First presented at the AI-Ready Data Summit, this talk tackled the part of AI-ready data that tooling alone can’t solve: getting busy people to actually adopt it.
AI-ready data is often framed as a technology challenge, but that framing misses the point. The real barrier often isn’t the tooling; it’s whether ML practitioners actually change how they work. A strong platform or a strong architecture only matters if teams adopt it, and data scientists and ML engineers can sometimes be resistant to changes that platform teams introduce.
That shift typically doesn’t happen through top-down mandates or “build it and they will come” approaches. It happens when organizations start with their users, understand what matters to them, and make the value of change obvious through practical, visible wins. Only then can AI-ready data move beyond being an initiative and become a natural, embedded part of how work gets done.
This article is about the people side of AI-ready data: what blocks adoption, what motivates it, and what organizations can do to move from good intentions to real usage.
The Main Barriers to Adoption
One of the biggest barriers is awareness. This sounds obvious, but it shows up all the time. Teams are often introduced to a new capability months after it could have helped them. It is surprisingly common to hear:
“We didn’t even know this existed. We wish someone had told us earlier.”
Another major barrier is existing habits are hard to break. Teams have existing processes, and individuals have tools and approaches that work for them. Even if those processes and tools are imperfect, they are familiar. If people feel that the current approach works well enough, they may not see a strong reason to change. A better method on paper is not always enough to drive a shift in behavior.
There is also a more structural problem: AI-ready data touches many teams, and those teams are measured differently. One may care about speed, another about compliance, another about cost, and another about model quality. Also, sometimes the transition cost, and the ROI of adopting new data tools aren’t distributed equally. One team might be asked to change their workflow to benefit another. That makes adoption harder, because what looks like progress for one team, can feel like extra work for another.
Best Practices for Driving AI-Ready Data Adoption in Your Organization
Start With User Understanding
To drive adoption, we need to understand the people being asked to adopt the change. In many organizations, that means data scientists and machine learning engineers.
These are highly skilled users, but they are also a uniquely demanding audience. They often have a great deal of autonomy. Their work is not always well understood by the rest of the organization, which means they are frequently given broad discretion over how they operate and which tools they use. That autonomy can be helpful, but it also means they are less responsive to top-down mandates.
The key point is that these users are usually not resisting the process for its own sake. More often, they are trying to do their jobs in environments where their needs are not fully supported. When governance feels like friction rather than support, they work around it.
So if we want adoption, we need to understand the world they actually live in.
What a Data Practitioner’s Day-to-Day Looks Like
A typical data scientist or ML engineer spends a lot of time context-switching. They may launch a training job for one project, then move immediately into data preparation or feature work for another while that job runs in the background. Their day is often split across several parallel tasks, each moving at a different pace.
They also operate in an environment where the rest of the business does not always understand the constraints of their work. Other teams may ask for things that seem simple on the surface but are unrealistic in practice. That mismatch creates pressure and makes people less patient with new systems or new expectations.
On top of that, their tooling is often highly fragmented. AI has been evolving so quickly that very few teams work in a stable, uniform stack. Models change, frameworks change, infrastructure changes, and working norms change. Many of these users are accustomed to navigating a messy, fast-moving toolbox rather than a standardized one.
There’s a cultural dimension, too. Some of these practitioners come from research backgrounds rather than software engineering backgrounds. That means they may not have been trained early on in the discipline of governance, repeatability, versioning, and process hygiene in the same way software engineers often are. If we ignore that, we misread the adoption problem.
The first best practice is straightforward but critical: talk to end-users directly.
Don’t rely on conversations with senior leadership to understand what practitioner teams are doing. You need to understand how practitioners actually work:
- What slows them down?
- What workarounds do they already use?
- What are they measured on?
- Where does friction show up?
And, most importantly, what would make their daily work meaningfully easier? Adoption only happens when people can see what’s in it for them.
Create Easy Wins
Once that understanding exists, the next step is to create easy wins.
