries and supporting evidence to make informed decisions. Such practices influence the rest of the organization, as employees must adapt to these standards and communicate using the same data-driven approach to gain recognition from senior leaders. The actions of top executives can drive significant changes in company-wide practices.

Select Metrics Strategically for a Data-Driven Culture

Leaders can significantly influence behavior by carefully choosing which metrics to track and setting clear expectations for their use. For example, if a company stands to benefit from predicting competitors’ pricing strategies, it should adopt a relevant metric like predictive accuracy over time. Teams should regularly make specific forecasts about price changes and monitor the accuracy of these predictions, which should improve with experience.

Take, for instance, a major telecom provider that aimed to enhance the user experience for key customers. Initially, it only had aggregated data on network performance, which provided little insight into individual customer experiences. By developing detailed metrics on customer experience, the company could quantitatively assess the impact of network improvements. Achieving this required a more rigorous management of data sources and usage than is commonly practiced—demonstrating the importance of precise data handling.

Avoid Isolating Your Data Scientists

Data scientists can often be disconnected from business leaders due to their isolation within the organization. For analytics to be effective and valuable, it must be integrated with the broader business operations. Companies that have successfully overcome this challenge typically use two main strategies.

First, they create flexible boundaries between data scientists and the business units. For example, the AIA Group rotates staff between specialized centers and operational roles to implement and scale proof-of-concept projects, before potentially returning to their original positions. Similarly, Walmart has established new roles in various departments that have connections to analytical centers of excellence. The key is to integrate domain expertise with technical skills.

Leading companies also employ another approach by bringing the business closer to data science. They encourage employees to become familiar with coding and quantitative concepts, fostering a culture where data literacy is widespread. While senior leaders do not need to become machine-learning experts, they must understand and engage with data-driven concepts to lead effectively in a data-centric environment.

Address Basic Data Access Issues Promptly

A frequent issue reported is that employees across different departments struggle to access even fundamental data, despite numerous efforts to make data more accessible within organizations. Without sufficient information, the analysis cannot be thorough, and the data-driven culture cannot take root.

Top-performing companies tackle this problem with a straightforward approach. Rather than implementing extensive and slow data reorganization projects, they pick a few critical metrics at a time and provide universal access. For instance, a major global bank, aiming to better predict loan refinancing trends, created a standard data layer for its marketing team that focused on essential measures like loan terms, balances, property information, marketing channel data, and overall customer banking relationships. The initial data made accessible should align with key metrics on the executive agenda. This strategic focus on important metrics can significantly enhance data usage across the organization.

Measure Uncertainty in a Data-Driven Culture

While absolute certainty is unattainable, many managers still request answers from their teams without considering the level of confidence associated with those answers. This oversight can be addressed by requiring teams to quantify their uncertainty, which has three significant benefits.

First, it compels decision-makers to directly confront potential sources of uncertainty, such as the reliability of data, the adequacy of sample sizes for models, and how to account for factors without available data, like emerging competitive trends. For example, a retailer discovered that declining response rates from their marketing models were due to outdated address data. Updating the data and implementing a process for maintaining its accuracy resolved the issue.

Second, when analysts must rigorously assess uncertainty, they gain a deeper insight into their models. For instance, a U.K. insurer found that its core risk models were not adapting well to market trends. By developing an early-warning system to incorporate these trends, the insurer was able to identify cases that would have otherwise been overlooked, thereby avoiding losses from sudden claim spikes.

Finally, focusing on understanding uncertainty encourages organizations to conduct experiments. As noted by a retailer’s chief merchant, many companies’ “test and learn” approach often amounts to “tinker and hope.” At his company, quantitative analysts worked with category managers to perform statistic-based, controlled tests of their ideas before implementing widespread changes.

Keep Proofs of Concept Simple and Reliable

In analytics, there are often far more innovative ideas than practical ones. The real challenge typically emerges when trying to implement proofs of concept into production. For instance, a major insurer conducted a hackathon within the company and awarded a winning idea—an elegant enhancement to an online process—only to abandon it later due to the high costs of necessary system changes. This can be disheartening and demoralizing for organizations.

A more effective strategy is to design proofs of concept with their practical viability as a core focus. Start with a solution that is both straightforward and robust, and then gradually increase its complexity. For example, when implementing new risk models on a large distributed computing system, a data products company began by creating a basic, functional process that handled a small dataset from start to finish. Once this simple model was operational and integrated well, the company could then enhance each element progressively—scaling up data volumes, incorporating more advanced models, and improving performance.

Provide Specialized Training When Needed

Many organizations invest in extensive training programs that employees quickly forget if they do not apply the knowledge immediately. While foundational skills like coding should be part of initial training, it’s more effective to offer specialized training in analytical concepts and tools just before they are required—such as before a proof of concept. For instance, a retailer trained its support analysts in experimental design just before their first market trial. This timely training ensured that the concepts, like statistical confidence, were retained and became a regular part of the analysts’ vocabulary.

Leverage Analytics to Support Employees, Not Just Customers

Data fluency can significantly impact employee satisfaction, though it is often overlooked. Enabling employees to manage data themselves can enhance their work experience, much like the approach described in the book *Automate the Boring Stuff with Python*. If employees view learning new data skills as abstract or disconnected from their daily tasks, they are less likely to engage with it. However, if they see immediate benefits—such as saving time, reducing repetitive tasks, or accessing crucial information more easily—they are more likely to embrace these changes. For example, years ago, an analytics team at a major insurer taught themselves cloud computing basics to experiment with large datasets and new models without waiting for IT support. This hands-on experience was crucial when IT eventually updated the firm’s technical infrastructure, allowing the team to not only describe but also demonstrate a functional solution for advanced analytics requirements.

Consider Sacrificing Flexibility for Consistency—At Least Temporarily

Organizations that rely heavily on data often have various “data tribes,” each with its own preferred information sources, custom metrics, and programming languages. This fragmentation can lead to significant inefficiencies. Companies may spend excessive time reconciling slightly different versions of what should be standard metrics. Moreover, inconsistent coding practices across departments can hinder collaboration and necessitate constant retraining, complicating the internal exchange of ideas. Instead, companies should establish standard metrics and programming languages. For example, a major global bank mandated that new hires in investment banking and asset management be proficient in Python, streamlining consistency across the organization.

Regularly Explain Your Analytical Decisions

Analytical problems seldom have a single definitive solution. Data scientists often face decisions with various trade-offs. It’s beneficial to regularly discuss how teams approach problems, the alternate solutions they considered, their understanding of the trade-offs, and their reasons for choosing one method over another. This practice helps teams gain a better grasp of their approaches and encourages them to explore a broader range of options or reconsider basic assumptions. For example, a global financial services firm initially thought a conventional machine-learning model for fraud detection was too slow for production. However, they later discovered that with a few adjustments, the model could perform exceptionally fast, leading to significant improvements in fraud detection.

Organizations often stick to familiar methods because exploring alternatives seems too risky. Data can support hypotheses, allowing managers to venture into new areas with more confidence. Merely aiming to be data-driven is not enough; companies need to foster a culture where this mindset thrives. Leaders can facilitate this shift by setting examples, adopting new practices, and establishing expectations for making data-driven decisions.

Technoforte is an IT Services and Data Management company with over three decades of experience in the industry. Read more about our Managed IT services and IT Staff Augmentation services.