Product design
AI functionality: Preserving a feature a new workflow would have quietly killed
A new feature was about to block AI functionality users relied on. No one had noticed.
Context
A core promise of the platform was that users could use AI to generate a required dataset. Running AI triggered creation of the master data container. A second, essential dataset was built manually and added to the container separately. A new workflow would automate the production of this second dataset, but in doing so would create the master data container on page load, inadvertently blocking users from accessing AI. I spotted the conflict during a session with the training team.
Challenge
The system only allowed one master data container. The new feature created it on page load, which meant AI could no longer trigger its creation. Users would lose AI functionality the moment the new feature shipped. Work was already underway, so fixing this meant convincing product and engineering to pivot mid-build.
Approach
I raised the conflict with product and was told a pivot was not possible. Within hours I produced prototypes showing the problem and a resolution: decouple AI from needing to create the master data container and let it generate a subset within the one the new feature had already created. I gathered feedback from the training team, who confirmed AI was central to onboarding less technical users and a key selling point to clients. I brought prototypes and evidence to product and engineering together. Engineering confirmed the approach was feasible.
Outcome
The team pivoted within two days. Both features were preserved in production. Users kept AI functionality without losing the efficiency of auto-generation.
Reflection
Sometimes collaboration means insisting. The evidence was clear and the cost to users was real. This was a moment to spend the trust I'd built, not wait for consensus.
Selected working artefacts from the project. Blurred intentionally to protect business context, shown to demonstrate the depth and shape of the work, not as polished deliverables.