Industries: Beauty

Customer service for Shopify beauty, skincare and cosmetics stores

Good beauty support answers product questions from your own listings and documents, and gives order and returns answers from live Shopify data. It never gives medical advice: when a customer reports a skin reaction, a person takes over. Fyncall lets you set those boundaries per channel, and send anything sensitive to your team.

Last updated: 19 November 2026 · By

Fyncall wrote this page, so weigh it accordingly. The rules described here are US rules, taken from the FDA's own pages and checked 19 November 2026. Nothing here is medical or legal advice. This page does not cover rules outside the United States.

What do customers ask beauty and skincare stores?

Common beauty questions fall into six groups: ingredients, shade matching, how to use a product, skin reactions, order status and returns of opened products. Each group needs a different answer, and only some should be answered by an AI.

QuestionExampleWhere the answer comes fromWho should answer
Ingredients "Is this serum fragrance-free?" Your product page and the ingredient list on the label AI, reading only what you have written down
Shade matching "I'm a 3W in another brand. Which shade is closest?" Your variants and your own shade guide AI for what the guide says; a person for anything it does not cover
Usage "Can I use the retinol and the vitamin C together?" Your written usage directions AI only if your directions answer it; otherwise a person
Skin reactions "My face is red and itchy after using this." Not a product question Always a person
Order status "Where is my order?" Shopify order, fulfilment and tracking data AI
Returns of opened products "I opened it and the shade is wrong. Can I return it?" Your written returns policy and the order AI for the policy; your team for refunds and exceptions

The pattern is simple. Facts that live in your catalogue, your documents or Shopify can be automated. Anything about a person's skin, health or body goes to a human.

Can an AI agent answer skincare questions safely?

Yes, for questions your own written content already answers, and only if you stop it from going further. The risk is not the order lookups. It is an AI that improvises advice about someone's skin.

Fyncall's AI answers from your knowledge base: it searches the documents you upload and writes its reply from the passages it finds. It also reads your synced Shopify products, including descriptions, variants, tags and stock. So the quality of its beauty answers depends on what you give it. A product page that lists full ingredients and clear directions gives it something accurate to say.

Then set the boundaries. Fyncall's guardrails are set per agent, and include:

  • Never Discuss. Topics the agent will not talk about. When a customer raises one, it redirects them to topics it does support. "Medical advice" is a sensible entry for a beauty store.
  • Must Follow. Rules applied in every conversation, for example "Never say a product treats, cures or prevents a skin condition" or "Never suggest a product is safe for a specific person's skin".
  • Never Say. Words removed from replies, such as terms your brand avoids.
  • Tool access. Switch off any tool the agent should not use, such as cancelling orders.

Guardrails reduce risk; they do not remove it. Any AI can make mistakes. That is why we suggest starting in Draft & Review, where the AI writes a reply and your team approves it before it is sent.

How should you handle a customer who reports a skin reaction?

Hand it to a person straight away, do not offer medical advice, and point the customer to a healthcare provider. The FDA's page on reporting cosmetic complaints says the first step is to stop using the product and contact your healthcare provider. The FDA also notes on that page that it does not provide medical advice, and neither should a store's support team.

A reply your team can adapt:

Hi [first name], I'm sorry to hear about your skin. Please stop using [product name] and speak to a doctor or pharmacist, who can advise you properly. We can't give medical advice, but we want to help with your order. Could you tell us your order number and the batch code printed on the product? A member of our team will follow up with you personally by [time or day].

The FDA's page on allergens in cosmetics says reactions most often appear as itchy, red rashes, and that anaphylaxis is rare with cosmetics but can be life-threatening: for anaphylaxis, the page says to call 911 and seek medical attention immediately. If a message sounds like an emergency, your reply should say that first.

Record each report. Under the Modernization of Cosmetics Regulation Act of 2022 (MoCRA), the "responsible person" must report serious adverse events linked to cosmetics used in the United States to the FDA within 15 business days. The FDA's page defines what counts as a serious adverse event. The responsible person is the manufacturer, packer or distributor whose name appears on the product label. If you sell your own brand, that may be you. If you resell other brands, it may be the brand owner, so tell them. Check which applies to you with a qualified adviser. This is not legal advice.

In Fyncall, turn a reaction report into a ticket with a priority (Low, Normal, High or Urgent) and an owner, so it does not get lost in the inbox. Tickets keep a full history, and agents can add internal notes. Ticketing with SLAs is part of the Pro plan.

How do you stop the AI answering reaction messages itself?

Combine a guardrail, a flow, escalation settings and a reply mode. No single setting detects every way a customer describes a reaction, so use more than one layer.

  1. Add "medical advice" to Never Discuss, so the agent redirects rather than advises.
  2. Write a flow in plain language, for example: "If a customer describes irritation, a rash, burning, swelling or any other reaction, do not give advice. Apologise, ask them to stop using the product and see a healthcare provider, ask for their order number, then escalate to the team." Flows can use support tools that create tickets and escalate; ticketing with SLAs is part of the Pro plan, so on Free have the flow escalate to the team.
  3. On web chat and WhatsApp, you can also add trigger keywords such as "rash", "reaction" and "irritation". On email, flows do not use keywords: they run from the scenario you describe.
  4. Keep the Explicit Human Request trigger on, so anyone who asks for a person gets one.
  5. Use Draft & Review while you start, so a person sees every reply before it is sent.
  6. Test the whole set-up. Fyncall has a sandbox where you can send the agent sample reaction messages and check that the guardrail and the flow work together before it talks to customers.

