Increasing number of purchases by customer through predictive modeling

Goal

  • Increase the Conversion Rate in those Customers with a Higher Propensity to Buy.
  • Our Client is an ecommerce with 1M purchases/year.

Methodology:

  • We will create a predictive model that will predict the purchase probability of each of our customers in the next month based on behavior and interactions with us in recent years.
  • With customers with the highest probability of purchase we will apply retargeting strategies to increase the possibility of closing the sale.

Phases

  • Data Exploration

    • Obtain, clean, and prepare data needed for the project.

  • Define and Implement Predictive Models (Machine Learning)

    • Featuring Engineering: Select/create  input variables
    • Comparison of two algorithms of predictive models in Artificial Intelligence and Machine Learning: Neural Networks using Tensorflow vs XGBoost.
    • Fine-tuning: Tune parameters / hyperparameters, iterate over the output, compare performance to finally selected best option.
    • Apply trained model to real data: The output is a list of Customers with High probability to Buy.

  • Define strategy for retargeting:

    • Define and implement A/B testing for email campaigns
    • Evaluate performance of the tests

Results

    • The application of the model drives to an increase of +25% in Conversion Rate
    • This CR drove to an increase of 4% Month over Month in Number of Purchases

Conclusion

The correct application of predictive models in our customer base has allowed us to detect specific behaviors at the customer level: customers with a high probability of buying during the next month. Actions on that information – sending emails with specific content – have substantially improved a key element of our business: the number of sales.

Marketing Channels

Life Time Value: Optimizing Marketing costs

Goal 

  • Optimize the marketing costs, spending the right money in the right Marketing Channel
  • Our client is an ecommerce  with online marketing costs of more than 10M per year.

Methodology 

  • We will calculate the Life Time Value (LTV) of each of the company’s clients and we will make an aggregation to find out the average LTV per Acquisition Channel.

Phases 

  • Data exploration

    • Obtain, clean, and prepare data needed for the project.

  • Calculate LTV by Acquisition Channel

    • Segment our population by Acquisition Marketing Channel 
      • SEM
      • SEO
      • Direct
      • CRM
    • Calculate Number of Purchases & revenue Margin of each cohort in next 12 months. 
    • We aggregate the information, obtaining the number of purchases and the average Revenue Margin per channel in each of the next 12 months.

  • Define strategy for Marketing Cost Optimization:

    • Based on the results obtained, we recalculate the budget and ROI of each Acquisition Channel.

Results

LTV resultados

  • We observe that although Payment Channels are more profitable in the short term (SEM), when we look at 12 months, Cheap Channels actually generate more purchases (Direct, CRM)
    • Direct overtakes SEM in terms of RM in month 7
    • CRM overtakes SEM in terms of RM in month 10

Conclusions

  • We conclude that customers from Cheap Channels are more loyal → Are we wasting investment in SEM ??
  • When we look at optimizing the costs of Acquisition Channels, we must look at the long term and not only at the time of acquisition.
  • Once we now understand the actual ROI of our Acquisition channels, we can recalculate and analyze the actual costs of each channel.