MIT SMR Connections
MIT SMR Connections is the custom content creation unit within MIT Sloan Management Review.
In this Q&A, Ishit Vachhrajani, enterprise strategist at Amazon Web Services (AWS), describes the cultural and organizational changes that enterprises must make to become genuinely data-driven. He discusses how to establish key leadership roles, address governance, overcome resistance, and generate widespread excitement about and support for AI issues.
This conversation has been edited for clarity, length, and editorial style.
Q: What does it mean to be data-driven, beyond just the technology and process considerations?
Vachhrajani: Being a data-driven organization means culturally treating data as a strategic asset and then building capabilities to put that asset to use not just for big decisions but also for everyday action on the front line. Organizations that are successful in this way focus on capturing, cleaning, and curating meaningful data, making that data available widely across the organization, having the skill set and proficiencies to convert that data into actionable insights, and then having the culture that allows people to take action on those insights.
Q: What kind of top leadership is important for building a data-driven culture?
Vachhrajani: Having a single-threaded leader helps a lot, especially when you are starting out. The way we define a single-threaded leader at Amazon is someone who wakes up and goes to sleep thinking about just this one thing. This could be the chief data officer; it could be the chief analytics officer; it could be any other senior leader.
I often advise customers to pick someone who is senior but also well respected within the business, because this change cannot simply be driven by authority. It has to be driven by influence. Having someone who is well respected within the business, but is also senior enough, really helps drive these initiatives forward. You may then decide to decentralize and make sure that these capabilities are actually built in the rest of the organization. The idea of the single-threaded leader is not to build an empire of data, but to enable and bring the rest of the company together.
Q: How about data leaders? What role should they play in creating and setting priorities for this type of culture?
Vachhrajani: Job one for data leaders is really driving education around data. What that means is: Demystify it. Make it relatable so that it is something that the whole organization can experiment with, touch, feel, and use on a daily basis, and then directly tie it to business outcome. Next is building the capabilities to allow data leaders to do that — the technology, the tools, and the right governance. Third is removing the impediments, working with other internal stakeholders, to make sure that information and data flow freely within the company.
Also, dealing with resistance is a big part of becoming data-driven. Resistance is real because while data is very, very empowering, it can also evoke strong emotion. A lot of times, people use data to justify what has already happened or been decided upon, rather than to actually guide them. There’s also a bit of a fear about losing the control of the narrative, the story, when it comes to democratizing data.
One big way to address that is by communicating the intent very clearly and strongly. This is similar to driving a lot of other types of changes, not just with data: You announce the intent from the top, then show action that ties to that intent.
This also brings up the issue of access. There are valid reasons why some people shouldn’t have access to data: compliance, privacy, and so on. But becoming data-driven also involves changing those gatekeeping functions to be more enabling, to treat data not as a departmental property, but truly as an organizational asset. In other words: Departments don’t own the data; they steward the data.
So how do you really change that gatekeeping role? The job of, let’s say, a finance or marketing or sales or operations or production person shouldn’t be that they own data and they decide who gets access to it. Instead, because they deeply understand that data, they take on the role of educators. They take on the role of publishing, cleaning, and curating meaningful data, and then making it easier for everybody else to consume.
Q: What do companies need to know about AI governance?
Vachhrajani: Governance is a very important topic. The way I think about governance is: Define your clear tenets, and your governance framework will follow. Before we talk about controls, what are the underlying guiding principles and tenets? This can be as simple as “Our goal is to enable more access to data, not to restrict it.”
It’s also important to ensure that data lineage is factored in from the get-go so that you aren’t implementing remedial controls that are reactive in nature when regulations change or when you have access issues. You are factoring in that data lineage all the way from where data originates to how it is used and consumed, and mapping that process out. You’re implementing privacy, security, and compliance considerations at every step of that process, by design.
Another issue is having a mechanism for data validation. I’m not just talking about technical validation; I’m also talking about business validation of data. That’s because often data may be mathematically correct, but when you apply the business logic, it doesn’t make sense. So it’s important to automate and integrate those aspects of validation in the governance process so that when the data is published, it is stamped not just from a basic quality check but also from a business validation standpoint for the consumer.
Finally, from a technical aspect, we always advise customers to encrypt everything. Encrypt all your data at rest. Encrypt data in motion. These are simple things, but putting them down as defining tenets can provide teams, especially smaller autonomous teams, a framework within which they can operate.
Q: What about some tactical actions that companies can take to build interest and support for these advanced capabilities throughout their organizations?
Vachhrajani: I think it comes down to the fact that most organizations, and most people, learn by doing. So I recommend picking one or two high-impact use cases, where there is a strong hypothesis that can be built, where you think data and analytics will actually make a difference in achieving a business outcome. Then run those use cases through this new approach that you’re taking, and communicate about that broadly. Even if those use cases may not be applicable to the whole organization, it’s important to communicate about them widely to build excitement, to build momentum.
We hear a lot about data literacy. I actually prefer the term data proficiency. That’s because it’s not just about being aware of the data; it’s about how to create a higher level of understanding and processes within the organization to put the data to use. Companies can build their data proficiency from the ground up by using creative ways to find education options.
For example, I’ve tried going beyond training and certification programs to offer data hackathons, where we would take publicly available data sets that everyone can relate to in their day-to-day lives, such as parking tickets, traffic violations, census data, and restaurant bookings. Then we bring in nontechnical folks from the front line to play and model with that data. We’ll provide them help and resources, then give them fun problems to solve. For instance: “Find me a restaurant in Manhattan on a Friday night where I have the lowest risk of getting a parking ticket.” This allows them to use data from everyday life to model multiple different scenarios, build visualization, build models, and combine data sets, which can then ultimately translate into actions within the business.
Q: Is there anything else that you’d like to add?
Vachhrajani: Talking about this brought one more point to the fore. We often talk about consuming data. When we’re thinking about becoming data-driven, most of the discussion is around finding data, bringing in more data, running analytics, things like that. It’s also important to have a very strong product and application strategy that ties into capturing data, because, especially in business, a lot of transactions still happen on the back of the napkins, in spreadsheets, from people talking to people.
One thing I do often is just observe the flow of information in the organization. Who is asking the question, who is providing the information, and who is making the decision? If you look at those three aspects, there are a lot of opportunities, before you even start to figure out how to consume that data, for bringing those lost insights into the workflow. It’s not just data as a byproduct of everything else. You should treat data as a product, something that you manage end to end throughout the life cycle.
Another aspect goes back to the resistance issue. Often, when companies start on this journey, the option to use the old way of doing things is kept open. If you do that, inertia will always pull you back to that old way. If you want to drive change, you’ve got to remove the option to go back. Of course, you have to do it in a mindful, intentional fashion, making sure things are working and running well. But ultimately, to succeed, you have to eliminate the options for going back and doing things the way you’ve always done them.