Portfolio Case Study / Conversion Optimization

Staking out a higher-converting lead form.

A controlled redesign of the Oregon Lawn Barbers quote form — replacing a single high-friction form with a progressive-disclosure wizard, then defining how the change would be measured in a controlled A/B test. Built in React. Instrumented for behavior. Designed to be judged by data, not taste.

Role
Design + Frontend
Stack
React / TS / Vercel
Method
A/B plan · 2-prop z-test
Result
Prototype validated
CASE STUDY STATUS: INTERACTION + MEASUREMENT PLAN VALIDATED|Reported figures below remain illustrative until a field sample reaches the documented threshold.
01

The case, in four measured plots.

Every optimization starts as a survey: define the ground, form a hypothesis, build the instrument, read the result. No step skips ahead of its evidence.

Plot 01 · Challenge

Friction at the ask

A local service page where the quote form is the whole funnel — and a wall of fields is the last thing a visitor sees before leaving.

Plot 02 · Hypothesis

Trade depth for momentum

Breaking one long form into staged steps — and returning instant value mid-flow — should lower abandonment versus the existing single form.

Plot 03 · Architecture

Built to be measured

Both variants engineered in React with per-field telemetry, deterministic bucketing, and accessibility held to a fixed floor.

Plot 04 · Verdict

Await field traffic

A planned two-proportion test will decide it. The result lands when the required field sample does — not before.

02

Control vs. Challenger.

The control reflects the Oregon Lawn Barbers form documented at the start of this study — a fair baseline, not a straw man. The challenger changes the structure of the ask, not the honesty of the comparison.

Variant AThe Control

Every field exposed at once. One decision point, high up front. This is the documented baseline for the test plan.

Full name
Email
Phone
Address
Service needed
Property size
Message
Submit Request
Variant BThe Challenger

Staged disclosure. Step one asks for a tap, not typing. Step two returns an instant estimate. Contact comes last, once momentum exists.

MAINTENANCE
DESIGN
HARDSCAPE
🛡 Licensed, bonded & insured · OR
Get My Estimate →
03

Engineered to hold up under inspection.

The implementation decisions behind a testable build — not screenshots alone, but defined behaviors and measurement hooks.

A · Assignment

Deterministic bucketing

Visitors hash to a stable variant and stay there on return. No flicker, no re-randomizing mid-session via a persisted visitorId.

B · Telemetry

Per-field drop-off

A useTelemetry hook logs step progression and the exact field_id where a session stalls — the raw material for a friction map.

C · Floor

Accessibility, non-negotiable

Full keyboard paths, visible focus rings, aria-live on the dynamic estimate, reduced-motion respected. Held constant across both variants so it never confounds the result.

04

Behavior, defined event by event.

The telemetry design captures the events that explain why a number moves — not just that it did. The trace below demonstrates the event schema with illustrative session values; it is not production traffic.

Representative telemetry · illustrative session
10:42:01 experiment_viewed { variant: "B" }
10:42:09 step_progression { step: 1→2, service: "maintenance", dwell: 7.9s }
10:42:15 estimate_rendered { sqft: 6500, label: "illustrative" }
10:42:28 interaction_friction { field_id: "phone", event: blur_empty }
10:42:41 conversion_achieved { variant: "B", steps: 3, total: 40s }
05

The verdict is a calculation, not a claim.

Completion rate tells you what happened. The significance test tells you whether to believe it. The test plan defines both before any field result is reported.

Completion rate
CR = (completed leads / total session starts) × 100%
Two-proportion z-test
z = (p̂B − p̂A) / √ p̂(1−p̂)(1/nA + 1/nB)

Sample size n calculated in advance from baseline CR and the minimum detectable effect. No early stopping in the test plan.

Verdict · pending field sample
Pending
No production lift claimed
  • Baseline CR (control)pending
  • Observed CR (challenger)pending
  • 95% CI on liftpending
  • p-valuepending
  • Required n / arm2,726
06

The interaction, prototyped end-to-end.

Alongside the shipped React build, the core progressive-disclosure flow was rebuilt and verified natively in Axure RP 11. The browser-accessible version below lets reviewers test the same service, estimate, validation, and success journey without an Axure account.

◇ Verified · Native Axure RP 11 + browser preview

From interaction design to working instrument

The native source file documents the interaction design process; the launchable preview demonstrates service gating, linked square-footage estimates, contact validation, backward navigation, success, and restart.

  • Four named dynamic-panel states
  • Native conditional cases and global variables
  • Boundary-tested estimate tiers
  • Complete mouse and keyboard reviewer flow
◇ Test the interaction →
07

Built in two honest phases.

The rig exists now. The result arrives when real traffic has spoken. The case study is transparent about which is which at every moment.

Complete
Phase 1 · The instrument

Build & validate

  • React control + challenger variants
  • Deterministic bucketing & persistence
  • useTelemetry event layer
  • Accessibility floor + reduced motion
  • Documented sample-size calculation
Planned
Phase 2 · The reading

Collect & conclude

  • Deploy to an instrumented test surface
  • Drive real traffic to both arms
  • Reach documented n, no peeking
  • Run two-proportion test + CI
  • Publish the real verdict here
Tony Martinez / Frontend & Conversion

The build rationale, method, and verification summary.

This presentation documents the working interaction, measurement architecture, accessibility decisions, and field-test plan. The production repository remains private.

Public case-study presentation based on a private production codebase. Client implementation details and repository history remain private.
Estimate ranges shown are illustrative interaction demos, not the business's quote engine. Metrics labeled "illustrative" or "pending" are placeholders until real data is collected.