Forecasting based on Machine learning ML

OUR CAPABILITIES IN MACHINE LEARNING

Data science

Data processing and analysis to find common patterns, dependencies and areas for optimization: preventing out of stock, analyzing the effectiveness of promotional activities, demand drivers’ decomposition, anomaly detection, etc.

Forecasting

Forecasting models based on machine learning algorithms: forecasting of demand, promotional sales, level of assortment cannibalisation, etc.

Optimization

Models maximize profits, revenue or reduce costs: optimizing promotional investments, optimizing stock in the warehouse or shop, smart pricing, etc.

ADVANTAGES OF ML FOR FORECASTING

MORE ADVANCED

Modern ML approaches can find implicit patterns between a target variable, such as a demand curve, and various influencing factors that cannot be identified using traditional models or expert opinion, learn from them, and use them to build a better forecast.

MORE FAST

With ML algorithms, huge amounts of statistical and variable forecast data are processed in minutes, freeing up experts’ time for more important strategic and analytical tasks.

MORE ACCURATE

Traditional forecasting methods are based on historical sales data, whereas ML forecasting supplements this data with other critical factors that influence demand: promotion costs, product inventory, weather, etc. ML forecasting enables you to explain changes in the numbers and identify new factors that were not previously evident.

MORE COST-EFFECTIVE

Improved forecast accuracy helps in reducing costs, improving profitability and efficiency across the supply chain, making better-informed decisions in the face of rapid change.

OUR FORECASTING SOLUTIONS

Even within the same industry, each project will be unique. Therefore, when creating a new customized ML model, we first help to analyze and, if necessary, improve business processes so that any obstacles to the launch and effective use of the new technology solution are eliminated.

BUSINESS PROCESS SETUP

Audit, analysis and setup of the target business process taking into account the specifics of the ML solution’s capabilities.

ML-MODEL DEVELOPMENT

Development, integration and launch of ML forecasting models.

BUSINESS AS USUAL SUPPORT

ML-model support after launch: technical support for ML-models and functional support for business teams.

KEY FUNCTIONAL AREAS AND USE CASES

Data science techniques and Machine Learning algorithms can more effectively resolve a number of key business problems - from demand forecasting to optimizing multi-million price promotion budgets.

Demand planning

Baseline sales

Total demand

History cleansing

Demand drivers’ decomposition

Trend and seasonality analysis

Outlier detection

Promo planning

Promo uplift evaluation

Regular sales cannibalization

Promotional sales

Non-price promotion effectiveness

Analysis of competitors’ promo strategies

Sales planning

Off-the-shelf sales forecast

Portfolio cannibalization analysis

Seasonal sales

Sales forecast driven by retailers’ co-investments

New product sales

Actuals decomposition and analysis

Supply planning

Inventory level projection

In store out-of-stock analysis and projection

Stock target calculation based on probabilistic forecast

Probabilistic demand forecast

MACHINE LEARNING METHODS USED

Based on the task, availability and quality of incoming data, our data scientists create a model with a unique set of features and algorithms. The goal can be solved by regression methods, decision tree algorithms, ensembles or a combination of the above.

REGRESSION MODELS

A simple yet powerful tool for time series forecasting, offering clear interpretations, ease of application, robustness and reliability, especially in situations with limited data.

DECISION TREES

A versatile and powerful machine learning algorithm, especially in scenarios where interpretability, non-linear relationships and feature selection are important factors.

ENSEMBLES

This technique simultaneously uses multiple learning algorithms, such as decision trees, linear models or neural networks, where a more accurate prediction can be obtained from an ensemble rather than from each algorithm individually. Flexibility in model architecture and hyperparameter tuning ensures fine-tuning for a specific problem.

HEURISTICS

The method accumulates the combined experience of a developers team in a particular industry, taking into account specific cases, and works to improve the efficiency of all the technologies used.

SEAMLESS INTEGRATION WITH ANAPLAN

It is possible to integrate models for forecasting based on machine learning algorithms with the Anaplan platform as well as with any integrated business planning solutions (IBP).