Fashion Simulator

Simulate first,
then order

Four simulations along the collection. The initial values are examples: set the sliders to the scale of your company. The calculation paths sit beneath each simulation.

From analysis to a testable simulation

Every collection starts with a bet: the pre-order decision is made months before the first customer sees the product. Three stages turn the bet into a calculation.

Stage 1
Analysis
Looks back and shows where the bet has cost money so far: in overstock running into markdowns and in sold-out styles.
Stage 2
Forecast
A model per style and market estimates demand before the order, with a stated error instead of gut feeling.
Stage 3
Simulation
You change a control variable and see what the intervention would lead to, before the order is placed.
Control

The fourth stage happens inside the company: after introduction, the simulated trajectory is compared against sell-through, returns and markdowns of the running season. What the measurement refutes gets replaced.

The pre-order quantity and the price of forecast error

Order too much and it runs into the markdown, order too little and demand goes unserved. The simulation weighs both error costs against each other and shows what a smaller forecast error per style is worth.

Every slider is an assumption, not a fact. All initial values are fictional examples: no customer data, no industry figures, not the values of the sample company from the six modules.

Expected demand per style
Forecast error today
Forecast error with model
Selling price
Purchase cost
Salvage value per unsold unit
Expected contribution margin over the order quantity: markers at both optima
Optimal quantity
today → with model
Contribution margin at the optimum
today → with model
Value of the error reduction
per style and season
Capital tied up in the pre-order
today → with model

Calculation: demand is normally distributed with the set mean and relative spread (assumption). Under-ordering costs selling price minus purchase cost, over-ordering purchase cost minus salvage value. Expected contribution margin per order quantity via the loss function of the normal distribution; the optimum is the maximum of this curve. One season per style, no replenishment. The calculation is meaningful as long as the salvage value sits below the purchase cost and that below the selling price.

Control

After introduction, the forecast error per style is measured on held-back seasons, and the ordered quantity is compared against actual sell-through.

Season contribution margin as a distribution

The same pre-order decision as a distribution instead of a point: 500 simulated seasons per scenario. The random number generator starts from a fixed seed, so every visit shows the same result.

Styles in the assortment
Mean demand per style
Forecast error scenario A
Forecast error scenario B
Median scenario A
season contribution margin
Median scenario B
season contribution margin
B ahead
out of 100 seasons each
Distribution of the season contribution margin across 500 seasons per scenario

Calculation: demand per style normally distributed, truncated at zero; order quantity per style equal to the forecast, without a safety margin (assumption). Selling price, purchase cost and salvage value come from simulation 01. 500 seasons per scenario, fixed random seed, independent random streams per scenario, so the metric compares self-contained runs.

Control

At the end of the season the measured contribution margin sits inside the band or outside it, and the deviation is reported, not explained away.

What an avoided return is worth

The simplest calculation on this page: returns, addressable share, achievable reduction and the savings per year.

Online orders per year
Return rate
Cost per return
Addressable share of returns
Achievable reduction within the addressable share
Avoided returns per year
Savings per year
Return rate
before → after
Savings per year as a function of the achievable reduction, with a marker at the set value

Calculation: returns = orders × rate. Avoided returns = returns × addressable share × reduction. Savings = avoided returns × cost per return. The simulation does not split by return reason and claims no cause shares, so the addressable share and the reduction are your assumptions.

Control

The return rate is measured per category and order cohort before and after the measure; the assumed reduction is replaced by the measured one.

Markdown: early and shallow versus late and deep

The old merchandising dispute as a calculation instead of an opinion. You set the sell-through assumptions of both scenarios yourself; no relationship between markdown depth and demand is claimed, because it would only emerge from your data.

Remaining stock at clearance start
Scenario A: early and shallow
Markdown from week
Markdown depth
Sell-through rate under markdown
Scenario B: late and deep
Markdown from week
Markdown depth
Sell-through rate under markdown
Cumulative revenue over 14 weeks: markers at both markdown start weeks
Revenue scenario A
incl. salvage value after 14 weeks
Revenue scenario B
incl. salvage value after 14 weeks
Difference
A minus B
Remaining units after 14 weeks
A → B

Calculation: before the markdown, each week sells 6 % of the then-remaining stock at full price (fixed value); from the start week onwards the set sell-through rate applies at the reduced price. Revenue = sum of weekly sales × price plus remaining units × salvage value. Selling price and salvage value come from simulation 01. Horizon 14 weeks. The chart shows cumulative sales revenue without the salvage value.

Control

For each markdown step the actual weekly sell-through rate is measured; the assumption yields to the measured curve, and the next season calculates with it.

What this page assumes, fully disclosed

Every number on this page follows arithmetically from the slider values. In addition:

  • Example values: all initial values are fictional. They are not customer data, not industry benchmarks and not the values of the sample company from the six modules.
  • Normal distribution: demand per style is normally distributed (truncated at zero in simulation 02), with a relative spread of at most 50 %. Real fashion demand has heavier tails, so the band understates extreme seasons.
  • Order quantity in simulation 02: per style equal to the forecast, without a safety margin. Fixed random seed, 500 seasons per scenario, independent random streams per scenario.
  • Returns: no cause split. The simulation claims no shares of size-related or other return reasons; addressable share and reduction are pure assumptions.
  • Markdown: the sell-through rates of both scenarios are your assumptions; no empirical relationship between markdown depth and demand is claimed. A full-price rate of 6 % per week and a horizon of 14 weeks are fixed values.
  • Not additive: the four simulations overlap in subject, so their results must not be summed, and this page does not sum them.
  • Nominal revenue: without process costs beyond those named, without discounting, without VAT.
Note on the context of this portfolio

This page is a basis for conversation, not an offer. All values are model calculations based on freely adjustable assumptions; no number is a commitment, a forecast for a specific company or an industry figure.

In a real project your data replaces the sliders: your sell-through history provides forecast errors and sell-through rates, your returns data the addressable share. The calculation paths stay the same.

All 6 modules: AI in fashion