How I replaced a fully manual compliance review process with an AI-powered audit pipeline using Claude + Figma MCP — cutting audit time from days to minutes.
44
CX criteria automated
27
XP criteria automated
~10m
Per audit turnaround
32+
Product teams enabled
01 / The Problem
Time Cost
2–3 days per audit manually cross-checking 27 experience criteria and 44 CX standards across Figma files one by one.
No Consistency
Results varied by reviewer. Two people auditing the same product could reach different conclusions on identical criteria.
No Self-Serve
Product teams queued for central DesignOps to schedule every audit. No way to check compliance before shipping.
Not Scalable
Manual process couldn't scale past a handful of audits per month across 30+ product teams.
The Old Process
02 / What I Built
CX Assessment Tool
Evaluates against 44 criteria across 15 CX standards. Generates an interactive HTML report with in-report approval workflows — reviewers approve or flag inside the report itself.
Experience Principles Verifier
Evaluates any Jio product against 27 criteria across 10 experience principles using live Figma design data extracted via Figma MCP. No screenshots — real structured data.
Outcome
Product teams run compliance checks without waiting on central DesignOps — anytime, before any release.
03 / Process
Figma file URL provided
Component data, tokens, and layout pulled live via Figma MCP
Figma MCP extracts design context
Real structured data — no screenshots needed
Claude AI evaluates against criteria
27 XP or 44 CX criteria — every one checked with a rubric
Pass / Partial / Fail scored per criterion
Each criterion gets a verdict with clear reasoning
Interactive HTML report generated
Downloadable, shareable — no special tool to open
In-report approval workflow
Reviewers approve, flag, or mark criteria inside the report
04 / Before vs After
Before — Manual Process
After — Automated Pipeline
05 / Learnings
01
AI works best with structure
Claude's evaluation quality improved significantly when criteria were written as specific, testable questions — not broad principles. Prompt engineering took as long as building the tool.
02
Figma MCP is powerful but selective
Live design context extraction works well for components, tokens, and layout. Nuanced experience calls still need a human reviewer — the tool doesn't replace judgement.
03
Adoption is a design problem
Getting product teams to trust an automated verdict required as much change management as engineering. The technical part was the easier half.
Honest note
This tool handles ~70% of what a manual audit covers. The remaining 30% — nuanced emotional and contextual judgement — still requires a human reviewer.
Open to consulting, advisory, or collaboration on AI-powered design ops.