Hourly traffic forecasts in stories

Problem:

In connection with the COVID-19 epidemic and potential customer concerns about the safety of their visits to the stores, one of the retailers in Poland wanted to provide their customers with precise information about the expected traffic volume in its brick and mortar stores.


Rozwiązanie:

Based on machine learning algorithms, predictive models were prepared which are capable of forecasting the number of transactions in hourly intervals 7 days in advance. A mechanism for regular automatic refreshing (teaching) was built to allow for quick adaptation of forecasts to new “post-epidemic” reality and new patterns of consumer behaviour.

Effects:

  • 7-day forecasts with average accuracy of 92%-95%
  • The system allows you to share the current forecasts with consumers through the website (integration of forecasts with the existing customer website)
  • Automation of the model refreshing process makes it possible to save about 60% of workload

Data base:

  • Sales data (historical)
  • Sales data (current, real time)
  • Weather data
  • Calendar data (days off, trade Sundays, promotions)

Customer database segmentation

Problem:

One of the leading retailers in Poland with an extensive network of outlets of various formats needed to create a communication strategy for its customers. One of the stages of the project was the segmentation of the consumer database.


Solution:

Using machine learning methods, 7 segments were identified based on 88 variables describing different aspects of customer behaviour. These segments were characterised and described for marketing purposes.

Effects:

  • The report describing the segments was used to adjust the offer and language to each 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 base:

  • Sales data (over 3.4 billion items)
  • Consumer data (over 5 million customers)
  • Product data
  • Data on the use of discount codes

CRM performance analysis

Problem:

One of the largest NGOs wanted to conduct a comprehensive review of CRM activities in the context of increasing the effectiveness of fundraising campaigns.


Solution:

An extensive analysis of the donor database, gathered by the organisation and the history of their interactions, was conducted. An extensive report was prepared, summarising the effectiveness of the organization’s current activities and indicating areas for optimisation. The donor database was segmented.

Effects:

  • The features of particularly valuable donors were identified, which made it possible to focus acquisition activities on people with an average donation value higher by about 60% than the rest
  • Reduction by about 10 percentage points of the percentage of people withdrawing from making the payments thanks to the recommendation that the retention activities should be accelerated

Data sources:

  • CRM Database
  • Data from the payment processing system
  • Fundraising activities calendar
  • Media data

Product recommendations on the website

Problem:

In connection with the development of the Internet sales channel, one of the retailers wanted to enrich its website with product recommendations.


Rozwiązanie:

Based on the basket analysis, a recommendation engine was built to power the website. The system fully integrates with the website, enabling its dynamic modification in accordance with recommended scenarios.

Effects:

  • Conversion increase by almost 14% (compared to the control group with disabled recommendations)
  • Increase in basket value by over 11%

Data sources:

  • Transaction data
  • Product data and business rules
  • Data on traffic on the website