The most effective organizations don’t launch adoption by announcing a platform and sharing documentation. They start by showing specific user groups how the new approach improves their work.
For one team, the easy win might be faster access to production-scale data without the old provisioning delays. For another, it might be the ability to create safe experimental environments without copying huge volumes of data. For a third team, it might be better debugging and recovery when a fragile data pipeline breaks. Or it might be reproducibility, so that experiments, training data, and outcomes are no longer lost when a teammate leaves or a compliance question arises.
The important thing is not to tell people that the new system is better, but to show them. The most effective adoption efforts are hands-on. They walk users through a real scenario, let them try the new workflow, and make the value visible. This is the pattern we see work with lakeFS, the control plane for AI-ready data: the easy wins (zero-copy branches for safe experiments, reproducible training runs, instant rollback when a pipeline breaks) are also the governed ones.
Build a Clear “What’s In It for Me?” Story
Every successful AI adoption effort needs a credible answer to one question:
Why should this person spend time changing the way they work?
That answer will not be the same for everyone.
For some users, the message is speed. For some, it is simplicity. For some, it is lower operational pain. For some, it’s reproducibility or compliance readiness. What matters is that the value proposition matches the user’s pain point when asked to change.
Organizations that do this well often build an internal menu of use cases. Not a generic sales pitch, but a practical set of examples that show how different teams benefit in different ways. That makes adoption much easier, because people don’t have to infer the value for themselves.
Then Scale Across Teams
Once there are early wins at the individual level, the challenge becomes scaling adoption across teams.
This is where context becomes essential. Most people understand their own job and their own team’s goals fairly well. What they often don’t understand is how their work affects other teams, or how their team contributes to broader organizational objectives such as faster innovation, lower costs, stronger compliance, or quicker delivery of models to production.
That missing context creates friction. If a team is asked to change a workflow but does not understand why the change matters beyond its own boundary, the request feels abstract and unnecessary.
So part of scaling adoption is helping each team see where it fits in the end-to-end lifecycle. Teams need visibility into how their work contributes to the larger system and where poor handoffs or poor practices create downstream problems.
Make the Right Thing the Easy Thing
A strong principle here is to make the right thing the easy thing.
Instead of trying to force teams into strict mandates, successful organizations often create what some call a paved road. That means templates, starter kits, sample workflows, code examples, and default patterns that make the governed and scalable approach faster and easier than the ungoverned one.
This is much more effective than trying to police every exception. If people can move quickly by staying on the paved road, most of them will. If going off-road is possible but harder, then teams will only do it when they have a real reason.
That is a much healthier adoption model than trying to control everything from the top.
Use a Pilot to Get Started
One of the most practical ways to start is with an end-to-end pilot.
Instead of trying to transform everything at once, choose one model or one project at the right stage and walk the full lifecycle. Look at data acquisition, transformation, annotation, development, evaluation, and deployment. Introduce the new practices there, with real users and real work.
This has two big benefits. First, it gives the organization a way to test the approach, refine the playbook, and learn what actually helps users. Second, it creates a measurable example of success that can be used to explain the value to other teams.
A pilot is not just a technical test. It’s also a communication asset. It gives you a story you can retell internally: what changed, who benefited, what improved, and why it matters.
Final Takeaways
The core lesson is that AI-ready data is fundamentally a behavior change problem, not a tooling problem. Driving adoption requires a deep understanding of individual users, especially those closest to the work: what slows them down, what they value, and what would make change genuinely worthwhile. From there, the focus should be on creating visible, practical wins that clearly demonstrate value.
Once those early successes are in place, teams need to be connected to the broader lifecycle so they can see how their contributions fit into the larger goal. The organizations that succeed aren’t necessarily the ones with the most ambitious rollout plans, but those that make change tangible, useful, and easy to adopt.
AI-ready data only becomes valuable when people actually use it. That makes adoption the central challenge, not a side concern. The path forward: start with users, deliver real wins, make the governed road the easy road, and scale with shared context across teams. That is how AI-ready data stops being an initiative and becomes how the organization operates.
