(US & Canada) | Practice a Culture of Progression Over Perfection — Microsoft General Manager, Enterprise Data

Karthik Ravindran, General Manager, Enterprise Data at Microsoft, speaks with Robert Lutton, VP at Sandhill Consultants, and Editorial Board Vice Chair at CDO Magazine, in a video interview about practicing a culture of progression over perfection, the key considerations and challenges in navigating technology choices or data governance; and implementing data governance for AI and AI for data governance, combined with humans in the loop for value creation.

The culture of progression over perfection

The final segment begins with Ravindran elaborating on practicing a culture of progression over perfection and how it impacts the culture and governance practices at Microsoft.

He explains from two perspectives:

1. Applying progression over perfection and navigating organizational change management

2. Progression over perfection in the context of applied data value

Explaining the former, Ravindran states that while venturing into the task of building the data office at Microsoft, it was seen as another centralized initiative. Initially, 20% of stakeholders across the enterprise wanted to connect data across the company and operate with company-wide data. These 20% started evolving the use cases by leveraging data and insights from cross-functional areas.

The next 60% of the stakeholders were on the fence, thinking of it as centralized and not being able to work for agile business needs. By applying progression over perfection for the first time, Ravindran decided to move forward with one positive partner instead of waiting to get a 100% buy-in.

Accordingly, he wanted a partner with whom he could build a data product or a set of valuable data services that would later demonstrate value to teams across the ecosystem. Ravindran says that the first step of progression over perfection was identifying the first stakeholder, and he ended up working with the marketing organization.

The marketing professionals had interesting use cases around connecting customer data across the company to personalize marketing campaigns. It could not be done earlier due to the highly siloed nature of the data.

Then, the first version of a connected customer data product was built in collaboration with marketing stakeholders. This, when applied to the marketing scenarios, became attractive to broader functions of sales, finance, and operations and became a magnet to attract more functions into the nucleus.

From customer data, Ravindran iterated and evolved to the employee domain. Then the ESG, sustainability analytics, data capture, and reporting, including manufacturing, supply chain, finance, and legal.

While 90% of the teams are now operating in the discussed model, some teams still operate in their own ways. However, they are extremely specialized and efficient in managing their operations versus integrating into something more connected and enterprise-wide.

Progression over perfection in terms of applied data

Moving on, Ravindran discusses progression over perfection in terms of applied data. Expanding on the example of building connected customer data products, he states that it was not an easy ride to perfection.

The team started with the most pressing problem of getting clean records of truth that could be categorized as customer master data information. While there were 50 systems across the company operating on account-level data in silos, when connected, the matching rate was less than 5%.

Then, the team started with the most basic customer information — contact and account information — and built a connected data product that became a single record of truth. Incrementally, Ravindran kept adding engagement signals and transactional data, ensuring that the evolutions were connected to applied business scenarios for demonstrable value outcomes.

In the progression over perfection approach, one needs to recognize it as a journey and take failure as a learning experience, says Ravindran. He recommends getting comfortable with the unknowns, and taking action is the only way to figure things out.

It will not happen while waiting to come up with a perfect architecture or middle-of-the-mass state solution on day one. One has to get comfortable with putting water through the pipes, learning by doing, and adapting on the go.

Navigating technology choices and data governance

When asked about key considerations and challenges in navigating technology choices and data governance across the enterprise, Ravindran elaborates on three considerations:

1. Usability

2. AI

3. Cloud-first mindset

The first consideration of usability ensures understanding the personas and recognizing and appreciating their complementary nature. The personas will include business users, operational users, specific data functions of data stewards, analysts, and data product owners.

Ravindran urges organizations to think through their experiences before delivering, as it must be a collective solution catering to the persona’s needs. For instance, if a certain technology is not adaptable, its adoption would fall apart.

Natural language as the language of data governance

Next, he discusses the AI consideration, stating that AI has some of the most pronounced applications in the space of data governance and management. He believes that with AI, the natural language could be the language of data discovery, understanding, and management.

Adding on, Ravindran believes that it is a massive opportunity for AI to bring to life the notion of natural language as the language of data governance. Further, he implores the community to assess where a technological solution for data governance is headed regarding enabling democratization through natural language.

The third dimension that Ravindran offers is having a cloud-first mentality and a hybrid-capable consideration. In this age of AI, when it comes to managing, governing, and processing data, he encourages leaders to ask the question “Why not cloud?” instead of “Why cloud?”

Whether it is massive data management, data quality management, or data health computation across an enterprise-wide data estate, these are the foundations on which AI and analytics are built.

These fundamentals require a certain scale of compute, storage, and processing horsepower, and it would be a massive advantage to leverage the cloud. Ravindran argues for the need to understand that the cloud-first approach must be hybrid-capable.

According to him, the starting point for most enterprises would be a hybrid estate that includes multiple clouds, private clouds, and on-premise systems. Ravindran assures that while having an on-prem estate, organizations can integrate the cloud in meaningful ways and get the best of both worlds.

This could be through virtualization mechanisms and highly efficient one-copy mechanisms, he adds. Ravindran insists that there are numerous architectural patterns, as well as technological advances, that enable getting started with data wherever it is and utilizing the benefits of the open standards of the cloud.

AI for data management

Commenting on AI being an accelerator for responsible data value creation and the role Microsoft plays in it, Ravindran states that there are tons of innovations happening through Microsoft’s generative AI and Co-pilot capabilities. He continues that almost every product or brand that the company ships is a user-facing product that has AI infused into it.

Delving further, Ravindran touches upon the ideas of data governance for AI and AI for data governance. In terms of data governance and data management for AI, it has been established that any organization's AI is only going to be as good as the organization's data.

Therefore, setting good data in place becomes fundamental to making AI efficient. Good data management practices, such as mass data management, data quality management, data health controls, data security compliance practices, and risk management, can help AI generate value on the application side.

Speaking of AI for data governance, or data management, Ravindran notes that the biggest application opportunities for AI lie in the data management sphere. This could include tackling master data management for massive data organizations exploding with structured, unstructured, and semi-structured information, which would be challenging if just based on human workflows.

However, Ravindran affirms that Microsoft strongly believes in the notion of humans in the AI loop. He says that AI cannot replace humans but can elevate a person to focus on higher-order semantic and contextual knowledge work.

Furthermore, Ravindran imagines a time when human expertise could be harvested into AI models, where humans play the critical role of data stewardship. He emphasizes how powerful it would turn out to be when human inputs translated into signals helped make the AI models better over time.

Ravindran maintains that starting with data stewardship, it extends to data engineers, analysts, and scientists, who can both contribute to and benefit from the AI outputs. Consequently, better data products will be built, which will then transcend to data consumers, who can provide input signals in the form of feedback through the activated loop.

In conclusion, Ravindran states that every single input would make the AI model richer. This opportunity to do data governance for AI and vice versa, combined with humans in the loop, is an amazing virtual cycle for organizations to really spearhead as an industry.

CDO magazine appreciates Karthik Ravindran for sharing his invaluable insights with our global community.