TBM (Deductor)

Project Objectives

The company management of "TBM" for Neyroproyekt company set the following tasks:

• Development of methodology of formation of potovarny monthly forecasts for mass articles in all branches of the Company.
• Automation of process of forecasting and its integration into a corporate IT system (Oracle platform).

The received solution should provide:

• the best possible quality of forecasting;
• cost reduction of a nedoprognozirovaniye (loss of demand) and reforecasting (excesses in a warehouse);
• reduction of the resources spent by the Company for forecasting process maintenance.

Solution

For the solution of assigned tasks the Consultant selected the Deductor Enterprise platform as best combining such aspects as:

• ease of development and automation;
• ample opportunities of integration;
• reliability of the created solutions;
• the built-in data warehouse supporting different DB;
• project cost, optimal for the customer;
• modern visualization tools and reportings.

All necessary initial data from several sources were consolidated in uniform storage based on Deductor Warehouse 6 for Oracle DBMS.

At a stage of creation of methodology by specialists of Neyroproyekt company the set of prediction algorithms and data scrubbing was tested. Ten best diverse methods which could supplement each other were selected, increasing reliability of a single system.

On the basis of the created methodology the large-scale automated system of forecasting of sales is developed.

Optimization of stocks in TBM company

Features of the solution

• The scale of the solution System Circuit covers tens and hundreds of thousands of articles in all branches of the Company. Taking into account multiple forecasting a system creates from several hundred thousands to several million forecasts in each cycle of work. Actually, the general cycle of work of the created system does not exceed several hours (2–3 hours) from which the most part of time is spent on service of the data warehouse, and work of several "heaviest" algorithms.
• Parallel processing the Solution of Neyroproyekt very technologically, and fully implements possibilities of parallel processing of information. An analytical part of a system is unrolled on a dual-processor HP server of Proliant (2*Xeon 5520, a total quantity of cores = 16). The created system distributes loading on different flows of processing, and processor opportunities of the server are utilized very effectively.
• Utilization of processor power Parallel processing — necessary attribute of implementation of solutions of an industrial scale. If necessary on the Deductor Enterprise platform even for such large-scale cases it is possible to achieve real-time operation.
• Group forecasting a System for each article in each branch every time builds several forecasts (different algorithms, different primary data). A system automatically keeps track of quality of each of them on background, and considers this information for obtaining the uniform forecast. In the analysis of retroforecasts the specific requirements of TBM imposed to sales forecasts are considered.

Unlike the simplest case when the only forecast from several is selected, the created solution has high reliability and higher quality of forecasting. During different time frames different factors, and circumstances can influence the forecast. The quality of work of different methods also changes. At the same time a system automatically ranges primary forecasts for the level of the current relevance, and creates from them the uniform forecast.

Result

In a project deliverable the following objectives and effects are achieved:

• The automated solution for forecasting of sales in all branches of the Company is created.
• High quality of forecasting is provided. For the goods which are on sale regularly the mean error of forecasting is lower than the expert's error for 1-5% for most branches. At the same time these forecasts are under construction automatically.
• The comfortable data warehouse allowing to carry out the analysis of sales without knowledge of language of requests of SQL is created.
• Diagram of forecasts

Feedback

... TBM company expresses gratitude to Neyroproyekt LLC for implementation of this project. During conducting works (November, 2009 - 2010) of Neyroproyekt company proved as the reliable partner fulfilling the undertaken obligations....

The First Deputy CEO M.S. Shakirov