Problem

Forecasting tools exist, but they’re made for data scientists.

Business users struggle with:

• unclear steps (“What do I do first?”)

• settings that feel technical and risky

• no clear explanation of what the forecast means

The product needed to enable non-technical users to upload data, generate forecasts, and trust the result, without reading documentation or knowing ML terminology.

Goals:

• Reduce friction for first-time users

• Make the forecasting process feel guided, not technical

• Show results in a way that’s explorable and trustworthy

Constraints:

• Dark theme required (existing design system)

• Must work for both simple CSV uploads and large structured enterprise data

• UI needs to scale to API + manual workflows

Primary workflows identified:

1. Upload dataset → run forecast → view output

2. Try a sample dataset before committing to own data

3. Review API usage, history, and billing as the product scales
   
   

Early Exploration:

Exploration A: Dashboard-first layout
Pros: Everything visible
Cons: Overwhelming for new users
—

—

Key Design Decisions

Decision 1: Split-pane layout

• Reason: Users needed to view input + chart + configuration without switching screens.

• Alternatives tested: modal workflows, wizard flows.

• Outcome: Split pane allowed faster iteration and fewer “where am I?” moments.

Decision 2: “Use Sample Data” CTA

• Insight: Users hesitate to upload their real dataset without seeing value.

• Result: Added a frictionless “Try With Sample” workflow.

• Impact: Reduced onboarding drop-off and made demos easier.

Decision 3: Invisible guidance through defaults

• Setting defaults for model type, confidence interval, and horizon made the product usable without ML knowledge.

• Users could edit settings later instead of being forced to decide upfront.