AI Pricing Playbook for Direct-to-Manufacturer

The number that matters in Atorie's story is not the $9.5 million seed. It is the move from roughly $5 million in 2025 sales to a projected annualized run rate above $55 million in 2026. The seed round and 40+ factory network are supporting proof of a direct-manufacturer model. Atorie aims to sell Italian-leather and cashmere goods at roughly 10% to 20% of typical retail prices, and it uses AI for trend analysis, color testing, demand forecasting, and material-shortage anticipation. The 2026 figure is a projection, not confirmed full-year revenue.

The useful lesson is that price can become the core product when the operator knows the cost floor, the demand signal, and the inventory risk before the item ships. Atorie's disclosed inputs are a $9.5m seed, 40+ factories, a 10%-20% retail target, AI trend and demand tools, and a consumer agent. The price-ownership playbook is separate from those disclosures: it turns cost, demand, and inventory risk into a weekly pricing loop.

Why price is the real product

When the cost model is built from factory price, freight, duties, quality loss, payment terms, and minimum order quantities, price stops being a category guess. The goal is not to publish the cost. It is to know the floor well enough to set a price band that can survive demand uncertainty. If the landed cost is unknown by SKU, color, and material, the first offer can look like a promotion even when it is the only real offer.

If your price is a guess, your margin is a lottery ticket.

The Price-Ownership Stack

Map the factory cost floor

The cost sheet is where the model either works or does not. Include raw material, labor, trim, packaging, freight, duties, returns, and write-down risk. For fashion, material is not a single number. A cashmere sweater in charcoal and one in a seasonal color can have different cost, lead time, and demand. As a hypothetical example, if a jacket's landed cost is $40 and your target price-to-cost ratio is 3.0, the floor is $120. If the forecast is weak, do not lower the price to $99 and call it a test. Either improve the forecast, reduce the cost, or cut the SKU.

Forecast by SKU, color, and material

At SKU-level operations, aggregate forecasts hide the problems. A category can look healthy while one color sits in the warehouse. Use trend analysis and historical sell-through to estimate demand by SKU, color, and material. Where possible, add leading indicators: search behavior, pre-orders, waitlists, and material shortage signals. Atorie's disclosed AI inputs include color testing and material-shortage anticipation, which is the kind of input that belongs in this step. The output should be a confidence range, not a single number. A high-confidence SKU can carry a tighter price band. A low-confidence SKU needs a smaller initial order, a faster review, and a clearer exit rule.

Set price bands from confidence

Price bands are more useful than single prices. A band gives you room to test without breaking the margin story. The trade-off is less initial volume in exchange for less write-down risk. As a hypothetical example, a $120 cost floor might support a $300 to $360 retail band if demand confidence is high. If confidence is low, the band may start at $300 with a smaller order and a 30-day review. The key is to decide in advance what price change means: a demand test, a clearance, or a correction to a bad cost model.

Hypothetical worked example: a $1,000 cashmere item

Assume a cashmere item has a typical retail price of $1,000. Atorie's stated 10%-20% target implies a $100-$200 price. If the landed cost is $40, the price-to-cost ratio is 2.5x at $100, 5.0x at $200, 3.0x at $120, and 4.5x at $180. A workable band might be $120-$180: $120 covers a low-confidence demand case, $180 leaves room for a high-confidence case, and the midpoint gives a clear test point. The band only works if returns, freight, duties, and write-down risk are already in the $40 landed cost.

Use AI outfit tests

A consumer-facing AI agent can be framed as a discovery feature. In a pricing system, it is a measurement tool. Atorie's consumer agent generates outfit recommendations; in a price-ownership model, that output can be used to test price elasticity, not just to generate looks. As a hypothetical example, if an outfit recommendation pairs a $320 coat with a $180 sweater and the customer accepts the bundle, you have evidence that the price point works in context. If the customer drops the coat but keeps the sweater, the coat may be the weak price, not the weak product.

Review margin leakage weekly

Margin leakage is the gap between the price you planned and the price you actually earned after discounts, returns, and inventory write-downs. Track it weekly against a target price-to-cost ratio. If a SKU is leaking margin, ask which input failed: cost, demand, price, or inventory. Do not let the weekly review become a blame session. It should be a correction loop.

Operating cadence and failure modes

The cadence matters more than the model. A weekly review should answer three questions: Which SKUs are above target price-to-cost? Which SKUs are below, and why? Which SKUs should be cut, repriced, or reordered? A monthly review should update the cost and demand models. A quarterly review should decide which factory relationships, materials, and categories are worth keeping.

Failure mode: cost floor is stale. Freight, duties, or material prices moved, but the price did not.
Failure mode: forecast is too aggregate. The category looks fine, but one color is a dead asset.
Failure mode: price tests are not labeled. A discount becomes a permanent price, and the margin story collapses.
Failure mode: AI recommendations are used only for conversion, not for price evidence.

First-week action: pick one SKU, calculate landed cost, set a confidence range, choose a price band, and schedule a 30-day margin leakage review. The review should compare planned price-to-cost ratio against actual earned price after discounts, returns, and write-downs, then decide whether to hold, reprice, or cut.