Pace
Running with friends
Year
2026
FOCUS
WELLNESS
PLATAFORM
IOS, ANDROID
We’re committed to shaping better futures putting people first — our clients, our employees, and the users we serve. We pursue excellence. We embrace growth. We create what doesn’t exist. We’re Riverside.
Xees, and the users we serve. We pursue excellence. We embrace growth. We create what doesn’t exist.
✦ TLDR
A fictional AI-assisted product design project for modern operations teams.
Northstar Ops helps customer success leads turn scattered account signals into reviewable action plans. The design challenge was not making AI feel magical. It was making AI outputs legible, editable, and safe enough to use in real customer decisions.
Fictional concept
Product category: B2B SaaS productivity. Audience: customer success managers, CS ops leads, and founders running high-touch accounts.
Scope
Risk triage dashboard and account detail workflow. AI-generated plan drafts with evidence and edit states. Design system for confidence, escalation, and review patterns.
✦ AI in this project
AI was treated as a collaborator with limits, not an authority.
I used AI for synthesis, draft generation, and critique during the design process. In the product itself, AI recommendations were designed to expose evidence, confidence, and missing context before asking users to approve anything.
01
Synthesis
Clustered interview notes, support themes, and account-review pain points to find repeated decision bottlenecks.
02
Draft generation
Generated first-pass recovery plans, then designed edit controls so teams could accept, revise, or reject each step.
03
Critique
Stress-tested empty states, risky copy, and overconfident recommendations before sharing the prototype with users.
AI callout
The interface never asks teams to trust a recommendation blindly. Every suggested plan includes source evidence, confidence, assumptions, and a clear review moment before customer-facing action.
✦ Problem + constraints
Customer risk was visible everywhere, but accountable nowhere.
Teams had enough signals to act sooner, but the signals lived across disconnected systems. The work was slow because people had to gather evidence, interpret it, draft a plan, and defend the decision in a meeting.
Fragmented context
Account health lived across CRM notes, usage data, tickets, meetings, and team memory.
Low trust in automation
Teams wanted AI support, but only when they could inspect why a recommendation appeared.
High-stakes output
A bad plan could waste executive time or make a customer relationship worse.
Fig. 04 — Decision spine mapped with the CS lead, week 2
Workshop artifact
✦ Approach
A compact process focused on decision quality.
Instead of documenting every activity, the case study shows the few moves that changed the product direction: mapping the decision, designing confidence states, and prototyping the review workflow.
Map the decision
I mapped the account review from signal discovery to customer-facing action, then identified where teams needed AI acceleration versus human judgment.
Journey map / decision spine
Design trust states
I created patterns for evidence quality, confidence level, missing context, and escalation urgency so recommendations felt inspectable.
System rules / UI states
Prototype the loop
I built a clickable prototype and lightweight UI proof to test if users could understand, edit, and approve AI-generated plans.
Prototype / front-end spike
✦ Solution
The final design centers on reviewable AI outputs, not black-box automation.
The product experience pairs a scan-friendly risk queue with an account workspace where evidence, draft actions, and approval states stay visible together.
Why this matters
The design makes the user decide what is true, what is useful, and what is ready to share.
01
Interview prep
Our Product Designer has multiple interviews each week. Dia creates a report for every candidate, so he walks in knowing exactly who he’s talking to.
02
To-do lists
03
Decision making
Charlie Deets
✦ Impact
Measured as speed, clarity, and fewer dropped escalations.
Because this is a fictional project, metrics are written as portfolio placeholder copy. They show the kind of outcome framing to replace with real evidence later.
31%
Faster risk review in pilot task simulation.
18%
Fewer unresolved escalations after weekly review.
2.4×
More plan drafts reviewed before customer meetings.
87%
Of testers understood why a recommendation appeared.
✦ Learnings + next steps
The strongest AI UX pattern was restraint.
The more consequential the workflow, the more the interface needed to slow down at the right moments: show evidence, reveal uncertainty, and make approval explicit.
Learning
Confidence is not just a number. It needs supporting evidence, missing-context cues, and language that does not overstate certainty.
Next step
Expand the prototype into a multi-account planning view and test how managers compare AI suggestions across a book of business.
Open with the product and outcome before explaining the process.
Replace fictional metrics with real proof, even if the proof is qualitative.
Keep AI usage specific: what it helped with, where it was limited, and how judgment stayed human.
← More work
Fictional portfolio reference page · No real project, client, or case study is represented.
Eduardo.
Portfolio
Home
Work
Contact
About
Bio
Process
Resume
Connect
Dribbble
Behance
