Over the last decade or so since the advent of the cloud, BI, and data analytics, businesses around the world have tried to navigate “Digital Transformation” with varying success. Sitting at the union of IT and business units, a solid data strategy is critical to ensure a harmonized, lockstep relationship where IT and business teams can jointly deliver the growth and business outcomes needed to stay competitive.

There are five keys to crafting a solid data strategy

1. Work back from desired business capabilities
2. Define persona journeys
3. Select a good architecture, but just get started
4. Create a prioritization framework
5. Don’t skimp on adoption, change management, and governance

1. Work back from desired business capabilities

• Start by working with the business and understanding what its fundamental needs are
• Spend significant effort to educate and envision the possibilities
• Work collaboratively to evaluate options and get the business’s input throughout the process

A lack of subject matter expertise makes it clear that the business should not be selecting the technology solutions, but there are some fundamental issues with the common IT-led/IT-only data modernization approaches in practice today.

For the most part, this approach works fine at first. However, we’ve seen numerous businesses, large and small, struggle to deliver the capabilities promised by the technology to the business where it can create value. Delivering data strategy to the business where, when, and how it needs to be is a complex problem that needs to be thoroughly evaluated before breaking ground on implementation.

Start with education. Business unit leaders are often bombarded by sellers from various data solution providers and often fail to understand the full range of options available, as well as their pros and cons. The “hodgepodge” approach of random solutions is only optimal from a convenience perspective and introduces so many issues that the ROI is greatly diminished.

Once the business leaders understand the landscape, they can provide valuable input on what really matters to them for the tech SMEs to make better design/platform choices

2. Define persona journeys

• Interview end-users and understand their experiences and workflows
• Identify the key points where data might help, and get the end-user to be specific about what they need and what they don’t (avoiding the moonshot to deliver everything)

Personas are a critical component of determining where the business value really lies. Data Strategy is used to automate or assist a human action, and only by understanding how, when, and why those decisions are being made can the right decisions be made about the technology to support them.

Building these should follow a simple process of interviewing key end-user groups and stakeholders to understand how they can do their job better, with a focus on identifying the low and high-value tasks that can be automated or augmented to drive value.

The goal is focus and detail, so being specific about what you are looking for, earnestly wanting to help, and circling back for feedback on prototypes will also go a long way when it comes to change management and adoption later on.

Understand that value in a scenario is often derived one of three ways: Automation for scale by multiplying the impact of human efforts,, or by lowering the operating costs of the infrastructure or licensing.

3. Select a good architecture, but just get started

• There are seemingly endless possible ways to build your infrastructure between cloud providers, and even a single cloud. Don’t get overwhelmed by the choices.
• The technology will continue to evolve at a rapid pace, so rather than locking into a specific tool, expect that the tools will improve and be replaced.
• In such an environment, focus on the core patterns of connecting, store, wrangle, analyze, and present.
• An agile development approach is essential to staying relevant, otherwise, services can end up outdated on day one.

The technology landscape today suffers from an abundance of choice. Even once a cloud is selected, there is still a paralyzing level of options. IaaS vs PaaS, SQL or NoSQL, Database or Databricks, and so on. Fundamentally, there are so many “good” ways to do things, it doesn’t really matter where you start as the tools will continue to change faster than implementations can keep up.

Instead, focus on the design patterns.

Let’s think of a simple example: You need a place to store the data in your new “data lake.” It doesn’t matter whether it’s Azure Storage or AWS S3 nearly as much as it matters how you design the structures and data strategy within it. You need a landing space for raw data, you need an archive at a lower cost tier, you need a surface layer for prepared and structured data, and you need a sandbox for data science. These patterns will draw from the requirements gathered from the business and personas, and working through these designs, you will limit the choices for tooling to a reasonable amount and make decision making much simpler.

4. Create a prioritization framework

• If early envisioning and education efforts are done well, the business will have dozens of potential use cases to investigate and develop personas or scenarios for. These need to be ranked by value and feasibility.
• There are many ways to prioritize, but we recommend a workshop-style approach where managers and leaders at various levels in IT and the business can form a consensus on the value and feasibility of each use case.

Prioritizing use cases can be a challenging task. There are often powerful, important voices that can dominate the conversation, while quantifying business value at the start is often a shot in the dark.

There is also a feasibility to consider. Scenarios that require advanced data capabilities and significant pre-work aren’t a sensible place to start, even if they are of high value.

   
We use these two dimensions and build a 2×2 quadrant where the high value, “low hanging fruit” scenarios shine through. For value, we ask the business to provide rough estimates of financial or strategic value, emphasis on rough, because at the start we’re just looking to set context relative to the other scenarios. A simple rating from 1-5 is often enough.

From there, we can construct an initial list and narrow it down to a few top scenarios to dive deeper on. Once the leadership agrees on their final scenarios, the data, IT, and analysis teams have a clear scope from which they can build their development roadmaps. Those roadmaps will continue to evolve, so reconvening between once a quarter and once a year will keep the team working on the highest value objectives.

5. Don’t skimp on adoption, change management, or governance

• Many of the desired business outcomes from data involve advanced analytics. Building these models is often easier than the change management required to leverage them fully
• As a part of the design and execution phases, ensure that the end-user groups are well defined and those users participate in the development. They’re the future evangelists you need to drive adoption and success.

Generally speaking, the end state for data maturity often involves some form of advanced analytics or AI. This may seem silly coming from one of the top data science consulting firms, but building Machine Learning models is the easy part — it’s getting (and keeping) them operational, and supporting the business how they’re intended, that is the bigger challenge. Not because the data science is just easy for us, but we’ve done enough POCs, pilots, and production implementations to know better than to assume that if we hand over a working model API and a nice PowerPoint, the customer can take it from there. Many projects often suffer from a lack of a clear home for the models and results, have limited change management efforts with end-users, or an inability to maintain and operate the model internally without support from our data scientists.

For the data science development team to step back and let the customer take over, a high level of operating and data maturity is necessary, so it’s important to engage with the end-user stakeholders and involve them in the process. Their fears are assuaged, they understand what the solution does and how it works.

Combine this with some modern DevOps, DataOps, or MLOps concepts, and any solutions deployed will have the support and governance structures to ensure production-grade reliability and auditable results.

Putting it all together

To wrap, a solid data strategy looks at the high-level desired outcomes and works back to the specific data, tools, approaches, and people needed to achieve them.

Quick wins are great, but they should accrete into a larger vision, ensuring everything is aligned to the enterprise transformation strategy. Keeping these 5 keys in your pocket will ensure you don’t navigate to the end of the data modernization labyrinth only to find a locked door at the end.