AI assortment and allocation in beauty retail

Walk the beauty floor of any chain and count the facings. A single foundation franchise can carry 40 or more shades across three formats — full size, travel, refill — before you add the limited edition that lands twice a year. Multiply that by a prestige skincare wall, a fragrance bar, a colour cosmetics gondola, and you are not managing an assortment. You are managing a combinatorial explosion that the buying team still tries to hold in a spreadsheet.

This is what makes beauty different from apparel or grocery. The unit of decision is not the product — it is the shade, in this format, in this door, this season. And the count of those units runs into the hundreds of thousands across a network that ranges from flagship to small-format kiosk. All of that before anyone has decided a single allocation.

Manual assortment breaks here for a structural reason, not a competence one. The merchandiser who can intuit a 200-line apparel buy cannot hold a 12,000-line shade-by-door grid in their head. So they collapse it: a single allocation curve applied to all doors, a tier rule applied as a filter, a tester budget set as a flat percentage. Every collapse trades precision for survivability — and every collapse is where the margin leaks.

The combinatorial scale

Beauty's SKU count is not large. It is multiplicative. Each axis would be manageable alone; stacked, they produce a space no human or spreadsheet can reason about line by line.

Start with the axes that define a single franchise:

• Shades. A foundation or lipstick range routinely spans 30 to 50 shades, with demand wildly uneven across them — a handful of core shades drive most of the volume, the rest form a long tail that still has to be carried to signal a complete range.
• Formats. Full size, travel, mini, refill, gift-set exclusive. Each format has its own demand curve, its own margin, and its own role in the basket.
• Limited editions. Holiday, collaboration, seasonal — high-velocity, short-lived, and impossible to replenish once gone.
• Prestige tiers. Mass, masstige, premium, prestige. The tier governs not just price but which doors are even allowed to carry the brand.

Now multiply by the store network. A chain with 80 doors does not run 80 copies of one assortment. It runs flagships that carry the full range, mid-format stores that carry the core plus a curated tail, and small-format doors that carry only the heroes. The decision space is shades × formats × tiers × door archetypes — and it changes every season as collections rotate.

A 50-shade range across 3 formats, allocated across 4 door archetypes, is 600 cells before you touch a single limited edition. One franchise. A beauty retailer carries dozens. The spreadsheet does not survive this, and everyone running one knows it. So the real assortment becomes a set of blunt rules applied uniformly, then patched by hand where someone noticed the damage.

Why allocation by store tier is the lever

The buy total gets the committee's attention. Store-level allocation is where the money is actually made or lost. The same units, bought correctly, produce a strong season in one allocation and a markdown pile in another. Beauty makes this sharper than most verticals because of the long-tail-versus-halo tension.

Carry too much of the long tail in the wrong door, and you tie up cash in shades that will never sell there. A deep-coverage foundation shade that moves in an urban flagship sits dead in a suburban small-format door for a season, then gets marked down or transferred at a loss. The tail is not free inventory — it is committed cash with a low probability of sell-through outside a narrow set of doors.

Carry too little, and you lose the complete-range halo. A foundation wall missing the shade extension reads as a brand that does not serve every customer. The shopper who cannot find their shade does not buy the core shade instead. They leave. The halo of a complete range lifts the heroes; an incomplete range depresses them.

This is the trap a flat allocation curve cannot escape. It over-carries the tail in small doors and under-carries it in the flagships that need the full statement.

The lever, then, is archetype-aware allocation: each door gets the depth that matches its real demand shape, not a fraction of a single global curve. That is exactly the discipline we argue for in the retail allocation engine — the season is decided at allocation, and replenishment only operates inside the box allocation set. Beauty raises the stakes because the box has thousands more cells, and the cost of a wrong cell is a dead shade rather than a slightly-off size run.

Tester and sample economics and prestige-tier rules as first-class constraints

Two beauty-specific realities sit outside every generic assortment model, and both are decisive.

