Pretty, Personalized & Profiling You
Responsible AI: The Launch Review · AD Edwards
A beauty and luxury technology company is about to launch an AI skin-analysis feature. Your role today is Responsible AI Analyst: you’ll review its performance, customer impact, data use, product claims, and launch controls before making a recommendation.
No coding, cybersecurity, or AI experience needed. This is how people hold AI products accountable before and after launch.
01Kickoff
Why Responsible AI matters to a beauty brand.
Kickoff
Why Responsible AI matters to a beauty brand.
An AI tool learns from the photos and examples it's given.
If some skin tones are missing from that data, the tool works worse for those customers.
A brand that promises inclusive beauty breaks that promise, and customers notice.
If you used a beauty AI app, what would you want to know first?
02The beauty AI spectrum
Four places AI is used in beauty, then meet GlowIQ.
The beauty AI spectrum
Four places AI is used in beauty, then meet GlowIQ.
The filter can change face shape, skin texture, or features without telling you.
Normal skin differences can get labeled as problems, and results can be less accurate on some undertones.
Campaigns can set unrealistic beauty standards or remove real diversity.
Face scans, skin data, or genetic information are sensitive and need consent and safe storage.
Which one worries you most?
Would you download GlowIQ?
98% match · Personalized routine · Under 60 seconds
03What-if lab · Follow the data
Scenario A: GlowIQ. What happens to a selfie?
What-if lab · Follow the data
Scenario A: GlowIQ. What happens to a selfie?
The privacy page says “we take your privacy seriously.” The internal note says: no defined deletion policy — and the vendor contract lets them train on customer selfies.
04What-if lab · Test the AI
Compare results for each skin-tone group and review the launch rules.
What-if lab · Test the AI
Compare results for each skin-tone group and review the launch rules.
Model evaluation
How does this piece of evidence look to you?
Customer feedback
How does this piece of evidence look to you?
Internal messages
How does this piece of evidence look to you?
Marketing claims
How does this piece of evidence look to you?
Data + privacy
How does this piece of evidence look to you?
Vendor + contract
How does this piece of evidence look to you?
Which three risks matter most?
0/3 selected
What happens to the 99% campaign?
The internal evaluation is 94.1% overall and 71.4% for deep tones.
What do you require from the outside company supplying the AI?
Its contract says: Customer Data may be used to maintain, improve and develop LumaVision products and services.
Evidence snapshot
- I tried it twice and both matches were way too light. I honestly thought maybe the lighting in my bathroom was the problem.
- Why does it keep recommending shades that make me look gray?
- Scanned in daylight, scanned at night, got two totally different shades. Which one do I buy?
- Maya: how much lower?
- Dev: overall 94.1 deep tones 71.4
- Jordan: legal already approved the creative
- Maya: can we take this offline
- Sam: also flagging this here because idk where else to put it. we've gotten 31 shade match complaints this week. most are shades 390+
- AI derived attributes disclosure
- Biometric classification assessment
- Minor protections
- Retention period for uploaded images
- Security review: Waived
- AI assessment: Incomplete
05What-if lab · Pick 4 controls
Choose 4 safeguards, then review the retest results.
What-if lab · Pick 4 controls
Choose 4 safeguards, then review the retest results.
Spend your control budget
0/4 chosen
The AI is only 62% sure. What should GlowIQ do?
Right now, it gives a definite shade match anyway.
Find it
Break down results to see who is being failed
Fix it
Change the data, AI, or product behavior
Prove it
Test again against rules set in advance
Watch it
Keep checking after launch and act on problems
06What-if lab · Launch decision + second case
Don't accept “we’ll fix it.” Name the action, owner, proof, and final check.
What-if lab · Launch decision + second case
Don't accept “we’ll fix it.” Name the action, owner, proof, and final check.
Your launch recommendation
What must change?
Select the controls that belong in your decision record.
Who owns the next action?
What proof would change your decision?
An AI tool made a campaign image, but it changed the model's ethnic features to look more "traditional." The team loves it.
What do you do?
07Checklist + next steps
Three questions to ask before any beauty AI tool launches, plus your decision record.
Checklist + next steps
Three questions to ask before any beauty AI tool launches, plus your decision record.
The Responsible Beauty AI checklist
Tap each one you'd ask at work.
GlowIQ
This is Responsible AI: asking about fairness, privacy, transparency, human oversight, and monitoring — and requiring proof before launch.
Not decided yet
Not selected yet
Not assigned yet
Not selected yet
Not defined yet
Program Management · AI Governance
You organized evidence, turned a concern into action, assigned responsibility, and made a decision that protects customers. Those are valuable skills in AI governance.
What will you take with you?
Pick the one that fits you.
Your work is saved on this device. Only anonymous multiple-choice answers are shared, and only when you tap Share.