Top Data Science Projects For eCommerce – Part 2

In the previous blog post we had explored some of the key Data Science projects in E Commerce that has practical value and significant impact on the business.

Let’s look through some more projects that are not only challenging but will bring about significant win for the business.

Top Data Science Projects

Promotional Forecasts

Generating accurate forecasts for sales and traffic not only helps in proper planning to meet the targets, but also gives a good understanding of what is likely to happen if you continue to run the business the same way as you did last year. Every year brings about significant changes to the business plan and it is very unlikely that you will operate the same way as last year. Especially for E Commerce where growth in year over year business is fact of life, it will be nearly impossible to achieve growth if you operate the same way.

One of the most critical levers in E Commerce is promotions and every year E Commerce businesses get creative about designing and running their promotions. Novelty is a key factor in designing promotions. While this is good news for the customers, it is bad news for forecasting.

Most forecasting algorithms are based on historical data and in case of totally new promotions introduced in a year there will be absolutely no data on which a forecast model can be built.

Coming up with a promotional lift model i.e how much lift in sales you are likely to get due to the newly introduced promotions is extremely crucial to know where you are headed and whether you can meet the targets.

This is extremely challenging since a model has to be built with very less historical data or no data at all. Using similar promotions data from history, using product features and customer behaviour to initially generate an estimate and fine tune that with actual data are some of the ways to solve this problem.

Assortment Intelligence

Product assortment planning is a key aspect of E Commerce and getting it right is crucial to achieve success in sales. In simple terms product assortment implies depth and variety of products available for sales in various categories. Predicting the depth, variety and bundles is a critical Data Science problem.

A typical medium sized E Commerce player can easily end up selling a million products in its marketplace. Coming up with the right assortments for millions of products is a scale problem that can easily be solved by a good prediction algorithm.

Using historical data can be a good starting point. Checking on competition and finding out their assortments – which will involve heavy duty page scraping, keeping tabs on product trends, customer behaviour trends, macro economic factors etc. are some of the features that can be used to build a predictive model that can predict what products to carry in different categories at different points in time during the year. This will help in-

  • Proactive planning
  • Onboarding the right merchants
  • Keep appropriate levels of stock, and
  • Coming up with a better pricing strategy than just depend on discounts

We have to constantly update this model with actual real time data since for fast moving products a prediction made say 3 months back may become completely invalid while it may work well for stable products.   

Read our blog on Common Challenges for Agile Testing Teams

Traffic Optimization

For E Commerce, getting the right customers at the right time for the right product is necessary to generate sales. So, this is a fundamental problem that has be solved really well. Businesses spend a lot of marketing budget on getting traffic. Hence, if it generates maximum sales with good ROI then it is worth the expense.

Traffic channels can be paid. E.g, ads in Google, Facebook or Google Search (Organic) or people may visit the website directly (Direct). Traffic may also come via partners (Affiliates), through Social Media and so on.

What is necessary is to predict how many users would be using these various traffic channels to visit the website or use the Mobile app. So, impose further budgetary constraints through an optimization algorithm. This will set the maximum thresholds per traffic channel so that the marketing spend per channel is well within budget.

It is extremely critical to choose the right paid channels. This is to ensure that the return on Ad spend (ROAS) is realized soon. In order to achieve this, we must undertake a thorough data analysis. This will identify the top paid channels with respect to high ROAS and allocate more share of the traffic to those channels.

Data science in brand building

For Unpaid channels investment in SEO and brand building exercises are a precursor to getting good quality traffic. So if the model predicts high volume traffic from Unpaid channels these investments will be necessary to achieve success.

We have to constantly update this model with real time data. So, it will be up to date with the latest signals and regenerates predictions based on that.

Downstream activities like-

  • Customer acquisition strategy
  • Allocating traffic to various categories
  • Conversion strategy
  • Fine tuning the purchase channel  etc., can be carried on based on the predicted data.

In a future post we will continue exploring more such projects. They will be innovative Data Science projects for E Commerce that have a significant impact on the business.

Blog Cover Photo by Bench Accounting on Unsplash

Leave a Reply

Your email address will not be published.