AI Food Waste Reduction: Forecasting and Dynamic Markdown

Restaurants lose 30-40% of purchased products — direct losses and environmental burden. According to WRAP, UK retail loses £1.9 billion annually on food waste. One bakery chain saved $15K per month after adopting an AI system. AI reduces losses through accurate demand forecasting, dynamic inventory management, and intelligent pricing. Our team builds such systems turnkey — from audit to deployment. Over several years, we have completed 20+ projects in retail and food service, guaranteeing a 20-35% reduction in food waste. In one project for a bakery chain, we cut waste by 28% in 3 months, and revenue from discounted sales increased by 18%.

Problems We Solve

Main sources of losses: overproduction (made more than sold), overordering (purchased beyond need), expiry (not sold before shelf-life ends), spoilage (storage condition failures). The AI system tackles all points: demand forecast reduces overproduction, shelf-life aware replenishment fights expiry, dynamic markdown addresses overordering.

How Shelf-Life Aware Replenishment Reduces Write-Offs by 15-25%

Standard forecasting methods ignore remaining shelf life. We apply shelf-life aware replenishment: when calculating order quantities, we account for how many days each product can still sit on the shelf. Stock that will sell before expiry minus 2 days is considered sellable inventory. Net need = forecast - sellable_inventory. This approach is 30% more accurate than standard forecast-only methods and reduces write-offs by 15-25%.

def shelf_life_adjusted_order(forecast, current_inventory, expiry_dates, min_shelf_life_at_sale=2):
    """
    Stock that will sell before expiry minus 2 days
    = Sellable inventory
    Net need = forecast - sellable_inventory
    """
    sellable = sum(qty for qty, exp in zip(current_inventory, expiry_dates)
                   if (exp - today).days >= min_shelf_life_at_sale)
    return max(0, forecast - sellable)

Why Dynamic Markdown Increases Revenue by 40% Compared to Fixed Discounts

When few days remain until expiry, the system automatically lowers the price to maximize revenue from remaining stock. A probabilistic model is used: survival_model estimates the probability of selling all units at the current price. If probability is below 80%, price_optimizer finds the optimal discount. Typical markdowns: 3 days to expiry — 15% off, 1 day — 30%, on expiry day — 50%.

def calculate_markdown(current_price, days_remaining, daily_demand, units_remaining):
    """
    Optimal discount: maximize revenue from remaining stock
    subject to selling everything before expiry
    """
    prob_sell = survival_model.predict_proba(days_remaining, units_remaining, daily_demand)
    if prob_sell > 0.8:
        return 0
    optimal_price = price_optimizer(daily_demand, price_elasticity, days_remaining, units_remaining)
    markdown_pct = (current_price - optimal_price) / current_price
    return markdown_pct

For restaurants, the mechanism adapts as Daily Specials: AI generates dishes of the day from surplus ingredients with expiring shelf life, boosting margin and reducing waste.

Why IoT Waste Monitoring Matters

Installing smart bins — scales and cameras over trash bins — enables real-time visibility into what products are thrown away and why. The chef receives a daily report:

daily_waste_report = {
    'total_kg': 12.3,
    'value_usd': 45.80,
    'top_wasted_items': [
        {'item': 'Salmon', 'qty_kg': 2.1, 'cause': 'overproduction'},
        {'item': 'Mixed salad', 'qty_kg': 1.8, 'cause': 'plate_waste'},
        {'item': 'Croissants', 'qty_kg': 1.4, 'cause': 'expired'}
    ]
}

This allows prompt adjustments to purchasing and menus. Our engineers integrate IoT data into the overall platform for dashboard display. The system supports integration with donation platform APIs (FoodCloud, Too Good To Go), automatically offering surplus to charity organizations considering logistics and shelf life.

What's Included in Our Work

We provide a full set of deliverables:

• Documentation: technical integration docs, operator manuals, API descriptions.
• Access: to the system dashboard, API keys, model repository.
• Training: hands-on workshops for the team (chefs, buyers, administrators), video tutorials and checklists.
• Support: 3-month warranty after launch, then according to SLA.

How We Guarantee Results

Every project starts with an audit of current losses and flows. We establish a baseline and then prove effectiveness through A/B testing at one location. Only after a successful pilot do we scale to all locations. The system learns on your data, so forecast accuracy improves over time. Our team holds certifications from leading vendors — your business is protected.

Process

1. Audit of current losses and product flows
2. Data collection and cleaning (transactions, inventories, shelf life)
3. Development of predictive model accounting for seasonality and shelf-life
4. Integration with POS/ERP via REST API or file exchange
5. Deployment of dynamic markdown and IoT monitoring
6. Backtesting on historical data and A/B tests
7. Go-live and staff training
8. Post-release support and optimization

Expected Results

Development timelines: basic system with demand forecast and markdown engine — from 4 weeks. Full platform with IoT and donation API — from 3 months. Pricing is calculated individually. Request a demo for your business — we'll show how AI cuts your losses. Get a no-obligation consultation.

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.