Each channel has its own agent with its own flows, so add the flow to every channel you use. The Free plan includes one AI workflow, so on Free the flow can run on one channel only; Pro has unlimited workflows. On channels without the flow, rely on the Never Discuss guardrail, the Explicit Human Request trigger and Draft & Review. Our page on human handoff explains the idea in more detail.

What do US cosmetics rules mean for your support replies?

The main one is that claims can change how a product is regulated. The FDA says whether a product is a cosmetic or a drug depends on its intended use, and that intended use can be shown by claims on the label, in advertising, on the internet or in other promotional materials.

The FDA gives examples on the same page. Claims that a product will restore hair growth, reduce cellulite or regenerate cells are drug-type claims, and the FDA says cleansers that are also intended to treat acne are regulated as drugs, or possibly as both drugs and cosmetics. Makeup and moisturisers marketed with sun protection are treated as both cosmetics and drugs. The FDA's page does not mention customer service replies, but a support message is written by your business, so the safe habit is to keep replies within the claims on your label and product page. If the label says "moisturises", the AI should not say "heals eczema". A Must Follow guardrail is a good place to write that down.

Ingredient questions have a related trap. The FDA's labelling guide says ingredients are listed in descending order of predominance, with some exceptions, and that fragrance compounds may be declared simply as "fragrance". So a customer asking "does this contain [ingredient]?" may not get a full answer from the label alone. If the label says "fragrance" and you do not know what it contains, say so, and do not guess.

This is a summary of FDA pages, not legal advice. Check your own claims with a regulatory adviser.

How do you answer shade-matching questions?

Give the AI a written shade guide and let it answer only from that. Without one, it has nothing reliable to match against, and it should say so and offer a person.

  • Upload a shade guide as a document: each shade's name, undertone and depth, and any comparisons you are happy to stand behind. Fyncall accepts PDF, Word, text and Markdown files up to 10 MB each.
  • Keep shade names in your Shopify variants consistent with the guide, so the AI can match them to products and stock.
  • If you do not publish comparisons with other brands, add a rule that the AI must not guess them.
  • Your team can send the right product or variant to the customer from the inbox, and build an order with a checkout link.

If customers ask about shades in Instagram comments, Instagram automations on the Pro plan can start a flow from a comment on a specific post or reel, a story reply or a story mention.

How do you handle returns of opened beauty products?

Decide your rule, write it down in plain words, and let the AI quote it rather than interpret it. Whether you accept opened products is your policy choice; the important thing is that customers hear the same answer on every channel.

Your returns document should cover:

  • Whether opened or used products can be returned, and on what terms.
  • What happens if the shade is wrong or the product arrives damaged.
  • What happens if a customer reports a reaction (hand over to a person, as above).
  • How long customers have, and how to start a return.

Fyncall's agent can look up the order, check whether it is eligible for a refund, and start a return request under your policy. It does not issue refunds: those stay with your team. For wording, see our return and refund reply templates, and for delivery questions the WISMO replies.

How do you set up Fyncall for a beauty store?

Install it from the Shopify App Store, upload your product and policy documents, set guardrails, and start one channel in Draft & Review.

  1. Install Fyncall. Products, variants, orders and customers sync from Shopify.
  2. Upload focused documents to the knowledge base, one topic per file: ingredient notes, usage directions, shade guide, returns policy and shipping policy.
  3. Set guardrails on each agent: Never Discuss medical advice, Must Follow rules about claims, and tool access.
  4. Add a reaction flow as described above.
  5. Connect your channels: WhatsApp, Instagram, email (including Gmail and Outlook) and the web chat widget.
  6. Start with Draft & Review on one channel, then move routine channels to Auto-Reply once you trust the answers.

The AI replies in the customer's own language, so you do not need a separate setup for each market you ship to.

Plan detailFreePro
Price$0$34 a month, or $370 a year
AI messages25 a month500 a month
Knowledge base files5Unlimited
AI workflows1Unlimited
Team membersUp to 2Unlimited
Guardrails and Draft & ReviewIncludedIncluded
Ticketing with SLAsPro onlyIncluded

Human replies are unlimited on both plans. A beauty store with ingredient notes, a shade guide and several policies will soon need more than five files, so check the pricing page against your own content.

Fyncall connects to Shopify, WhatsApp, Instagram, email and its own web chat. It does not offer Messenger, phone, SMS, or integrations with returns, reviews or email marketing apps. If you rely on those, factor that in. Other stores with similar questions: fashion and supplements.

Common questions

Should an AI ever recommend skincare for a customer's skin type?

Only by repeating what your own product pages say, such as "formulated for oily skin" if that is your claim. It should not judge whether a product suits a particular person's skin or condition. Send those questions to your team.

Do I have to report a customer's skin reaction to the FDA?

MoCRA requires the responsible person, the company named on the label, to report serious adverse events within 15 business days. Whether that is you depends on your label. Customers can also report problems to the FDA directly. This is not legal advice.

Can the AI read ingredient lists?

It can read what is in your Shopify product descriptions and the documents you upload. If ingredients only appear in product photos, add them as text.

Can the AI check where an order is?

Yes. It looks up orders and tracking in Shopify and gives an estimated delivery date. It can also update the address on an unfulfilled order after the customer confirms, if you allow that tool.

Where can I learn more about AI support in general?

See the AI support agent page and our AI customer service guide, which covers testing an AI in a sandbox before it talks to customers.

Ready to try Fyncall?

Install it tonight, let the AI draft your first replies, and switch it on one channel at a time as you start to trust it.

  • Free plan, no credit card
  • 14-day free trial on Pro
  • Unlimited human replies on every plan