ARM Dashboard
PRODUCT DESIGN
Designed an AI-integrated campaign hub that reduces Time on Task by surfacing the right insight at the right moment.
DOMAIN
Ad-tech
ROLE
Product designer
TIMELINE
5 weeks
Product Context
Advertising Resource Management (ARM) platforms exist to give brand-side media managers control over their campaign portfolios. The market includes tools like Mediaocean, Prisma, and Allocadia — but all of them share the same fundamental flaw: they surface data, not decisions. Meridian is a concept ARM dashboard designed to close that gap — built around a single premise: a media manager should be able to assess their entire portfolio health in under 60 seconds, and act on an anomaly without leaving the platform.
The Problem
Media managers make expensive decisions with delayed, fragmented data. Brand-side media managers typically discover campaign problems from weekly agency reports — 3 to 5 days after the issue started. By then, budget has already been wasted.
My Role & Team
End-to-end pfeaturedesign — from research to production-ready system.
THE APPROACH
01
Discover
Audited five competing ARM platforms and mapped the distinct jobs a media manager performs, each with a different cognitive mode and urgency level.
02
Hypothesize
Turned research into five structural insights — from "triage over analysis" to "AI needs visible reasoning to be trusted" — each pointing to a specific design decision.
03
Redesign
Rebuilt the experience around proactive insight surfacing and a dedicated AI trust layer, rather than a passive, query-only dashboard.
04
Test
Validated the mobile experience as a distinct job, not a shrunk desktop view, rebuilding the information architecture from that job up.
05
Ship
Delivered a production-ready design system with atomic architecture and semantic tokens, ready for developer handoff.
06
Measure
Defined the three target metrics — Time on Task, Error Rate, Time to Insight — to measure real impact once live.
Research Methods
Every structural decision in Meridian traces back to a research question answered before any wireframe was drawn.
Competitive audit: I analyzed five ARM platforms — Mediaocean, Prisma, Allocadia, Google Campaign Manager, and an independent agency tool. The consistent finding: all of them show everything to everyone at all times. The result is cognitive overload, workaround spreadsheets, and users who stop trusting the platform.
Jobs to Be Done mapping: I mapped three distinct jobs Marco performs at different times with different cognitive modes.
Data hierarchy analysis: I categorized every available metric by decision-driving value vs. reporting habit. Most platforms show both equally. Meridian surfaces only what drives decisions — everything else is accessible on demand.
Key Insights
01
User doesn't analyze — he triages.
He opens the dashboard 2–3× per day for under 2 minutes. His first question is always: "Is everything okay?" not "Show me all the data." → Design for the triage job first.
02
Pacing without context is meaningless.
"40% budget spent" is fine at day 40. It's critical at day 60. A number alone doesn't communicate urgency. → Encode both dimensions visually — spend and timeline together.
03
Mobile is a different job, not a smaller screen.
On mobile, user has 90 seconds between meetings. A 13-column table at 390px doesn't answer his question faster. → Rebuild the mobile IA from the mobile job up.
04
The platform is passive when it should be proactive.
Every existing ARM tool waits to be queried. User discovers problems from his agency — days late. → Surface anomalies before user looks for them.
05
AI recommendations need visible reasoning to be trusted.
A budget recommendation that says "increase by 38%" with no explanation will be ignored or second-guessed. → Show the reasoning chain before the action button.
User Flow
Three primary flows. Each designed for a distinct job.
Main Hub: My Campaigns
The portfolio-level view — Marco's daily starting point. A single page that answers: "Is everything okay across all my campaigns?"
AI Alert Strip above the fold
The platform has access to real-time campaign data. But it stays silent until the user decides to look. User discovers underpacing campaigns from weekly agency reports — 3 to 5 days after the problem started. The dashboard that should protect him is just watching. Decision: A persistent AI strip positioned between the page header and the KPI line — the first piece of content user encounters on every dashboard open. 5 insight types with distinct UX behavior.
Campaign Detail: Brand Full Funnel
The investigation-level view — User's deep dive into a single campaign. Answers: "Why is this happening, and what should I do?"
AI Trust Design Patterns
AI recommendations involve real financial decisions. A budget change of 38% needs more than a button. Without visible reasoning, users either ignore recommendations or apply them blindly — both are failures. Decision: Four trust design patterns integrated into the AI layer. Each addresses a different moment in the trust-building journey.
Role Contract
Establishes AI boundaries on first launch. Without clear boundaries, users either over-trust AI (apply everything blindly) or under-trust it (ignore everything). The Role Contract calibrates expectations before the first interaction — not after the first mistake.
Contextual Entry Points
A generic "Chat with your data" modal has no context. Marco has to explain what campaign he's looking at before every question — cognitive overhead that kills adoption.
Reasoning Glimpse
Makes the AI's internal logic visible before action. Why collapsed: Marco can act without reading it. But if he's unsure — the logic is one tap away. This is progressive disclosure applied to AI reasoning. Design principle: reasoning before action, always.
Confidence Thermometer
hows AI certainty level — but only when it matters. Why show only non-high confidence: Showing "High confidence" on every recommendation trains users to ignore it. Showing uncertainty only when it exists makes the signal meaningful.
Mobile as a different scenario
A 13-column table at 390px is not a smaller table. It's the wrong answer to a different question. Decision: Rebuilt mobile IA from scratch for the 90-second check-in job. Table → Campaign Cards. Sidebar → Bottom Navigation Bar. 5-tab switcher → Dropdown selector. Single view → Segmented Control.
Design system
Atomic architecture means fixing a component state propagates automatically to every screen. Semantic tokens mean Dark mode works without touching a single component. Developer handoff includes token names, not hex codes.
THE OUTCOME
Time on Task
Error Rate
Time to Insight
A NOTE FROM THE DESIGNER
The best onboarding doesn't ask users to trust the product blindly — it shows them, immediately, what they're getting.