Retail Data Science and Demand Forecasting Training Course

Course overview

Retailers hold transaction, promotion, product, store, inventory and digital-channel data, yet many planning teams still forecast with spreadsheets, simple seasonal averages or models that cannot explain exceptions. The result is avoidable stockouts on fast-moving lines, excess inventory after promotions, poor allocation between stores and weak confidence in category plans. This course equips retail professionals to turn granular demand data into forecasts that are accurate enough to support replenishment, assortment, allocation and promotional decisions.

Participants learn a practical retail data-science workflow, from defining the forecasting grain and auditing point-of-sale data to engineering calendar, price, promotion, availability and hierarchy features. They build and compare baseline, seasonal, regression and machine-learning forecasts using Python, pandas and scikit-learn, then assess performance with MAE, RMSE, WAPE, bias and forecast-value-add analysis. The course addresses retail-specific challenges including intermittent demand, new-product launches, stockout distortion, cannibalisation, promotional uplift and forecasting across SKU-store, category and channel hierarchies.

Teaching combines instructor demonstrations, guided coding labs and retail case work based on realistic weekly sales and inventory datasets. Participants use Jupyter notebooks to clean data, construct features, train models and communicate results through a Power BI forecast dashboard. They leave with a reusable demand-forecasting project: documented data assumptions, a model-comparison report, forecast accuracy measures, exception rules and a 90-day implementation plan for applying the approach within their own retail operation.

The course is designed for analysts, planners and data professionals who already work with retail data and need to move from reporting historical sales to producing defensible forward-looking demand recommendations. It also gives managers responsible for inventory, trading or analytics a structured basis for evaluating forecasting initiatives and model outputs.

Course objectives

By the end of this course, participants will be able to:

• Profile retail sales, inventory and promotion data to identify missing values, stockout periods, outliers and aggregation errors
• Define an appropriate SKU-store-channel forecasting grain and forecast horizon for replenishment and trading decisions
• Engineer calendar, price, promotion, availability and product-hierarchy features in pandas
• Build seasonal-naive, moving-average and regression baseline demand forecasts in Python
• Train and tune tree-based machine-learning forecasting models using scikit-learn
• Evaluate forecast accuracy with MAE, RMSE, WAPE, bias and forecast-value-add measures
• Design exception rules for intermittent demand, new products, promotions and stockout-distorted sales history
• Produce a forecast dashboard and documented implementation plan for a retail demand-planning use case

Benefits of attending

Target competencies

Who should attend

• Retail Data Analysts — who need to convert point-of-sale and inventory data into decision-ready demand forecasts
• Demand Planners — who need more reliable SKU-store forecasts for replenishment and inventory positioning
• Merchandise Planners — who need to quantify seasonal demand, promotional effects and category-level forecast risk
• Supply Chain Analysts — who need to connect demand signals with allocation, fulfilment and stock-cover decisions
• Category Managers — who need to challenge forecast assumptions and interpret expected sales outcomes
• BI and Analytics Professionals — who need to build retail forecasting datasets, models and management dashboards

Requirements and prerequisites

Participants should be comfortable working with tabular business data and interpreting basic charts, percentages and averages. Prior experience using Excel for analysis and basic familiarity with Python syntax, such as variables, lists and reading a CSV file, are expected. The course uses Jupyter notebooks, pandas and scikit-learn, so attendees should be able to follow and edit short code examples. Prior knowledge of machine learning, time-series modelling, retail planning systems or Power BI is not required; these are introduced in the retail forecasting context. Participants should bring a laptop able to run a modern web browser and Python environment.

Training methodology

Each day combines focused instructor-led explanation with guided work in Jupyter notebooks using a realistic retailer’s sales, inventory, price and promotion extracts. Participants clean data, write and adapt Python code, compare model outputs and interpret the commercial consequences of forecast errors. Small-group case discussions address decisions such as allocating limited stock, planning a promotion and treating new items. The final day uses a workshop format: participants assemble their forecast report and Power BI dashboard, then define data, ownership, validation and adoption actions for a 90-day workplace application plan.

Course outline

Day 1: Retail demand data and forecasting design

• Retail demand-planning decisions and forecast use cases
• Forecast grain across SKU, store, channel and week
• Point-of-sale, inventory, price and promotion data structures
• Demand versus sales: identifying stockout-censored observations
• Data-quality profiling with pandas DataFrames
• Retail calendars, trading weeks and holiday event mapping
• Exploratory demand analysis using trend, seasonality and distribution plots

Workshop: Participants audit a multi-store sales dataset and produce a data-quality log, demand-profile summary and proposed forecasting grain.

Day 2: Retail time-series baselines and validation

• Time-series components: trend, seasonality, cycles and noise
• Seasonal-naive and moving-average benchmark forecasts
• Exponential smoothing for level, trend and seasonal demand
• Rolling-origin back-testing and holdout-period design
• MAE, RMSE, WAPE and bias calculation
• Forecast-value-add analysis for planner overrides
• Intermittent-demand patterns and Croston-style forecasting concepts

Workshop: Participants build baseline forecasts for a category-store portfolio and create an accuracy scorecard comparing methods by demand segment.

Day 3: Feature engineering and machine-learning forecasts

• Lag, rolling-window and prior-year sales features
• Calendar, holiday and local-event feature construction
• Price index, discount depth and promotion-mechanic variables
• Product hierarchy, store cluster and channel attributes
• Train-test separation to prevent temporal data leakage
• Random forest and gradient boosting regression models
• Hyperparameter tuning and feature-importance interpretation

Workshop: Participants engineer a retail feature set and train competing baseline and tree-based models, producing a model-selection recommendation.

Day 4: Retail forecasting exceptions and decision use

• Promotional uplift measurement and post-promotion dip effects
• Cannibalisation and halo effects within product categories
• New-product forecasting using analogues and attribute similarity
• Stockout adjustment and lost-sales estimation assumptions
• Hierarchical forecasting across SKU, category and total business levels
• Forecast segmentation by volume, variability and lifecycle stage
• Exception thresholds for forecast review and planner intervention

Workshop: Participants resolve a promotion and stockout case, documenting adjusted assumptions, exception rules and implications for inventory allocation.

Day 5: Forecast communication and operational deployment

• Forecast output tables for replenishment and merchandising users
• Power BI forecast-versus-actual dashboard design
• Bias, accuracy and exception visualisations for management review
• Model documentation, assumptions and reproducibility controls
• Forecast refresh cadence and data-pipeline requirements
• Human override governance and forecast approval workflows
• Ninety-day retail forecasting implementation roadmap

Workshop: Participants present a completed forecast notebook, Power BI dashboard and 90-day implementation plan for the retail case organisation.

Tools & standards covered

Python, Jupyter Notebook, pandas, Microsoft Power BI

A typical training day

Live online deliveries follow the same structure in the East Africa Time zone, with shorter screen blocks and longer breaks.

What the fee includes

How you can take this course

Certification

Participants who complete the full five days receive the Skillset Development Certificate of Completion, stating the course title, course code, dates and delivery format — suitable for professional-development records and employer reimbursement.

Frequently asked questions