Author: Marta Wodzińska

Segmentation of the customer database

Posted on by Marta Wodzińska

Problem:

One of the retailers, in order to diversify its offer and communication, wanted to distinguish business-relevant segments in a database of almost 4 million registered consumers.

Rozwiązanie:

Using machine learning methods, 5 customer segments were identified based on nearly 100 variables. These segments were characterised and described for marketing purposes. Each customer in the database was assigned a segment with the possibility of periodical automatic refreshing.

Effects:

  • Automation of the cyclic segmentation refreshing allows to save time for marketing and analytic tasks by approx. 40%
  • Adjustment of the offer and language to the customer segment
  • Tracking customer migration between segments in time
  • Possibility to assess the increase in effectiveness of marketing activities addressed to a specific segment

Data sources:

  • Sales data
  • Product data
  • Promotion calendar

Choosing the optimal distribution network

Posted on by Marta Wodzińska

Problem:

When introducing a new product on the market, one of the FMCG companies needed to decide which distribution networks to cooperate with. The aim was to ensure product availability in target groups in the most important cities in Poland while limiting the number of partners.

Solution:

A tool was built to simulate target market coverage depending on the selected network and other parameters. Business users can change simulation assumptions and test different scenarios themselves.

Effects:

  • Possibility to simulate and evaluate different decision options
  • Optimisation of distribution costs

Data Sources:

  • Sales data
  • Demographic data (Central Statistical Office of Poland, external suppliers)
  • Point of sale network data; Distances/time to arrive

Selection of areas for catalogue distribution

Posted on by Marta Wodzińska

Problem:

High printing and distribution costs mean that paper catalogues are prepared in limited editions.  It is necessary to decide in which areas to carry out the distribution.

Solution:

A tool for the marketing department has been developed, creating a ranking of areas based on several variables with the possibility of determining their weights by a business user. The solution was nationwide (it concerned all the regions in which the retailer operated).

Effects:

  • Saves time and effort (by about 30%) for the marketing department
  • Increase in efficiency of operations (in the first year, thanks to the recommendations of the model, approximately 20% of the circulation was reallocated)

Data sources:

  • Transaction systems; Loyalty program
  • Demographic data (Central Statistical Office of Poland, external suppliers)
  • Internal surveys and studies

Forecats of number of shoppers

Posted on by Marta Wodzińska

Problem:

One of the leading retail marketers in Poland needed short-term forecasts of the number of shoppers for better store work planning.

Rozwiązanie:

Using modern machine learning algorithms, the models were prepared to forecast traffic in stores 10 and 30 days in advance. The forecasts are provided daily in the form of interactive dashboards and email alerts.

Effects:

  • Forecast of traffic in each store (accuracy of about 5% for 30 days ahead)
  • Possibility to understand what 4 factors affect the traffic in the stores and how
  • Identification of the 3 most important promotions affecting traffic in stores
  • Possibility to identify days with increased traffic 30 days in advance

Data sources:

  • Transaction systems
  • Weather database
  • Calendar of promotional campaigns