A vertical pack changes what Refik asks, what it refuses to claim and which product fields it reads. Only packs that exist in the code are listed here.
One beauty packOne neutral fallbackOne base rule set every store gets
What a pack actually changes.
A pack is a data file. It changes six concrete things, and the console shows you every one of them after setup.
The questions it asks a shopper
The beauty pack carries three questions and fourteen answers. Eight answers point at a tag in your own catalogue; the other six carry a label only, on purpose. A pack that points at a tag outside your catalogue is rejected when it loads, so every live answer option has products behind it.
The sentences it refuses
Twenty-four claim rules, read against the draft answer before the shopper sees it. Three of them apply in one market only — a phrase a regulator forbids in one country is the name of a real collection in another, so the rule carries the market it belongs to and fires only there.
The questions it hands to you
Thirteen patterns that mark a question as one for a person: pregnancy, a named skin condition, a prescription, a procedure someone had last week. Refik says so on the spot and offers to pass the conversation to you, with everything the shopper wrote attached.
Which tools the model is even offered
Three switches: shade matching, remembering a returning shopper’s profile, and noticing a product that is running out. The beauty pack opens all three. The neutral pack closes all three, and a closed tool is absent from the list the model is handed, so the model can only offer what the pack opened.
What your product fields are called
Forty-six product roles and twenty-eight needs. These are fallback names only: where your own catalogue already names a tag, your name wins. A pack tag that matches zero products in your catalogue is counted and shown in the console with the number it matched, so you can see which of the names your catalogue answers to.
The words in your storefront
Forty-six storefront strings — the greeting, the example questions, the survey label, the entry cards. They are written to app-owned metadata when you confirm the pack. Where a pack has an empty storefront block the generic wording stays, and the same write clears the text it replaces.
All six describe a file, and every number above was counted from that file. A wrong one is wrong about something you can be shown.
The packs that exist.
Two, and one of them is the fallback. This is the list, in full.
Ships today
Beauty
Hair care and makeup, and with them skin care, beard and shaving care, and fragrance. The advisor description Refik proposes for this pack comes out of the pack itself; nobody retypes it per store.
Product roles
46
Needs
28
Survey questions
3
Claim rules
24
Handover patterns
13
Ingredient terms
39
Alternative spellings
50
Replenishment intervals
13
Storefront strings
46
The alternative spellings are why a shopper who types the Spanish or Turkish word for a shampoo, a mask or a foundation is understood while your catalogue stays in its own language. The replenishment intervals are in days, per role, which is why a shampoo and a beard oil come up on different schedules.
Zero survey questions, zero roles, zero needs, an empty storefront block, and all three optional tools closed. Refik introduces itself as a general product advisor, and its wording stays at that level of claim.
It is the smaller claim: Refik still answers from your catalogue, your stock and your policies, it still refuses to assert an absence, and it still carries the base rule set below. What it leaves out is the three survey questions written for somebody else’s industry.
Two situations bring a store here — the catalogue is still being read, or it was read and every pack stayed under the three gates — and the console names which one.
The registry holds two packs today. What we can say is the cost of a third, because the answering code was rebuilt to stay put while a pack is added: one data file, one entry in the registry, and the list of words that would identify that industry in a catalogue.
Until that file exists, a store outside beauty gets the neutral pack. This page lists an industry on the day its file lands, so every pack row here is a file you can be shown.
An industry without a dedicated pack still gets a page, and that page says so in the first screen. What those pages describe is the neutral pack in that trade — which of the twelve tools answer there, which base rules fire, and what a dedicated pack would add. They carry the sentence “No dedicated pack is built for this sector yet” where you can read it, not in a footnote. There are 10 of them today, and none of them is a pack:
Every store carries the base rule set, whatever it sells.
This was measured. The neutral pack shipped with an empty rule list, which made the store we knew least about the one answering with the fewest rules in front of it — and an empty list measures our knowledge of a store where a full one measures its risk. Sixteen rules and nine patterns were written for that gap.
Every store gets, whatever it sells
16 claim rules and 9 handover patterns, in English, Turkish and Spanish
Treating, curing or healing a condition — the line that turns a product into a medicine
Standing in for a medicine, a prescription or a doctor
Preventing a disease, or killing germs, viruses or bacteria
Proof language — clinically, scientifically, lab-tested — where a document would be required
Promising that a result is certain, or that it is total
Absolute safety — the pack’s own words for it: “no side effects”, “causes no reaction”, “safe for everyone”
A legal or tax assurance beyond what a store can give, such as customs or duty being waived
Lifting an age check, or calling an age-restricted product harmless
Pregnancy, allergy, a chronic condition or a current medication — handed to you with the conversation
What the base set does with a pack’s own rules
Sits behind them. The base list is appended after the pack’s, so a line caught by both is reported with the pack’s rule first.
Stays quiet where the pack already spoke. A base rule whose match overlaps the words a pack rule reported is skipped, so one line produces one report.
Covers the sentences any store could get wrong, which is why it applies to a store of any trade, including one we have room to place later.
Leaves market exceptions to the pack. A rule true in one country only lives in a pack file, where it carries that market with it.
