The Problem

User Problem

D/Cipher processes rich psychographic and communication data, generating scores and trait profiles for individuals and teams. Early prototypes presented this as dense tables and raw score grids. Enterprise users — HR leaders, organizational consultants, and researchers — found these impossible to act on. They either over-trusted a number without understanding what it meant, or dismissed the product entirely as a black box.

Business Problem

Without interpretable outputs, the product could not demonstrate ROI during enterprise trials. Trust in AI-generated behavioral insights is not given — it has to be designed. This was the core challenge: make complex probabilistic data feel credible, clear, and human at the same time.

Constraints

• No design system existed. I had to establish conventions while simultaneously shipping features.

• AI outputs carried inherent uncertainty. The UI had to represent confidence levels honestly without alarming non-expert users.

• Users ranged from PhD-level researchers to executive stakeholders with no statistical background — both had to succeed with the same interface.

• Small engineering team meant every pattern I designed had to be reusable. Custom one-offs were a luxury we couldn't afford.

Discovery & Research

I conducted structured interviews with 3 distinct user types: enterprise decision-makers, internal researchers, and domain analysts. I mapped their mental models of behavioral data — most thought in narrative terms ('this person tends to...') rather than statistical ones ('confidence: 0.72').

Key insight: users trusted the product more when insights were framed as hypotheses to explore, not conclusions to accept. This became our core design principle, and it cascaded into every information hierarchy decision.

Design Process

1. Defining Product Design Principles

Before touching Figma, I worked with the founders to establish 4 non-negotiable principles: Clarity over density. Interpretability over raw precision. Confidence without intimidation. Consistency across all analytical views. These became the decision filter for every design review — not aesthetic opinions, but product values.

2. Reframing the Information Architecture

I restructured dashboards around user questions — 'What drives this person's communication style?' rather than around data categories ('Trait X: 68%'). This shift reoriented the product from data-first to insight-first, and reduced time-to-first-meaningful-action significantly in prototype testing.

3. Designing for Uncertainty

I developed a visual system for representing AI confidence — subtle saturation gradients and contextual annotation patterns — so users could distinguish high-confidence insights from exploratory signals without needing a statistics background. I tested 3 different approaches with internal users before landing on the final pattern.

4. Building the Design System from 0

I established the full component library: analytical card containers, chart states (loading / empty / partial data / error), behavioral scoring modules, comparison layouts, and progressive disclosure patterns. The system covers ~40 components and allows engineers to ship new insight types without design involvement.

5. Cross-Functional Process

I ran biweekly design reviews with the data science team to ensure visualizations accurately reflected model outputs. I also facilitated insight walkthrough sessions — users narrating their thinking aloud while interacting with prototypes — to surface comprehension gaps that structured usability testing might miss.

Key Decisions & Tradeoffs

Progressive disclosure over full data exposure

Stakeholders wanted all model outputs visible upfront to demonstrate product depth. I pushed back: showing everything created decision paralysis and made the product feel overwhelming rather than intelligent. I designed a layered system — summary insights by default, with drill-down available for power users. Tradeoff: advanced users had an extra click. Compromise: a density toggle satisfied both groups.

Hypothesis framing over score-first presentation

I reframed all insight copy from 'Score: 74 — High openness' to 'This person tends to explore new approaches before committing.' This required close collaboration with data scientists to write language that was accurate and non-overstated. It took 3 rounds of iteration with researchers to calibrate the right specificity. The result: users stopped asking 'what does this mean?' and started asking 'what should I do with this?'

Neutral palette as a values-driven design decision

I deliberately chose a restrained, low-saturation color system — explicitly avoiding red/green for behavioral traits — to prevent users from assigning moral judgment to behavioral patterns. This was an ethical design choice, not an aesthetic one, and I documented it explicitly in the design system.

Outcomes

• Established a complete design system from 0, now covering 40+ components across all product surfaces. This reduced average design-to-handoff time per feature.

• Moved the product from prototype stage to active enterprise trials.

• Designed the full visual and interaction language - now the primary asset in investor demos and client pilots.

What I'd Do Differently

I'd have established a lightweight usability testing cadence earlier. In the first 3 months, I relied heavily on stakeholder feedback as a proxy for user feedback — a structural mistake. Even a monthly panel of 3–5 real users would have surfaced comprehension gaps faster and reduced iteration cycles. I now advocate for this in every project kickoff.