How are UX metrics verified in regulated financial product design?

Three-tier verification: primary source (the actual measurement event — Mixpanel funnel event, Hotjar session, usability session recording), calculation method (the formula and sample size — e.g. pooled standard deviation for two-sample UX comparisons, Cohen's d for effect size), and confidence interval or caveat (e.g. 'n=15, within-subjects design, p<0.05 on task completion time'). Every metric in the portfolio has all three tiers documented. If the calculation relies on an estimate or assumption, that is disclosed explicitly — for example, engineering-week calculations are described as 'approximately X weeks based on [method]' not presented as audited figures.

What is the correct way to measure usability improvements in institutional trading UI?

Task completion time is the wrong primary metric for institutional trading surfaces because a faster but less confident trading decision is a worse outcome — not a better one. The correct paired metrics are task completion time AND task confidence ('How certain are you about this decision?' on a 5-point scale). A redesign that improves completion time while dropping task confidence has failed, even if the efficiency number improved. This was learned empirically from a Finlogix redesign in 2022 that measured +22% task completion speed but -31% task confidence — the redesign was pulled, and the direction was rebuilt around a different question: how do we make the existing information density faster to parse, not how do we show less.

How long should you measure financial product design changes after launch?

Minimum 90 days post-launch before drawing conclusions on regulated financial products. Two reasons: (1) institutional trader behaviour has a learning period — the first 30 days often show inflated task times and error rates as users adapt to any change, regardless of quality; (2) regulatory-adjacent metrics like disclosure acknowledgement rates, support ticket volume for compliance questions, and EDD completion rates take 60–90 days to stabilise after an onboarding change because the cohort mix shifts as new users arrive. KYC funnel metrics in particular need a full 90-day window to capture the tail of users who abandon at EDD and return after finding their source-of-wealth documentation.

How do you calculate the engineering cost savings of a compliance design system?

The calculation is straightforward but requires honest sourcing of inputs. Pre-system baseline: count the engineering hours spent on a comparable regulatory update (not estimated — pull from sprint logs or engineering retrospectives). Post-system measurement: count the same metric after the compliance primitive architecture is in place. For ACY Securities, pre-system regulatory updates ran 3–4 engineering weeks per update across five platforms; post-system updates run 3–5 days. Across eight documented regulatory changes (ASIC leverage cap updates, ESMA inducement ban, FCA risk warning changes), the compression yields approximately 20 engineering weeks of avoided spend. This is presented as 'approximately' because the per-update variability is real — some changes were simpler and some more complex than the mean.

Data Integrity Commitment

How I verify every metric in this portfolio

Q: What is the correct way to measure usability improvements in institutional trading UI?

Task completion time is the wrong primary metric for institutional trading surfaces because a faster but less confident trading decision is a worse outcome — not a better one. The correct paired metrics are task completion time AND task confidence ('How certain are you about this decision?' on a 5-point scale). A redesign that improves completion time while dropping task confidence has failed, even if the efficiency number improved. This was learned empirically from a Finlogix redesign in 2022 that measured +22% task completion speed but -31% task confidence — the redesign was pulled, and the direction was rebuilt around a different question: how do we make the existing information density faster to parse, not how do we show less.

Q: How long should you measure financial product design changes after launch?

Minimum 90 days post-launch before drawing conclusions on regulated financial products. Two reasons: (1) institutional trader behaviour has a learning period — the first 30 days often show inflated task times and error rates as users adapt to any change, regardless of quality; (2) regulatory-adjacent metrics like disclosure acknowledgement rates, support ticket volume for compliance questions, and EDD completion rates take 60–90 days to stabilise after an onboarding change because the cohort mix shifts as new users arrive. KYC funnel metrics in particular need a full 90-day window to capture the tail of users who abandon at EDD and return after finding their source-of-wealth documentation.

Q: How do you calculate the engineering cost savings of a compliance design system?

The calculation is straightforward but requires honest sourcing of inputs. Pre-system baseline: count the engineering hours spent on a comparable regulatory update (not estimated — pull from sprint logs or engineering retrospectives). Post-system measurement: count the same metric after the compliance primitive architecture is in place. For ACY Securities, pre-system regulatory updates ran 3–4 engineering weeks per update across five platforms; post-system updates run 3–5 days. Across eight documented regulatory changes (ASIC leverage cap updates, ESMA inducement ban, FCA risk warning changes), the compression yields approximately 20 engineering weeks of avoided spend. This is presented as 'approximately' because the per-update variability is real — some changes were simpler and some more complex than the mean.