Reads the assistant’s draft. Your own product pages stay exactly as you wrote them.
The merge happens inside the one function that loads a pack, so every store that has a pack has the set. An empty set makes the app refuse the pack outright at load time — because an empty rule set is the exact state this base set was added to end.
How the pack gets chosen.
Refik measures your catalogue and tells you what it found. You are the one who acts on that measurement, and it stays a suggestion until you confirm it.
You choose it
The first screen of setup is called What Refik advises on. The pack is written once you confirm it, together with the advisor description — and that description is the pack’s own sentence, shown to you to accept or rewrite before it is saved.
The catalogue is measured separately
Every pack has a list of words. Refik counts how many of your products carry any of them and shows you the count, the ranked score for each pack, up to eight of the words it actually matched, and up to three example products for each word. You see the working, and the verdict stays yours.
Three gates, each measured on its own
A pack is only suggested if the share of matching products clears a threshold, at least three different words matched, and at least five products matched. One word appearing in one collection stays one word, and a store with two products produces a ratio too thin to read.
Your choice wins, and it asks once
Pick a pack the measurement left out and it is still your store. It asks you to confirm, and the server checks that confirmation by running the measurement again; the one thing it reads from the page is your intent.
It can be changed afterwards
The pack is editable later from the Configuration screen by an account administrator. The change is written to your account’s audit record in the same database batch, a second open tab is told the value moved, and the product knowledge is queued to be rebuilt for the new pack while the old tags stay in place — which the screen tells you, so your catalogue keeps answering right through the rebuild.
The measurement reads your own tags, titles and collections, and skips the tags a pack itself produced. Scoring a catalogue the beauty pack has already labelled as beauty would make the answer an input to the question, and the resulting confidence would be a number about our own labelling.
Five outcomes, and the console names the one you got.
Four of the five say the measurement stopped, and each of the four stops for its own reason. The console prints that reason by name, because “the catalogue is still loading” and “the score stayed under the threshold” are different problems with different fixes.
Catalogue still loading
A setup step is still running, so the score has no input to work from. The neutral pack here is a waiting room and the console says so by name: the reason belongs to the step, and your products come into it once they are read.
Too few products
The catalogue is small enough that any ratio taken from it is a rounding error. Two products out of two describes two products, and treating it as an industry would put a whole store behind a pack chosen by that rounding.
Below the threshold
The highest-scoring pack covered less of the catalogue than the threshold asks for. That threshold sits far below what a real store in that industry scores, because the two mistakes cost different amounts: sending a beauty store to the neutral pack costs you one choice on one screen, while telling a furniture store it is a hair care advisor costs a shopper a wrong recommendation and a wrong refusal.
Too few distinct words
The share of matching products was high enough, but it rested on one or two words. A single recurring word — in an all-products collection, say — can carry a whole catalogue over a ratio while describing one collection.
A match
All three gates cleared, and the pack is put in front of you as a suggestion with its score and its matched words. It is still a suggestion, and your confirmation is the thing that writes it.
The thresholds were measured on three catalogues. In two real beauty catalogues, 139 of 142 products and 138 of 141 carried at least one of the pack’s words; in an unrelated catalogue of furniture and electronics the count was zero out of ten. The separation is wide, and the threshold sits nearer the low end of it on purpose.
Questions this page should answer.
Including the ones with a one-word answer.
Is my industry supported?
If you sell hair care, makeup, skin care, beard and shaving care or fragrance, there is a pack for it. Everything else gets the neutral pack. That is the whole answer: this page lists two packs, and both of them have a file behind them.
What does the neutral pack cost me?
The plan price is the same under every pack: the billing and quota code reads your plan and your usage, and a pack name appears nowhere in it. What it costs is specificity — the survey, shade matching, the remembered shopper profile and the replenishment reminder are the four things a pack opens, and the generic storefront wording stands in for wording written for an industry. Everything on the surfaces page still works.
What if I pick the wrong one?
Change it from the Configuration screen. Refik then queues a rebuild of the product knowledge against the new pack, keeps the old tags until that finishes, and says on screen that it is doing so. One case is deliberately refused there: a store set up before the App Store listing existed already has live storefront text coming from its pack, and that screen leaves such a storefront alone, because the merchant would be editing it blind.
Will you add a pack for my industry?
What we can tell you is the shape of the work, because the answering code was moved out of the way precisely so that it would stay put: a data file, one entry in the registry, and the words that would identify that industry in a catalogue. When one exists it appears on this page; today this page says two.
Does the pack decide what Refik recommends?
Your catalogue does. The pack supplies names and rules; your catalogue supplies the products. Every option Refik offers has passed two gates in your own store — sellable, and in stock — and a pack tag matching zero products is reported to you with the number it matched. The products a shopper sees are the ones you carry today.
Does a pack let Refik say more?
It lets it say less, more precisely. A pack only ever adds to the base set, so the longest list inside a pack is the list of sentences it refuses. The part of a pack that makes the assistant sound like it knows your trade is the part that stops it from claiming things about your trade.
If something on this page turns out to be wrong, it is wrong about a file you can be shown. That is the only kind of claim this site is willing to make about industries.