I began with sessions and stakeholder interviews (PMs, engineers, data scientists, client leads). We mapped workflows, pain points and expectations.

Key findings:

• Model outputs were massive. Users struggled to digest raw data.
• Clients want flexibility but not infinite complexity.
• Some features were repeated in bespoke forms across clients.
• Any architecture change needs to support future expansion.

We built role-based personas for forecasters, risk analysts, and operations.

The goal was alignment across shared flows. Each role touched different tools and insights, then handed outcomes to the next step in the journey.

We built personas based on different roles (forecasters, risk analysts, operations and more) to keep us aligned while solving interconnected flows on which multiple participants can iterate on one specific flow, each one of them might use a different tool, insight or have a completely different goal that might be required by any other user later on their journey.

The map exposed approval bottlenecks, repeated manual edits, and risky handoffs. It also showed where an adjusted forecast should become a portable unit. InnoPods solved that handoff.

Interactive prototypes were created to validate the new modular workflows and information architecture. We tested them with stakeholders and end users to check if InnoPods moved smoothly across tools and if tasks like creating, adjusting, and approving forecasts were clear.

Feedback from these sessions drove rapid iteration. Each cycle helped refine interactions, edge cases, and system logic before development started. The prototypes became both a validation tool and a shared reference point for design, product, and engineering.

To validate the visualization workflow, we ran an A/B test comparing two layouts for configuring forecasts:

• Variant A: a dense, expert-oriented pattern where all configuration options are visible and grouped by detail.
• Variant B: a progressive, guided pattern where configuration is simplified into drag-and-drop, lowering cognitive load.

The test wasn’t only about visual placement, it was about how users interact with complex data configuration. One pattern favored speed and control, the other helped clarity and accessibility.

We measured:

• Time to set up a new visualization.
• Error rate when selecting dimensions.
• User preference for clarity vs. speed.

Findings:

The test compared two different configuration patterns. Variant A gave expert users faster access to all options at once, while Variant B simplified the workflow into drag-and-drop steps, reducing cognitive load. Variant B proved easier for new users, cutting errors by 20%.

Decision:

We adopted Variant B as the default pattern to prioritise clarity and adoption. To balance expert needs, we added advanced toggle settings to the backlog for later release.