Every number cited here will be questioned by Legal, validated by Product, scrutinised by Engineering. This page is the answer. Four evidence tiers, transparent formulas, attribution-honest claims. Confidential employer data is paraphrased, not exposed — the same discipline I'd bring to your data.

4 Evidence tiers

3 Live formulas with sample size + limitations

100% Sources attribution-honest

0 Confidential leaks · ever

View Complete Verification Manifest (JSON)

Machine-readable structured data — every claim, source, formula, and sample size in one file.

The Framework

Four evidence tiers — every claim sits in one of these

Public evidence first, NDA paraphrase last. The tier is always disclosed alongside the number. Reviewers can verify Tier 1 themselves; Tiers 2–4 unlock progressively in interviews.

Tier 01

Public evidence

Verifiable by anyone through public sources. Awwwards nominations, ASIC public records, jurisdictional disclosures visible in website footers.

VERIFIABLE NOW

Tier 02

Analytics screenshots

Hotjar, Google Analytics 4, Search Console. Sample size and date range disclosed. Live dashboard walkthrough in video interviews.

SCREENSHOT ON REQUEST

Tier 03

NDA-protected (paraphrased)

Confidential business metrics presented as relative changes ("+35% improvement") with methodology disclosed but exact values withheld. Final-round disclosure under bilateral NDA.

CONFIDENTIAL

Tier 04

Executive collaboration

Sanitised investor-deck templates, FY26/27 financial framing artefacts, C-level strategic feedback summaries. Walkthroughs in final interviews.

FINAL ROUND

Why this page exists

Quantifying design isn't proof — it's trust infrastructure

I'm a designer, not a data scientist. But every design decision I make has consequences Legal, Engineering, and the CFO will ask me to defend. This framework is how I answer them — with rigor, attribution, and the same confidentiality I'd carry into your team.

Question 01 · Legal

"How do users actually see this risk disclosure before they commit?"

Answered with eye-tracking samples, scroll-depth analytics, and interaction logs. Not a screenshot of the disclaimer; evidence the disclaimer was read.

Question 02 · Engineering

"Does this redesign improve task time or is it just prettier?"

Answered with controlled usability data, before/after times, sample size, and statistical caveats. Cohen's d only when the test design supports it.

Question 03 · CFO

"What's the ROI of the design system investment?"

Answered with implementation-time delta against baseline period and explicit confounding factors (TypeScript migration ran in parallel — not all gain is design system attribution).

Attribution Discipline

What I measure, what I don't claim, how I handle confidential data

Most designer portfolios over-attribute team outcomes. Mine doesn't. The line between "I designed this" and "the company reported this" is drawn explicitly on every claim.

Principle 01

What I measure

User behaviour (task time, error rates, time-on-task), system adoption (component reuse rate, design system uptake), and compliance outcomes (zero design-related regulatory findings).

DESIGN-ATTRIBUTABLE

Principle 02

What I don't claim credit for

Revenue, user growth, conversion rates. These are team outcomes driven by Product, Marketing, Engineering — not design alone. Company-wide metrics are attributed to source, not to me.

TEAM-ATTRIBUTABLE

Principle 03

How I handle confidential data

Three tiers: public evidence anyone can check, analytics screenshots in interviews, confidential data shown as directional estimates with methodology. I won't compromise a previous employer's confidentiality to make my portfolio look better.

PROFESSIONAL ETHICS

Calculation Methodology

Three live formulas with sample sizes and limitations

Every quantitative claim in the portfolio is backed by a transparent formula. Below: the math, the sample, the confounders that should keep you skeptical of the headline number.

Formula 01

Design system implementation speed

Baseline 3 days/component (n=10, 6-month pre-system) vs. 2 days (n=20, 9-month post-system). Reported as range to absorb engineer-level and component-complexity variance. Confounders: TypeScript migration ran in parallel — not all gain attributable to design system. Directional, not causal.

