Get Started

3 Practices to Help Build a Strong Data Culture

Data Basics, Scaling AI, Featured Shaun McGirr

Let’s be frank — creating a lasting data culture in your company isn’t going to happen overnight. No technology you install or datasets you gather will do that for you. You need time and, as we’ve seen across pop culture, it usually takes a new idea or innovation (or an old idea packaged as new) to change culture. This change usually falls on data leaders to drive because they have a unique perspective across data, technology, and the organization.

But speak to these leaders long enough and “changing culture” surfaces as a key barrier to successfully scaling AI. The following three foundational practices are distilled from my discussions with data leaders across different industries; I hope they help you spark the organizational change that will help you build a strong data culture.

strong foundation

1. Recognize how the company makes decisions and what needs to be changed.

If 10 data people are building five complex, moonshot use cases that won’t be implemented or used, someone has to change something about the process and the power has to shift. Organizations need to understand and go beyond the surface-level value/complexity trade-off from a purely data perspective, and think about their data culture holistically, observing how colleagues make decisions about changing processes. Ultimately getting a “model in production” from the data team’s perspective is just a small part of a larger change that could take months or even years. 

You also need to recognize that data people are, in their own way, a foreign culture to most organizations, coming in from the outside to help drive change. This is especially true for those just starting out on their data and AI journeys, with whom you might share very little language in common. One way to find that common ground is to avoid devoting all your time to moonshot projects that the business may not be able to adopt, due to the magnitude of follow-on process change required. You might create more long-term change by bringing a slightly new perspective to an existing problem (think a cover version of a famous pop song!) which may feel more mundane, but will be less foreign and thus easier to adopt. Over time, those many small changes add up to a shift in culture that will drive the business to ask you to attack much more ambitious problems.

2. Embed business translators and data product managers into the organization.

Many data leaders (especially CDOs I’ve spoken to recently) are embedding business translators into the organization to effectively translate business needs into data needs, with the goal of making AI pervasive and not a “special event.” A range of profiles can fit this bill, as Gartner® states, “This could be a business-savvy data scientist or citizen data scientist, an analytically minded business person or a process engineer (process modelers or business analysts focused on process design) who is mindful of business optimization opportunities derived from analytical assets.”*

By injecting these translators into the appropriate pockets of the business to identify data requirements, oversee data workstreams, and act as mediators and go-to points of contact for technical and business stakeholders, organizations will better identify true business needs from data and AI.

In addition to translators, who could well be already running a service desk where people send in requests for data, reports, and insights, data product managers are also increasingly popular. They are sharp data storytellers with a specific lens on understanding the lifecycle and development of data products. Compared to translators who discover demand in the business and reflect it back to data teams, data product owners are accountable for delivering more concrete value, co-sponsored by business and data teams.

Organizations need to develop a keen understanding of who these ambassadors are to ensure the value of data projects and products are appropriately communicated to the business. It’s important to remember, though, that data initiatives can’t exist as a stand-alone island, even with translators. To move beyond the typical relationship where the business just sends requests to the data team, both parties need to adopt a different way of thinking — which will change the culture for good over time and establish bridges in both directions. Hiring data product managers demonstrates a specific commitment to changing your data culture from casual interest (“insights are useful if they are low effort”) to proactive seeking (“I will invest effort to ensure I have the right insights”).

3. Don’t give your models names, build an ideas factory.

ideas factory

An organizational role of data practitioners is to generate and test ideas, and keeping data team members engaged and happy means taking time to try new things, especially if they are trendy and spur learning. But it’s important to understand that not all of them may be translatable to business value, or well received. When this gap between business and data team cultures is too wide, data people could spend a lot of time translating or re-translating ideas that don’t actually make sense for the business in the long run. A classic sign of this cultural gulf opening too wide is when data teams give their projects, or even specific models, cute names as a way to “humanize” early experiments. While this can garner interest from curious business stakeholders, this superficial interest might hide a lack of business value, and giving something unproven a cute name simply makes it harder to abandon later on.

Instead of trying to generate ideas and eventually models within the data team, and then having to name them cutely as a way to get attention, set up the conditions for an ideas factory with a low barrier to entry, but a higher barrier to continuation. Build the capacity within and outside the data team to garner ideas from everywhere, and for many different profiles of people to try out those ideas. When an idea both performs well, and resonates with business stakeholders on its own terms, the buy-in for further investment will flow naturally, and contribute to a stronger data culture over time as more people take ideas from initial seed to eventual harvest.

*Gartner - Use 3 MLOps Organizational Practices to Successfully Deliver Machine Learning Results, Shubhangi Vashisth, et al, 2 July 2020. GARTNER is a registered trademark and service mark of Gartner, Inc. and/or its affiliates in the U.S. and internationally and is used herein with permission. All rights reserved.

You May Also Like

Why AI Is Essential to the New Energy Model

Read More

What Is Mature MLOps? A Perspective

Read More

How to Move From Brick and Mortar to Online With Dataiku

Read More

3 Inherent Challenges to the Modern Data Stack

Read More