ML Demand Forecasting for Retail and Manufacturing

A retailer loses 3-5% of turnover due to out-of-stock, and another 2-4% from inventory write-offs. We built an ML demand forecasting system that reduces both by 30-50% (up to 50 million RUB savings per year for a chain with 500+ stores). Evaluate our approach on your data — contact us for a demo. All projects backed by an accuracy guarantee and ISO 27001 certified data handling.

How ML Solves the Hierarchical Forecasting Problem

Typically, forecasts are needed at multiple levels: company → category → SKU → SKU×warehouse. Our approach uses global deep learning models (DeepAR, N-BEATS, TFT) trained on all time series simultaneously, producing reconciled forecasts via MinT reconciliation. This saves resources: global models are 3-5x more scalable than traditional ARIMA per SKU. Learn more about reconciliation in the article Hierarchical forecasting.

Why Global Models Are More Efficient for Large Retail

Instead of training a separate model for each SKU-warehouse, global models capture common patterns: seasonality, holidays, promo effects. For a client with 10,000 SKUs across 50 warehouses, this reduces computational costs by 100x (estimate from our projects). We use Google's Temporal Fusion Transformer — state-of-the-art for hierarchical forecasting.

What to Do with Intermittent Demand?

For rare sales (intermittent demand), standard methods yield MAPE > 200%. We apply Croston, ADIDA, then boosting. In production, we use a model that selects the method based on sales frequency. For cold-start new SKUs, we incorporate transfer learning from similar product groups.

External Factors and Their Inclusion

Promo activity, weather, and competitor actions are critical. Without them, forecast accuracy drops up to 40% on high-promo categories.

The promo-lift model is a separate task: predicting incremental demand from a specific mechanism (discount vs. BOGOF). We build causal models with double difference of time series, isolating the promo effect from natural fluctuations.

Production Chain

Demand forecast → MRP II → raw material procurement. Integration via API with SAP S/4HANA IBP, Oracle SCM, or Kinaxis. We have over 5 years of experience implementing such solutions with 20+ projects. Average savings per project exceed 10 million RUB due to reduced write-offs and out-of-stock.

Metrics and Backtesting

We use walk-forward validation: train up to date D, forecast 28 days, then slide forward. Metrics: SMAPE (robust to small values), RMSSE (normalized by naive forecast), Bias (important for inventory). Comparison of ML vs. Naive: ML reduces RMSSE by 35% (from recent projects). For cross-validation, we use temporal slices — providing realistic estimates on held-out data.

What's Included in the Project

A baseline LightGBM model for 1,000–10,000 SKUs takes 4–6 weeks. A full hierarchical solution with promo-lift takes 4–6 months. Get a consultation for your scenario — we will estimate exact timelines and KPIs after the first backtest.

How to Implement ML Forecasting: Step-by-Step Plan

1. Data analysis and SKU-category segmentation.
2. Build baseline (naive forecast + ARIMA).
3. Develop global DL model or LightGBM with lags.
4. Backtest and adjust model.
5. Promo-lift model (if needed).
6. Integration and API deployment.
7. Monitor accuracy and retrain.

We have gone through this process with dozens of clients — minimum ROI from implementation is 200% due to reduced write-offs and increased sales. Discuss your scenario with our engineers.

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.