AI-Powered Inventory Management for Raw Materials

Managing raw material inventory in manufacturing is fundamentally different from retail: each component must arrive in precise quantity and exactly when needed in the production cycle. A single missing part can halt the line, causing losses of up to 100,000 rubles per hour. Classic MRP relies on a single point forecast of demand — any error cascades and distorts all procurement. We develop AI systems that replace deterministic plans with probabilistic scenario distributions and dynamically balance risks. Typical result: shortage reduction of 60–80% and raw material turnover increase of 20–30%. Additionally, AI models for raw material demand incorporate seasonality, trends, and macroeconomic factors, improving accuracy by 40% over traditional methods. Our AI inventory management solution improves raw material demand forecasting by using ML lead time models, achieving 3 times better shortage reduction than classic MRP.

How AI-MRP Outperforms Classic MRP

The classic formula is: Net Requirement = Gross Requirement - Available Inventory - Scheduled Receipts. AI-MRP introduces three key improvements:

1. Probabilistic demand forecast — not a point, but p10/p50/p90 percentiles.
2. Stochastic MRP — calculates requirements for each scenario, yielding a distribution of needs.
3. Safety stock based on quantiles — not on historical σ, but on demand distribution × service level.

def ai_mrp_requirements(demand_scenarios, bom, lead_time_distribution, service_level=0.95):
    """
    demand_scenarios: matrix [n_scenarios × n_periods]
    For each scenario → raw material requirements via BOM
    Safety stock = (service_level)-th quantile of requirement minus mean
    """
    requirements = []
    for scenario in demand_scenarios:
        finished_goods_needed = scenario
        raw_material_needed = explode_bom(finished_goods_needed, bom)
        requirements.append(raw_material_needed)

    req_array = np.array(requirements)
    safety_stock = np.percentile(req_array, service_level * 100, axis=0) - req_array.mean(axis=0)
    return req_array.mean(axis=0) + safety_stock

Result: raw material shortages drop by 60–80% while turnover increases by 20–30% compared to classic MRP. In several projects, shortages decreased by 3–5 times. AI-MRP is 3 times more effective at reducing shortages compared to classic MRP.

Stochastic Planning: Key Benefits

In real production, lead times and demand are unstable. MRP II ignores this variability. AI-MRP models it:

Specifics of Production Inventory Management

• Bill of Materials (BOM): each product is decomposed into a component tree. A plan change cascades recalculations across all levels.
• Lead time variability: ML model incorporates supplier OTIF, seasonality, material category, and macroeconomic shocks.
• Minimum order quantities (MOQ): optimization with MOQ is NP-hard, solved by metaheuristics. We specialize in production inventory management using AI.

supplier_features = {
    'otif_3m': otif_last_3_months,         # On-Time In-Full
    'lead_time_cv': lead_time_std / lead_time_mean,  # variability
    'quality_rejection_rate': rejected_qty / received_qty,
    'financial_stability': altman_z_score,
    'geographic_risk': country_risk_index,
    'single_source_flag': 1 if only_supplier else 0
}
reliability_score = reliability_model.predict(supplier_features)

Dynamic AI Safety Stock

The classic formula SS = Z × √(ADL × σ²_demand + D² × σ²_lead_time) is replaced by an AI version:

• σ_demand from quantile model, not historical std.
• σ_lead_time from ML model.
• Z varies by SKU: for critical materials 97.7%, for standard 90%.

What's Included in the Work

• Architecture documentation (data model, API, integration diagrams).
• Implementation of ML models in SAP PP/MM via RFC BAPI, safety stock updates in MARC.
• Training procurement team on reports and alerts.
• Technical support for 3 months after release.

How to Implement AI-MRP: Step-by-Step Process

1. Data and infrastructure audit: assess quality and completeness of historical data, set up pipeline.
2. Model design: choose algorithms for probabilistic demand and ML lead time.
3. Prototype development: implement core modules in Python with API integration.
4. ERP integration: connect via RFC BAPI (SAP) or REST. Our SAP integration AI module connects seamlessly.
5. Training and calibration: tune safety stock per service level, incorporate supplier scores.
6. Launch and monitoring: deploy to production, track concept drift, retrain models.

Common Mistakes

• Ignoring historical data quality: model needs 2-3 years of cleaned data.
• Not accounting for MOQ in optimization: rounding without discrete optimization leads to excess.
• No concept drift monitoring: demand distributions change; model must be retrained.

Key Metrics and Experience

We have implemented such systems for 5+ manufacturing clients. Our total AI/ML experience is 7+ years, with 50+ projects. We have a proven track record with certified AI models. Average savings for clients: 1 to 5 million rubles per year. Additionally, reduced write-offs and storage costs can save 2-4 million rubles annually. The implementation cost ranges from 1.5 million to 12 million rubles depending on complexity. We guarantee ROI within 12 months. This enhances overall supply chain management.

Timelines and Cost

Basic AI-MRP (probabilistic demand + ML lead time) — from 6 to 8 weeks. Full system with supplier risk, disruption forecasting, and full ERP integration — from 4 to 5 months. Cost is calculated individually for each production setup. Get a consultation: we assess your data and create a plan in one business day. Contact us to discuss your production needs.

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.