Scenario Comparison AI

Challenge

The whole planning suite was moving to our next-gen design system (Modern Harmony), but Qlik can't consume our component library out of the box. Every screen gets rebuilt from scratch, and user experience suffers.

• Modernize all ~40 dashboards across Demand, Supply, and S&OP.
• Improve user experience across all the modules, pass WCAG 2.1 and hold up to real usability.
• Do it all for the Gartner 2026 review, under a month.

Solution

The real bottleneck was time. Designing 40 dashboards by hand wasn't an option. I ran the full double diamond on the Scenario Comparison, then built the agentic AI workflow that audits dashboards, applies Qlik constraints and Modern Harmony, and delivers an approval-ready design I sign off before handoff.

Heuristic Audit

Before a single pixel moved, I ran a Nielsen-based heuristic evaluation across the existing Scenario Comparison.

• Chart styling varies across modules, feels like multiple products.
• KPI cards lack hierarchy, can't tell which model is better at a glance.
• Active filter state is ambiguous, and scenario selection resembles just another filter.
• Heat-map red/green logic is misread against the planner's mental model.
• Inconsistent labels and units, no in-context legend.

Competitive Analysis

I evaluated 6 BI and Planning products on five UX dimensions showed a clear split: BI tools shine on polish, planning tools on workflow. The gap — planning depth with BI polish was the opportunity, and it pointed straight at a Modern Harmony Qlik.

Usability Test

I ran 1:1, moderated, semi-structured sessions over Teams video (~45 min each) with 12 end-users. Tasks given were to filter a scenario pair, identify the better model, inspect MAPE-by-lag, drill into a heat-map cell, and share an insight.

1. KPI hierarchy collapse, filter ambiguity, tab switches lose context.
2. Heat-map red/green directionality was inverted in users' mental models on first read.
3. Users exported to build the narrative the dashboard couldn't.

SME Interview

I spoke with Subject Matter Expert; a planning consultant who works alongside planners daily, in a semi-structured, 45–60 minute session. We covered the weekly cycle review, what "a good model" means, decision rituals, workarounds, trust signals, and recurring support tickets.

• The planners run the same comparison every week, so scanning speed matters more than feature breadth.
• Exports happen because the dashboard can't support a narrative.
• Trust drops the moment styling feels inconsistent across modules.
• Active reviewers drill into errors, passive monitors only scan KPIs.
• Planners rarely stop at two models.

User Persona

The research distilled into Maya; a Demand Planner, active reviewer with 4 years of experience who runs the weekly cycle reviews and needs to pick the better model fast, defend it, and catch anomalies.

"Just tell me which one is better and let me drill in if I want to."

Defining the Problem

The problem was never the data, it was the reading of the data.

How might we make scenario comparison feel decision-ready instead of data-heavy?

Three guiding questions shaped the solution:

1. How do we surface the "better model" at first glance?
2. What builds trust: consistent styling, colour-with-meaning, predictable filter state?
3. How do we let users pivot to more scenarios without losing context?

Design Principles

I translated Modern Harmony into five anchors that guided every decision:

1. Hierarchy before density - one clear answer first, detail is earned
2. Colour carries meaning - every hue is a signal, never decoration
3. One primary action per region - drill, don't dump
4. System over novelty - Modern Harmony tokens beat per-screen creativity
5. Decision-first, detail-on-demand - surface the verdict, let curiosity reach the rest

How / Wow / Now

Ideas were triaged by ease × impact to focus the redesign on the highest-value moves first.

Journey Map

I mapped Maya's weekly scenario review on the existing dashboard to see where the experience fell apart. Confidence drops at filter state, KPI hierarchy, and the moment Excel takes over.

Agent workflow & setup

Here's where the story turns. Scenario Comparison was my proof, it went through the full double diamond hands-on. The AI setup I built lets the other 39 dashboards get the same second diamond; define, ideate, prototype, validate, powered by each agent's knowledge and skills, with the designer still in the approval loop.

The whole workspace lives on GitHub; agents, skills, a knowledge base, and an iteration folder all versioned and shareable with the team.

Agentic Automation

Instead of one giant prompt, I split the work into specialized AI agents, each with a clear job and its own knowledge base. The designer stays in the loop, every output comes back to me for review before it touches a developer.

Multi-Agent Design

The workflow runs the second diamond on autopilot: the orchestrator delegates, the designer handles discovery and Modern Harmony, the developer builds, looping until zero issues remain, then routing through 3 approval gates.

Qlik-expert agent

The Qlik-expert agent is the reality check. It validates every design for feasibility in Qlik Sense/Cloud: limitations, visualization types, layout constraints backed by a knowledge base.

Designer agent

The designer agent is the quality gate. It runs every output through the Modern Harmony and dashboard-design skills, loaded with Design System library, design heuristics, patterns, and WCAG 2.1 guidelines; my design brain, cloned.

Developer agent

The developer agent owns the handoff. It translates approved designs into buildable Qlik prototypes, armed with Qlik's limitations, visualization types knowledge, and a versioned prototypes/iterations folder; no surprises at handoff.

AI generate design

With the workflow in place, Scenario Comparison was the proving ground; if it could fix the hardest dashboard, the assumption was it could fix anything.

The orchestrator starts with discovery: asking the planner how often they'll use the dashboard, what decisions it supports, how many scenarios exist... The answers become the design brief.

From there it produced a 3-tab design spec an at-a-glance Overview, a Trending analysis tab, and a drill-down Detail pivot with every component mapped to its Qlik type.

Modern Harmony tokens were applied directly; purple for Scenario A, teal for Scenario B, semantic green/red, a clear type hierarchy, and a 16px spacing grid and the v1 prototype went live.