The shine is wearing off a little bit when talking to business leaders about AI. The realization is fully settling in that differentiation is in the data. With the barrier to entry for accessing large language models (LLMs) rapidly approaching zero, the early competitive advantages are starting to diminish.

The old mission was “We must use AI.”

And the new mission is “We must use AI accurately and effectively.”

💡 Accurate and effective AI end up having a lot to do with the organization's data strategy.

Data strategy is a fancy term that emerged shortly after the proliferation of extremely affordable data storage. With the storage constraint lifted, companies could retain all kinds of data, in all kinds of formats, and for all kinds of reasons. More data is great, but to be accurate (actually representative of the business processes creating it), and effective (able to be transformed, curated, or otherwise utilized to advance core business objectives), companies needed to develop long-term strategies.

Like:

• Identifying important data

• Developing ways to capture that data

• Storing that data

• Transforming that data

• Operationalizing that data, and

• Governing that data

If it sounds like chores, that’s because it kind of is. The trouble today is that many companies haven’t done this foundational work. They’re trying to leap to level 5 — AI — when organizationally they’re still on level 1.

The Four Horsemen Re-Emerge

Data strategy — and everything that makes it challenging — will likely return to the forefront of the discussion among business leaders. Specifically, we will return to contending with

1. Process

2. People

3. Tech Stack

4. Culture

How to ride like a cowboy 🤠

1 Process

Do you have data capture at critical points? Are your workflows for collecting, transforming and analyzing (ETL) data effective?

Example: Your company measures engagement with your custom app by counting user logins. The product team recently implemented persistent login—a feature upgrade that keeps users signed in, eliminating the need to log in every time. But the change wasn’t communicated to the data team, and now executives are wondering why daily login counts are suddenly plummeting in the reports.

Solutions:

• 🛤️ Release trains or structured deployment processes: Product changes follow a predictable cadence, with checkpoints to flag potential data impacts.

• 🤝 Tight collaboration with the data team: The data team is looped into product planning and release cycles, not just left reacting to surprises.

• 🔍 Awareness of downstream effects: Teams are trained to think beyond their own domain—understanding that a product change can affect reporting, analytics, alerts, and even executive dashboards.

• 🧭 Change tracking & documentation: All changes are documented in a centralized place (like a change log or product release notes) accessible to analytics and ops teams.

2 People

Do you have the right team (skill + will + capacity) to execute a given strategy?

Example: Your company has a single data analyst that exclusively uses Power BI. They're fabulous at what they do, but as the lone data viz person, their capacity is limited. On top of that, they aren't comfortable "being a beginner," which makes it hard to learn new skills.

Solutions: 

• 💡 Re-frame and neutralize : Ensure psychological safety for your analyst by re-framing why you're bringing on more sophisticated tools, techniques, & team members. Emphasize that this maturation is a testament to their great work, not an indictment.

• 🦸 Position everyone for their "highest and best" use of time: Be strategic and vocal about each person's unique role and expected contributions.

• ✏️ Craft high-signal skill assessments: "High-signal" assessments are those that give the hiring manager a view into how a candidate approaches problem solving. Especially today, the tools are less important than the approach. You can learn tools, but learning to think is much harder.

3 Tech Stack

Are the tools efficient, scalable, and well integrated into the strategy?

Example: Your company's first low-code analytics tool has grown to contain a lot of legacy business logic. The new data engineer proposes an open source, version-controllable, ETL tool. It's a pain to migrate, but the low-code tool is slow and difficult for new hires to understand.

Solutions:

• 🌱 Embrace the growth mindset : Tools and technologies change. Think of tool migrations like "spring cleaning" - a fresh shot at...

• ✅ Audit, document, plan, and track: Take a close look at the old tools. What needs changed, deleted, or documented? Make a plan and track progress with milestones. Don’t forget to celebrate.

• 👯 Duplicate and cut over: For high-value data assets, ensure compatibility before deprecating the old version and promoting the new version.

4 Culture

Does the organization - from leadership to the front lines - value data when making decisions?

Example: A VP wants to run an experiment on a new UI when the customers log in. The analysts put together an experiment and a tracking dashboard. One week into the test, the VP declares the new version as better than the old, and tells the product team to switch over. The dashboard gets viewed 8 times in total.

Solutions:

• 🤝 Use data with intention: Many leaders fall into the trap of tracking data points that don't matter. When this happens, the data, dashboards, and experiments lack intention. Leaders must differentiate between when they want to take direction from the data, versus when they want data to support a decision they've already made (the case in this example). Either approach is acceptable, as long as the leader's words are consistent with their actions.

• 🧪 Some experiments fail, and that's a good thing : Leaders that embrace failed and successful experiments as "wins," create cultures that truly value data-driven decision making.   

Key takeaways

1. Great AI implementations start with great data foundations. Tackling each of the Four Horsemen sets you, your team, and your company up for long-term, sustained success with data (with and without AI).

2. Move swiftly, but not anxiously. This is a marathon, not a sprint. Starting, experimenting, learning, and adapting, is key.