Install or upgrade by 20 November for BFCM60 plus a free done-for-you setup.Install by 20 Nov · BFCM60 + free setup

05d 19h 35m 35s05d 19h

See the offer
Face Age

How to Validate AI Skincare Recommendations Across Ages, Skin Tones and Skin Concerns

To validate AI skincare recommendations, write down what a sensible product list looks like for each type of shopper, then scan a small panel of real people who cover different ages, skin tones and concerns, and compare. In Face Age, most fixes happen in product mapping: severity levels, AND/OR conditions and exclusions. Recommendations are on every Shopify plan.

That's the short version. The longer version is below, because "check it works" is easy to say and easy to skip. A setup can look perfectly fine on the founder's own face and still be odd for half the customer base.

Key terms

  • Recommendation validation: checking that the products a scan suggests fit the person who scanned, for every kind of customer you sell to.
  • Test panel: a small group of people chosen to cover your customer range, not a random sample.
  • Expected list: the products you'd want each panel member to see, written down before you scan them.
  • Severity level: low, medium or high, the intensity Face Age assigns to each visible concern.
  • Exclusion rule: a rule that hides a product when a quiz answer matches, such as sensitive skin. Exclusions always win.

The scan itself is a cosmetic analysis for shopping guidance. Not a medical diagnosis; does not replace a dermatologist.

Why test recommendations separately from the scan?

People tend to blur two questions. One is "does the analyzer read faces consistently?" That's about the scan, and we covered it in how to test a skin analyzer before you trust it. The other is "do the right products come back?" That's about your catalog and your rules.

The second one is where most problems live. A scan can correctly find medium dryness and high pores, and the store can still show a clay mask first, because the mask was linked to "all concerns" during a rushed setup. The shopper doesn't blame the mapping. They just stop trusting the result.

Step 1: Write the expected lists first

Before anyone scans, sit down with whoever knows your products best and write a short table. Rows are shopper profiles. Columns are "should see," "may see" and "should never see."

ProfileShould seeShould never see
20s, oily, visible pores and breakoutsLightweight gel moisturizer, BHA tonerRich night balm
40s, dry, fine linesHydrating serum, richer creamStrong mattifying products
50s+, deep wrinkles, dark spotsRetinoid-style serum, SPF, brightening serumHarsh scrubs
Any age, sensitive, visible rednessGentle cleanser, barrier creamStrong exfoliating acids
Deeper skin tone, uneven pigmentBrightening and tone-evening productsProducts described only for "redness"

Keep it to your real range. The point isn't to be exhaustive. It's to have something to compare against, so "this looks off" becomes "row 3 should never see this product."

Step 2: Build a test panel that covers your customers

This is where ages and skin tones come in. You want coverage, not volume.

  1. Ages: at least one person in their 20s, 30s to 40s, and 50s or older. Mature skin tends to show deep wrinkles and eye-area concerns that younger panel members won't trigger.
  2. Skin tones: include light, medium and deep skin. Pigmentation and redness read differently across tones, so the products that come back for them are worth checking closely.
  3. Concern profiles: oily and dry, breakout-prone and not, sensitive and not.
  4. Consent: staff, friends and family are fine, but ask first, explain what the scan does, and let them see their own result.

Eight to twelve people is usually enough for a store. If your customers skew heavily toward one group, add more people from that group.

Step 3: Scan in fair conditions

Bad photos give bad results, and you'll waste time chasing mapping problems that are really lighting problems.

  • Use soft daylight facing the person, no strong backlight.
  • No heavy makeup, no filters, hair off the face.
  • Scan each person twice. If the two results are far apart, fix the photo setup before judging products.

Face Age checks lighting, position and whether the person is looking straight before it accepts a photo. Our photo tips for a usable skin scan post has a version you can give shoppers too.

Step 4: Compare the product lists to your table

Now go row by row. For each panel member, note three things: products that should be there and are, products that shouldn't be there and are, and products that should be there and aren't.

