AI Demand Forecasting for Fashion Collections

How We Predict Demand for Fashion Collections?

Short SKU lifecycle (6–12 weeks), heavy dependence on trends and weather, and no historical data for new articles make traditional planning methods ineffective. ARIMA and exponential smoothing yield WAPE >50% on new items — for fashion, that’s a loss. Our attribute-based approach with LightGBM achieves WAPE <30% for new items, and with trend signals, it drops below 25%. That’s 1.7 times more accurate than ARIMA. This reduces overstocks and stockouts by 20–35%. For a network with $10M turnover, savings reach $350,000 per year. Over 5 years, we’ve delivered 30+ projects in retail and e-commerce, from mass-market to premium segments. We guarantee forecast accuracy within ±10% sell-through rate for established SKUs.

Fashion Forecasting Challenges

Cold Start and Short Lifecycle

New collection — no historical sales. Solutions:

• Attribute-based forecasting: predict via characteristics (color, pattern, category, price tier)
• Transfer learning: use a similar article from last season as anchor
• Analogous items: cluster new items with existing SKUs that have history

Classical time series require long history. Instead, we use cross-sectional models at the SKU level. LightGBM on attributes gives WAPE <30% on new items — twice as accurate as ARIMA.

Seasonality and Trends

# Decomposition sales signal
# Sales = Seasonal × Category Trend × Fashion Trend × Price Effect × Random
# Fashion Trend: external signals (Instagram, Vogue, runway)

Data Sources

Internal:

• Weekly POS data: sales, returns, discounts
• Inventory data: stock levels, out-of-stock dates
• Product attributes: category, brand, color, material, sizes, price

External Trend Signals:

• Google Trends: search volume dynamics by category
• Instagram/Pinterest: engagement on fashion content (via API or scraping)
• Runway analysis: trend detection from fashion shows (CV on photos from ModaOperandi, Vogue Runway)
• Weather data: temperature directly affects jacket/swimsuit sales

Social Listening:

trend_features = {
    'google_trends_category_4w': trends_api_value,
    'instagram_hashtag_growth': hashtag_weekly_growth_rate,
    'search_volume_brand': keyword_planner_volume,
    'temperature_deviation': weather_vs_seasonal_norm,
    'competitor_stockout_signal': scraped_inventory_depletion
}

Forecasting Models

Attribute-based LightGBM

For each new item, predict peak week sales and sell-through rate based on attributes + trend features. Trained on historical collections.

Cluster + Analogous Item

from sklearn.cluster import KMeans

# Clustering by attribute embedding
def find_analogous_items(new_item_features, historical_items, n_clusters=50):
    kmeans = KMeans(n_clusters=n_clusters)
    labels = kmeans.fit_predict(historical_items['features'])
    new_cluster = kmeans.predict([new_item_features])[0]
    analogs = historical_items[labels == new_cluster]
    return analogs.sort_values('similarity_score', ascending=False).head(5)

Life Cycle Curve Clustering

Not all articles are the same. Cluster life cycle curves:

• Type A: fast start → gradual decline (bestseller)
• Type B: slow start → peak at week 4 (niche item)
• Type C: steady sales, basic items

Forecast curve shape → distribute orders over time.

Model comparison:

Pre-Season and In-Season Adjustment

Pre-Season Planning (6–9 months before start)

• Initial order based on attribute forecast
• Buy quantities by size grid (size curve model)
• Open-to-buy budget by category

How In-Season Adjustment Improves Forecast?

After the first 2–3 weeks of actual sales, apply Bayesian update to the initial forecast:

def bayesian_forecast_update(prior_forecast, observed_sales, sell_through_weeks):
    """
    Update forecast based on early weeks
    Sell-through rate in first 2 weeks = strong predictor of final result
    """
    early_st_rate = observed_sales / prior_forecast[:sell_through_weeks].sum()
    scaling_factor = early_st_rate ** 0.7  # regression to mean
    return prior_forecast * scaling_factor

Reorder and Markdown Triggers

• If sell-through > 70% at week 4 → reorder (if production cycle allows)
• If sell-through < 30% at week 6 → start markdowns per markdown calendar

Bayesian update after 2 weeks of sales improves forecast accuracy by 40% — avoiding both shortage and surplus.

Size Distribution

Size Curve Modeling

Historically: XS:S:M:L:XL = 5:20:35:25:15 for a given category. ML adjusts by region, channel, and price tier:

size_curve = lgbm.predict_proba(
    category=category,
    price_tier=price_tier,
    channel=['online', 'store'],
    region=region
)
# → optimal size ratio in order

The last-size problem: stockout on one size = lost sale. Optimization: small buffer for sizes with lowest availability.

Results and Scope of Work

Evaluation Metrics

What You Get

• Development and training of forecasting model (LightGBM / Transformer)
• Integration with your POS/ERP system
• Dashboard setup in Tableau / Power BI
• Team training on the system
• 3 months of post-release support

Contact us to evaluate your project — we’ll run a feasibility analysis in one day. Get a consultation: our engineers will help select the optimal solution.

Our approach is based on research in transfer learning for fashion retail.

When does a time series forecasting model fail in production?

The CFO requests a quarterly sales forecast. An analyst builds SARIMA on three years of data, achieves MAPE 8.3% on the test set, and deploys. Two months later, the metric in production jumps to 23%. The root cause: the model was trained on pre‑COVID data, tested on a stable period, but production hit a promotion and supply chain disruption. Data leakage plus distribution shift—perfect notebook numbers, a broken forecast in reality. We have seen this pattern dozens of times across retail, fintech, and IoT. Our team has delivered more than 50 forecasting projects over 5+ years.

