01
Validate the job before the technology
Describe what the user is trying to complete and what currently makes it slow, expensive or unreliable. Interview users and prototype the narrowest valuable workflow.
Do not add AI because it is fashionable. Add it only where probabilistic output is suitable and where errors can be detected or corrected.
- Identify the repeated task
- Map required inputs and outputs
- Define unacceptable errors
- Prototype before full development
02
Design for interrupted connectivity
A mobile connection can change during a task. Decide what the app can cache, what needs a live request and how an interrupted action will recover.
Never show success before the server has accepted an important action. Prevent duplicate submissions and explain when data is waiting to sync.
- Preserve safe drafts
- Make sync state visible
- Handle retry without duplication
- Keep critical screens lightweight
03
Treat AI output as a proposal
AI features may summarise, classify or assist with retrieval. Show the source or relevant record where practical and allow a person to confirm high-impact actions.
Log enough information to investigate failures without exposing sensitive prompts or customer data.
- Set confidence or review rules
- Provide a correction path
- Restrict data access
- Monitor quality after release
04
Plan the product after launch
The first release should prove the core workflow. Define analytics, support and update responsibilities before publishing.
Measure successful task completion, failed attempts and retention for the workflow—not downloads alone.
- Set an owner for incidents
- Plan app-store updates
- Test supported devices
- Budget for maintenance
Sources
Primary references
Technical and regulatory points were checked against the following primary or official references.
This guide was prepared with AI-assisted research and drafting, then checked against the cited sources and reviewed for unsupported claims before publication.