When something's wrong, the cause is usually one of these:

  • A product is linked to too many concerns. It shows up for everyone. Narrow it to the concerns it's actually for.
  • Severity levels are too wide. A strong product selected for low, medium and high will reach people who only need something mild. Select medium and high only.
  • Two rules in Condition should have been OR. In Face Age, rules inside Condition use AND logic, so a product set to "dark circles, medium" and "eye bags, medium" needs both. If either should be enough, split it.
  • No exclusion for sensitive skin. Add an Extra Condition that excludes exfoliating acids when the quiz answer is "sensitive." Exclusions always win over includes.
  • Automate AI linked things loosely. If you used automatic linking, review every product it touched. Its strictness setting can be high, medium or low.

The pillar guide on how AI skin analysis picks products explains each of these settings in detail.

Step 5: Check the edge cases on purpose

The middle of your range usually works. The edges are where people get odd lists.

Visible redness and sensitivity. These shoppers are the most likely to react badly to a wrong suggestion. Run the checks in gentle product matching for redness and sensitive skin.

Mature skin. Make sure your anti-aging range shows at the right severity, and that it doesn't also show to someone in their 20s with low fine lines. The signs of aging matching guide has a mapping table you can copy.

Pigment on deeper skin tones. Check that brightening and tone-evening products appear when the scan finds pigmentation, and that the copy around them is respectful. Our post on dark spots and pigment product matching goes further.

Shoppers with almost nothing to fix. Someone with low severity everywhere should still see a sensible basic routine, not an empty page or your entire catalog.

Step 6: Read the result page as a shopper

Numbers and product lists aren't the whole experience. Read every line on the result page as the panel member would.

  • Does the wording describe what's visible, or does it sound like a verdict on their health?
  • Do product reasons match what the scan showed? The Product Insight block on the product page should explain the match in plain language.
  • Is the disclaimer visible without scrolling?

If anything sounds medical or alarming, rewrite it. You control most of the copy.

Step 7: Keep validating after launch

A recommendation setup that was right in March can drift by June. New products get added with default mappings, old ones go out of stock, and nobody re-checks.

Put three habits in your calendar:

  1. After every catalog change, scan two panel members whose profiles the new products target.
  2. Monthly, look at the Analytics page: product display rate, top products added to cart after a scan, and products shown after analysis. A product shown to everyone but rarely added is worth a mapping review.
  3. Quarterly, rerun the whole panel and update the expected-list table.

On the Enterprise plan, confidence guardrails can route uncertain results to human review, which suits brands with a large or regulated catalog. For most stores, the panel routine above does the job.

What not to do

  • Don't judge the whole setup on your own face.
  • Don't promise shoppers results from the products. You're checking fit, not outcomes.
  • Don't skip exclusions because "the scan will handle it." The scan reads the face; it doesn't know which of your products are harsh.

Tools you'll use in Face Age

Everything above happens in Shopify Admin > Apps > Face Age: product connections with Condition and Extra Condition, the quiz builder for sensitivity and skin type questions, and the Analytics page. The product recommendations feature page shows how matching looks to a shopper, and the Shopify app page covers installation and plans.

FAQ

How should beauty brands validate AI skincare recommendations? Write expected product lists per shopper profile, scan a panel that covers your ages, skin tones and concerns, compare, fix mapping, and repeat after launch.

How many test people do I need? Eight to twelve chosen for coverage is a workable start for a store.

Is a bad recommendation usually the scan's fault? Usually not. Most odd lists come from mapping: too many concerns, wide severity levels or missing exclusions.

Which Face Age plan includes product recommendations? Every Shopify plan, starting with Start at $59 a month.

Does validating recommendations make the scan a medical tool? No. Not a medical diagnosis; does not replace a dermatologist.

Share
Ask AI for a summary
Article byJJ SEOadmin

Put a skin scan on your store today

Add Face Age to your site with the SDK, a plugin or the Shopify app — and start with a 7-day free trial.

Choose which cookies Face Age can use on this site. You can change this any time from “Cookie preferences” in the footer. Privacy Policy