Conversion Forecasting: Quilytics' Predictive Analytics Elevates Sales Strategy for a Leading E-commerce Platform

Client Introduction

The client is a pioneering fashion brand renowned for its sustainable and innovative apparel. They use cutting-edge technology to create fabrics that repel liquids, stains, and odours, offering functional and stylish clothing that reduces environmental impact. With collections available in major cities and select international markets, they provide high-quality, eco-friendly fashion to discerning consumers.

Business Problem

The client encountered a major obstacle in accurately predicting overall demand and product-wise demand, stemming from the variability in their marketing campaigns, which hindered their ability to set effective sales and marketing goals. Without a clear grasp of forecasting website purchases through their e-commerce platform, they struggled to stock in-demand products during peak periods.

This challenge led the client to seek expert business analytics consulting to harness valuable, data-driven insights into their sales strategies. Recognizing the need for a predictive analytics solution in business analytics, they aimed to cater to better forecasting processes and optimize their overall performance in terms of product demand and ad campaigns.

Project Goals

Identify and prioritize key parameters that statistically impact website purchases, ensuring a thorough understanding of their influence on marketing campaigns.

Evaluate and select the most effective predictive model to accurately forecast conversions based on the identified parameters and test it effectively on unseen data.

Develop in-depth insights by analyzing the data and impact of each parameter on website purchases.

How did Quilytics help?

A) Identify key parameters

How marketing campaigns impact the conversions using exploratory data analysis:-

Harnessing our team’s extensive expertise with statistical analysis and visualizations, we undertook a thorough examination to unravel the complexities of the client’s campaign landscape and its impact on conversions. This involved a deep dive into channel-level spend and performance, campaign-level performance indicators and user engagement dynamics.
With exploratory data analysis, we visualised the intricate relationships between media channels, and campaign metrics alongside user demographic metrics and conversions using Python. Our goal was to gain preliminary insights to develop an accurate forecasting model.
We used the data from Google Analytics 4 that we implemented for the client to track their conversions and user behavior.

B) Building a forecasting model using advanced machine learning algorithms to forecast conversions

The next step was the process of model selection, which entailed a meticulous exploration of advanced time series algorithms and online tools to deliver precise insights into conversion dynamics.

Through the application of advanced machine learning techniques, we built a reliable and robust model for forecasting conversions using metrics such as page views, user count and the channel as well as campaign-level metrics like spend, impressions and clicks.

We trained the model on different periods to check the impact of seasonal changes on the predictability of the model. Upon performing data analysis, it was discovered that time ranges with at least one cycle of season.

Our testing data: training data ratio was 1:3. The metrics which did not add to the prediction were also dropped, for eg. Gender did not increase the accuracy or provide a significant inference, thus it was dropped.

Our model’s accuracy increased from 56% to 89% through this process. We first built simple CART models and later utilised ensemble techniques to reduce the error produced during predictions.

This enabled our team to forecast their conversions into the future using the optimal parameters for the next holiday season. This helped the client understand which SKUs to include in their inventory with their quantities.

Furthermore, we also used Vertex AI provided by GCP to further fine-tune our predictions, improving the accuracy to 95%.

c) Using Novel approaches to understand the results and derive further inferences on which ad campaign has the most effect on the model results

After the predictive model was trained, the model results were further fed to frameworks, which used game theory algorithms to understand the impact of each parameter.
This enabled the client to further check for which ad campaigns resulted in a higher demand allowing for the optimization of clients’ ad campaign groups.

Result

The deployment of our predictive analytics solution yielded significant benefits for the client’s purchasing division. The client was equipped with reports about the future demands based on their given ad campaigns which helped them prepare their inventory in advance.

By gaining comprehensive insights through exploratory data analysis into their marketing campaigns and the ability to forecast the conversions, the client was equipped with data-driven decisions to optimize marketing campaigns by identifying ineffective channels through the model which allowed the client to increase their ad spending in effective marketing channels.

The study helped the client predict demand, keep relevant product SKUs and increase their conversion rates by 15%.

The inferences derived post modelling and forecasting the data provided a real-time view of which campaigns contributed the most to conversions, this enabled the client to run more campaigns with metrics to increase the overall conversions.