Machine
learning
We combine artificial intelligence, industry expert experience and a flexible planning platform into a one forecasting process.
HOW WE FORECAST
Traditional methods use the simplest statistical methods, small data sets, samples and assumptions.
TRADITIONAL STATISTICAL METHODS
FORECASTING BY MACHINE LEARNING
Complex single and multivariate models that process large amounts of data,
with the selection of the optimal forecasting method
according to available data
and determining the degree of influence of external factors
on the forecast.
PREDICTIVE ANALYTICS
Machine learning, in conjunction with traditional statistics methods. It helps to analyze the set of events in the past,
and predict
future events are highly accurate as positive,
negative as well (to be avoided).
WHAT TASKS WE SOLVE
FORECAST OF SALES CURRENT
PRODUCT RANGE
For forecasting sales of a wide range of products with a wide geography of presence, automatic selection of univariate forecasting models is ideal
with minimum standard error
forecasting.
FORECAST OF SALES OF NEW PRODUCTS AND PRODUCTS WITH A SHORT LIFE CYCLE
Traditional statistics will not help to forecast sales of new products on the market. There is no historical data and fine tuning of forecast models is required
using data on related products (similar parameters) or new market data in comparison with peers.
SALES FORECAST BY PRODUCTS THAT DEPEND ON WEATHER AND OTHER EXTERNAL FACTORS
Products whose sales are sensitive to weather, temperature, humidity, light level
and other external factors, are more suitable for forecasting using multivariate models than the use of traditional statistics, especially
on a short planned horizon.
PLANNING PROMO ACTIVITIES
High-quality forecasting of the effects of promotions helps to save billions of dollars in marketing budgets and improve the quality of working with working capital. For this, the methods of searching for correlations, clustering, and machine learning based on multifactorial forecast models are combined.
HOW WE WORK
1. COLLECTING OF HISTORICAL DATA FROM SOURCES
CRM, ERP, retail data, market research, data from social networks.
2. CLEANING DATA
AND HARMONIZATION OF DIRECTORIES
Identification and correction of errors, data inconsistencies in order to improve data quality.
3. SELECTING OF ANALYSIS METHOD
TIME SERIES
WITH THE LEAST FORECAST ERROR
Automatic ranking and recommendations.
4. MONITORING
AND MODELING OF THE MODEL UNDER NEW CONDITIONS AND FACTORS
Identifying differences and conditions for successful application of the model.
5. CREATING A PLATFORM FOR COLLABORATION WORK AND DATA EXCHANGE
IN THE FORECASTING PROCESS
Accounting for the expertise of industry professionals.
WHAT METHODS WE USE
HOLT WINTER’S
EXPONENTIAL SMOOTHING
Holt Winter's Exponential Smoothing (HWES) is a triple exponential smoothing model, also known as the Holt-Winters method, which allows you to take into account the trend and seasonality of the time series.
XGBOOST
XGBoost is a gradient boosting algorithm for decision trees. Gradient boosting is a machine learning technique for classification and regression problems that builds a prediction model in the form of an ensemble of weak predictive models, usually decision trees. The model is new and very promising.
FACEBOOK PROPHET
The Prophet library is a model developed by Facebook to predict time series data based on an additive model in which non-linear trends are consistent with annual, weekly and daily seasonality, as well as holiday effects.
LONG SHORT-TERM MEMORY
Long short-term memory (LSTM) is an artificial recurrent architecture
neural network (RNN) used in the field of deep learning. Models of this class, inspired by the structure of the human brain, are complex and require more time and resource-intensive training, but at the same time make complex forecasts depending on many third-party factors.
SARIMAX
SARIMAX (Seasonal AutoRegressive Integrated Moving Average with eXogenous regressors model) is the Box-Jenkins model,
integrated autoregressive model
and moving average. Fixes a set of different time structures
in data, for example, reveals a trend, seasonality, influence
third-party factors.
WHAT WE OFFER
APPLY FOR A WEBINAR TO STUDY THE MODERN FORECASTING OPPORTUNITIES FROM TRADITIONAL STATISTICS TO NEURAL NETWORKS
ORDER A PILOT TO STUDY THE APPLICABILITY OF MACHINE LEARNING ALGORITHMS FOR YOUR FORECASTING CASE
TEST YOUR PREDICTION ACCURACY AGAINST CURRENT MACHINE LEARNING METHODS AND TECHNIQUES
DOWNLOAD MACHINE LEARNING BROCHURE (on Russian)