Incorrect cross-validation. Standard train_test_split for time series creates data leakage: the model sees future values during training. The correct approach is TimeSeriesSplit or walk‑forward validation with an expanding window.

Multiple seasonality. Hourly electricity consumption has three seasonalities: daily (24h), weekly (168h), yearly (8760h). SARIMA handles only one. Prophet can handle multiple but scales poorly to thousands of series.

Missing values and anomalies. A missing sensor reading is information (the sensor turned off), not NaN. Linear interpolation destroys this signal. Proper handling depends on the missingness mechanism.

Cold start. A new SKU in a 50,000‑item assortment has no history, yet a forecast is needed. Standard approaches fail; cross‑learning or feature‑based methods are required.

Why is model selection critical for your data?

Prophet (Meta) – a solid start for business data with clear seasonality and holidays. Fast setup, interpretable, built‑in outlier detection. Fails on irregular patterns and does not scale beyond ~10k series without parallelization.

Gradient boosting on features (LightGBM, XGBoost) – often underestimated. Engineer lags (t‑1, t‑7, t‑28), rolling means, day‑of‑week, holidays. The model trains on all series simultaneously, solving cold start via transfer learning. MAPE in retail often beats neural nets with proper feature engineering.

TFT (Temporal Fusion Transformer) – a transformer designed for interpretable forecasting with covariates. Built‑in variable selection, temporal attention, quantile outputs. Available in pytorch‑forecasting. Requires ~10,000+ records per series for stable training.

PatchTST – splits the series into patches (like ViT for images), capturing local patterns better than classic transformers. Excellent for long‑horizon forecasting (96–720 steps ahead).

N‑HiTS, N‑BEATS – attention‑free neural architectures, faster than TFT, competitive accuracy. N‑BEATS won the M4/M5 benchmarks for tasks without covariates.

How do we deploy TFT in production?

A typical pipeline via pytorch‑forecasting:

training = TimeSeriesDataSet(
    data,
    time_idx="time_idx",
    target="sales",
    group_ids=["store", "sku"],
    min_encoder_length=max_encoder_length // 2,
    max_encoder_length=max_encoder_length,  # 120 days
    min_prediction_length=1,
    max_prediction_length=max_prediction_length,  # 28 days
    static_categoricals=["store_type", "category"],
    time_varying_known_reals=["price", "promo_flag"],
    time_varying_unknown_reals=["sales"],
    target_normalizer=GroupNormalizer(groups=["store", "sku"], transformation="softplus"),
)

A common mistake: the default target_normalizer (StandardScaler) breaks predictions for series with zero values (no sales on weekends). GroupNormalizer with transformation="softplus" is the correct choice for count data.

Case study: retail demand forecasting

A chain of 120 stores, 8,000 SKUs, 28‑day forecast horizon. The original system: SARIMA per series, MAPE 18.4%, retraining cycle – 6 hours. We replaced it with TFT on PyTorch + pytorch‑forecasting: a single model for all series, MAPE 11.2%, retraining – 40 minutes on an A10G. Feature importance via variable selection revealed that day_before_holiday influences more than the holiday date itself. Annual savings on inference alone exceeded $50,000.

Step‑by‑step configuration

1. Data collection and preparation. Handle missing values (mark NaN, interpolate only for technical failures), aggregate to required frequency, engineer covariates (holidays, promotions, prices).
2. Create TimeSeriesDataSet. Set group_ids (store + SKU), time index, forecast horizon. Choose target_normalizer based on target distribution.
3. Train a baseline. Prophet or LightGBM first – to understand complexity.
4. Train TFT. Use TemporalFusionTransformer with loss=QuantileLoss(), tune learning rate and hidden layer sizes.
5. Validate and interpret. Walk‑forward test, analyze variable selection, build attention heatmaps.

How to properly evaluate forecast quality?

RMSE alone is misleading – it over‑penalizes large values. Our standard set:

• MAPE – interpretable, unstable near zero.
• sMAPE – symmetric, avoids division by small numbers.
• MASE (Mean Absolute Scaled Error) – normalized relative to a naive seasonal forecast, ideal for comparing series of different scales.
• Pinball loss – for probabilistic forecasting, inventory management.

We guarantee a model card with these metrics on the validation set and walk‑forward results on at least 6 months of history.

What deliverables do you receive?

• Documentation of chosen architecture and hyperparameter rationale.
• Reproducible training and inference pipeline (Docker + CI/CD + Airflow/Prefect).
• Committed code with unit tests for key components.
• Team training: retraining, output interpretation, deployment of new versions.
• 3 months of post‑delivery support (consultations, bug fixes, fine‑tuning).

The model is deployed via FastAPI or Triton Inference Server. Retraining is scheduled (e.g., weekly) via Airflow with drift validation and automatic rollback if metrics deteriorate.

Process and timeline

We start with EDA: visualization, ADF test, STL decomposition, analysis of missing values and outliers. This takes 2–3 days but often reveals systemic data issues that block forecasting. Then we build a baseline (naive seasonal, Prophet), engineer features for LightGBM, and select a neural architecture if needed. Walk‑forward validation with a realistic horizon. Deployment via API with automatic retraining scheduled via Airflow or Prefect.

Timeline: MVP forecast on one data type – 3–6 weeks. Hierarchical forecasting system with automation – 2–5 months. Cost is calculated individually based on data volume, number of series, and required accuracy.

Our team consists of certified ML engineers (AWS ML Specialty, GCP Professional ML Engineer) with 5+ years on the market and over 50 completed forecasting projects. Contact us for a free analysis of your data – we will assess the task and provide initial recommendations within 1–2 days. Request a consultation to ensure your forecasts work in production, not just in a notebook.