OBSERVATIONAL DIRECTIONAL ONLY

Formula 02

Order placement flow time reduction

Moderated lab test, n=15 (5 novice / 7 intermediate / 3 expert). Within-subjects, ±0.2s manual timing. Cohen's d = 2.47 (very large) — expected, since steps were cut 6→2. Limitations: learning effect not counterbalanced, lab ≠ live trading stress.

CONTROLLED LAB d = 2.47

Formula 03

TradingCup SEO performance

187K organic clicks ÷ 31M impressions = 0.603% CTR over 12 months (Google Search Console). Attribution: SEO strategy was Marketing + Content. I claim only the UX-side scaffolding — page structure, Schema.org markup, Core Web Vitals.

PUBLIC TIER 1 UX SCAFFOLD ONLY

Honest Assessment

What this evidence means — and what it would take at institutional scale

Small samples done well are still small samples. The portfolio's strongest study (Finlogix, n=15 within-subjects) is a directional signal, not a guarantee. Naming that gap honestly is part of the methodology.

What I have · current portfolio

Exploratory + one well-designed comparison

Finlogix n=15 within-subjects, p<0.001, d=2.47
Nova n=8, HorizonSync simulated scenarios — proof-of-concept
Production analytics (n=100K+) used only for IA-level claims
Every directional finding labelled directional, not validated

What I'd add · institutional scale

Proper randomised + longitudinal protocol

n=200+ per cell with random assignment (95% / ±7%)
Counterbalanced order, controls for trader experience & market regime
Effect-size + confidence intervals, not just p-values
3–6 month longitudinal tracking to catch learning effects
Multi-market replication — confirm results aren't one-region artefacts

Why this matters for institutional roles: on retail platforms (100K+ users), fast exploratory testing drives shipping. At institutional scale — where one UX choice can affect billions in AUM or expose the firm to regulatory risk — design needs rigorous testing, peer review, and reproducible methods. I'm ready to partner with quantitative researchers and compliance teams to validate decisions to that bar.

Real-World Application

Four claims, four verification paths

Each example below is one tier of evidence. Tier 1 examples can be checked right now; Tier 2 unlocks in interviews; Tier 3 stays paraphrased.

Google Search Console — TradingCup organic search traffic over 3-month period showing clicks and impressions — Google Search Console · TradingCup · 3-month window. Tier 1 public evidence — live dashboard walkthrough available in interview. SEO strategy attributed to Marketing + Content; UX scaffolding mine.

Tier 01 · Public

TradingCup · 187K clicks

Google Search Console screenshot above. Live GSC dashboard walkthrough on request. CTR 0.6% / position 21.7 demonstrates discovery effectiveness — UX side, not the SEO strategy itself.

VERIFIABLE NOW

Tier 01 · Attribution

100K+ traders · 40+ jurisdictions

Company-wide metric, publicly stated by my former employer. I designed for a platform operating at this scale — I do not claim user-acquisition credit. Jurisdiction count verifiable in website footer regulatory disclosures (ASIC, FCA, CySEC, FSA).

PLATFORM-STATED FOOTER-VERIFIABLE

Tier 02 · Analytics

User behaviour · Hotjar & GA4

Hotjar heatmaps + session recordings + GA4 engagement metrics drove design prioritisation. What I won't show: internal dashboard screenshots, session recordings, geographic distributions — these belong to my prior employer. Methodology walkthrough in interview; detailed data under NDA in final round.

NDA-PROTECTED METHOD ON REQUEST

Tier 01 · Live demo

Macro Signal · FRED + EDGAR pipeline

Every metric on Macro Signal Network is derived from public APIs — FRED, SEC EDGAR, U.S. Treasury FiscalData, BLS. Reproducible end-to-end by any reviewer with an API key. Zero NDA risk; nothing paraphrased from a prior employer's dashboard.

FULLY PUBLIC SOURCE

Why I take the confidentiality stance: If I share a previous employer's internal data to land a new role, why would you trust me with yours when I eventually leave? This page demonstrates I can measure design impact rigorously while honouring professional confidentiality. In financial services, that's not optional.

Related work

Verified data in production case studies

Case study Macro Signal Network Live FRED pipeline with Tier-1 public-source citations on every metric. Case study Finlogix — Institutional Market Data Institutional-grade data with explicit source attribution on every field. Case study TradeX Institutional Real-time FIX 4.4 execution with venue timestamps and audit trails. Case study LogixPanel Margin and exposure calculations with fully auditable data lineage.