The standard objection to GenAI projects: we don’t have the data foundation. The reasoning sounds responsible — without clean upstream data, the model has nothing to work with. The conclusion most teams draw: wait, build the foundation, then add AI.
That conclusion gets the constraint wrong.
Implementing GenAI is easier than people think. A Streamlit app built in Python creates its own clean records as it goes. The structure the AI needs comes from the workflow the app defines — not from a pristine data warehouse upstream. Build the app, and the data layer comes with it.
The real barrier sits somewhere else.
It’s an imagination problem. The harder work is seeing what GenAI now makes possible — tasks that couldn’t be automated before because they required judgement, handled ambiguity, or took different paths. Agents change the category of what’s automatable. Most teams haven’t yet absorbed what that unlocks.
And then there’s workflow redesign. Most “AI-enabled” workflows are still shaped by what was possible without it — the AI bolted on at the end, doing what an agent could have run end-to-end. The shape of the workflow has to be rethought for the new capability, not retrofitted around it.
So the constraint isn’t the data. The constraint is leadership willing to imagine the use case, redesign the process, and let the implementation prove easier than expected.