Do your data inform your marketing tactics?
Wag! is an on-demand dog walking and boarding app available in all 50 states and 4,600 U.S. cities. The company recently sought to improve its marketing efficiency by more effectively targeting its marketing campaigns and other activities – specifically by leveraging Machine Learning (ML) to forecast nation-level dog walking demand and predict drops, knowledge which, in turn, would inform its marketing efforts. Wag’s data scientists needed the ability to make more accurate predictions more quickly. In an effort to amplify these efforts, Pariveda and AWS partnered with the company to conduct a seven-week ML Jumpstart project.
A jumpstart to improved accuracy and results
A team from Pariveda assisted Wag’s data scientists to assess its data and identify relevant features for ML experimentation, introduced an ML algorithm to generate improved accuracy and correlation results, and leveraged the Amazon SageMaker platform to speed up ML experimentation 15 times faster through automation and concurrency. Over the course of the project, Pariveda compressed 45 days’ worth of computation time into just three days.
A foundation for future ML experimentation
The Pariveda team developed for Wag! a ML model that reasonably predicts whether dog walking demand will increase or decrease within the next seven days. Results were as follows:
· 93.8 percent accuracy for the training data set;
· 80.2 percent accuracy for the validation data set; and
· 55.5 percent accuracy for the test (most recent) data set; this figure includes new, non-walking services that will need to be distinguished
· Approximately 70 percent accuracy across both validation and test data sets
Pariveda also identified for Wag! additional, high-probability data features to incorporate in future ML experimentation and defined the productionalization next steps for the current ML process.