The first is tester and sample economics. Testers are not marketing fluff — they are a hard cost line that drives trial and conversion. A fragrance counter without testers converts a fraction of the foot traffic of one with them. But testers consume real units, carry their own breakage and theft rates, and have to be allocated per door against a budget.

The decision "which doors get testers for which SKUs, at what depth" is an allocation problem with its own economics. And it is tightly coupled to the sellable assortment: a tester with no stock behind it is wasted, and stock with no tester underperforms.

The second is prestige-tier rules. Brands enforce where they will and will not be sold. A prestige house may grant a chain its line only in flagship and premium-format doors, with explicit exclusions, minimum-presentation standards, and planogram tiers that dictate facing counts. These are not preferences a model can optimize away. They are contractual constraints — break one and the brand pulls the line.

Here is the failure mode that mirrors what we see across retail. A model generates a "mathematically optimal" allocation, then a filter strips out everything that violates a tier rule or blows the tester budget. What is left is not a coherent plan — it is a remainder. The merchandiser sees instantly that the system does not understand how beauty actually trades, and goes back to the spreadsheet.

The architectural point is the one we make in AI assortment planning: business rules must be inputs the optimization sees, not filters applied to its output. When the tester budget, the prestige exclusivity, and the planogram tier are inputs, the allocation is executable by construction. When they are after-the-fact filters, the plan degrades into something no category manager will sign.

What a decision layer does for beauty assortment and allocation

A decision layer is the part of the stack that turns the demand signal and the chain's own rules into an executed allocation, shade by shade, door by door. Not a recommendation someone has to manually reconcile against reality.

For beauty specifically, that means four things the spreadsheet cannot do:

• It reasons over the full combinatorial space. Shades × formats × tiers × door archetypes, scored per cell against local demand shape — not one global curve fanned out uniformly.
• It carries the rules as constraints. Prestige exclusivity, minimum-presentation standards, planogram facing counts and tester budgets are inputs the optimization respects, so the output is allocatable on day one.
• It separates the hero from the halo. Core shades get depth tuned to velocity; the long tail gets carried where it earns its halo and trimmed where it only ties up cash.
• It stays live. A new prestige agreement, a re-tiered door, a tester budget cut — each updates the next allocation without an IT release or a rebuilt spreadsheet.

The merchandising conversation changes as a result. The category manager stops arguing line by line about a grid they cannot fully see, and starts arguing about the constraints. Whether the tester budget should rise in conversion-sensitive doors. Whether a small-format cluster should be re-tiered to carry more tail. Whether a prestige exclusivity is worth the door restriction it imposes.

Those are the decisions that move a beauty season. The shade-by-door numbers fall out once the constraints are explicit.

This is the same logic Solya is built around, applied to beauty's particular shape. Not an assortment tool with a tab for business rules, but a decision layer where prestige tiers, planogram standards and tester economics are first-class inputs. The allocation a category manager receives is one they can execute. And one the downstream replenishment and transfer logic can operate on coherently.

The question to sit with

If you run beauty assortment, ask what happens when your team rejects an allocation. If the answer is "the forecast is off", the conversation is about model accuracy. If the answer is "it would break our prestige agreement" or "it ignores the tester budget", the conversation is about architecture. And no amount of forecast tuning closes that gap.

Beauty is, at its core, an assortment-and-allocation problem at a scale no spreadsheet survives. The chains that solve it are the ones that stop treating their hardest constraints as afterthoughts and start treating them as the first thing the system has to know.

---

Is your beauty assortment surviving its own scale?

At Solya, we offer beauty merchandising and category leaders a personalized 30-minute diagnostic. We assess, on your own franchise mix and door network, where the shade-by-door allocation leaks cash to dead tail and missed halo. And identify which constraints need to move from filter to first-class input.

You'll walk away with:

• A read on where your current allocation over-carries the long tail and under-carries the complete-range halo
• A map of the tester-budget and prestige-tier rules your model treats as filters rather than inputs
• The first architectural changes to make before the next collection rotation