A financial-data assistant can produce a useful answer and still disappoint its users. An open-ended chat box invites questions the AI cannot handle; vague promises make errors feel like broken trust. User experience (UX) is the design of how people reach the AI’s value, understand its outputs, and retain control when those outputs are uncertain or wrong. It belongs alongside data and models in the solution—not after them. section 1.2 section 1.3.2
Start with decisions that shape the task, not the appearance of the screen:
- Choose a conversational, graphical, or hybrid interface that fits what users need to do. section 3.3
- Decide how much work the AI should automate and where a person should remain in control. section 3.3 section 3.4
- Guide requests toward capabilities the AI can support, and communicate limitations without inflating expectations. section 1.2 section 1.5
- Show when AI is processing information, help users judge uncertain outputs, and let them modify or reject results. section 10.4
Make a forecast usable
A demand forecast chart may look clear yet leave a supply-chain manager wondering how certain the prediction is. A shaded confidence band or percentage range can communicate uncertainty beside the prediction rather than asking the user to treat the line as a fact. The design choice turns a model output into information someone can judge. section 12.1.4
Treat UX design as an iterative process. Define a use case with a measurable task; research users’ needs, attitudes toward AI, and tolerance for errors. Test prototypes with plausible AI failures rather than only ideal responses. After launch, collect explicit and implicit feedback to find behaviors the team did not anticipate, then revise both the experience and the AI. UX research, interaction, interface, conversational, and content design may be separate roles; their work must stay aligned with technical capabilities and domain needs. section 10.4 section 12.1.4
