Defining and launching an AI productivity product from zero to one

The opportunity

The initial idea was broad: use generative AI to help users work more effectively. Before selecting features or technology, the challenge was to identify a specific recurring problem where AI could provide meaningful value rather than becoming an additional chatbot without a clear purpose.

Product discovery

• Which users experience the problem most frequently?
• What part of their workflow creates the greatest friction?
• Which actions should AI perform and which should remain under user control?
• What evidence would demonstrate that the product was genuinely useful?
• Which capabilities were essential for the first version?

Product hypothesis

We believe that helping users convert unstructured information into a clear, editable action plan will reduce time-to-completion and increase repeated use of the product.

MVP scope

• Included one clearly defined core workflow, structured AI output, editing, confirmation, fallback states, usage measurement and feedback capture.
• Postponed broad tool collections, autonomous irreversible actions, complex collaboration, extensive integrations and enterprise permissions.

AI product trade-offs

• Accuracy vs cost and latency.
• Automation vs user control.
• Personalisation vs privacy risk.
• Flexible prompts vs predictable output.
• Safety controls vs interaction friction.

Success framework

• Activation: users completing the core workflow.
• Time to first value and repeated weekly usage.
• Percentage of AI output accepted or edited.
• Task completion time, user-reported usefulness and cost per completed workflow.

What I learned

The hardest part of building an AI product is usually not connecting the model. It is defining the right level of automation, designing for uncertainty and identifying an outcome users value enough to repeat.