Predictive Analytics: The Next Big Thing in E-Tailing

Posted in Operations & IT Articles, Total Reads: 1221 , Published on 01 May 2015

As defined, Predictive Analytics (PA) describes a range of analytical and statistical techniques used for developing models that may be used to predict future events or behaviors. It is different from descriptive or historical analytics in which the emphasis is on the summary of what happened in the past and how business was impacted in the past.

Predictive analytics and e-commerce have walked hand in hand since the time e-commerce was a mere adolescent. Although predictive analytics has been adopted by many industries including healthcare, insurance, telecommunication and banking, yet online retailers were the first to employ it. However, even though e-tailers know how Predictive Analytics works, the complete knowledge of PA stands more or less underutilized with e-tailers not putting such knowledge to substantial use.

Let us first take a look at the current challenges faced by e-tailers or online retailers in today’s times as put forth by Mark Richmond Consulting:

 Competition in similar verticals: With many companies offering the same products priced at par with each other, customers have thousands of options to choose from.

 High marketing budgets: E-tailers do manage to save on showroom costs as well as recruitment costs. However, they have to compensate by spending all their money on heavily advertising their website.

 High cost of customer acquisition: Customers are whisked away by different attractive discounts, combo offers etc. too quickly. Their attention span is too less. Thus, acquisition of a customer takes a long time and requires lot of effort from the e-tailer’s end.

 High Customer attrition: Again, as customers are enticed by a variety of websites offering new products, they no longer remain brand loyal to a particular e-tailer.

 Low profit margins: As companies indulge in providing heavy discounts to the customer at the cost of building a big customer base; their profit margins are low as compared to the local brick and mortar stores.

Companies like Amazon and Walmart are already on the verge of overcoming the above obstacles by incorporating PA into their system. Walmart purchased a predictive analytics firm by the name Inkuru in 2013 whose PA platform helped Walmart’s retailing business by building merchandising and marketing campaigns targeting shoppers when they are most likely to buy from the website. Similarly, Amazon in January 2014, filed a US patent for a ‘Method and System for Anticipatory Package Shipping’ based on predictive analytics. With these online retail giants taking such big steps, even the smaller emerging companies will follow suit in the future.

The biggest advantage of using PA is predicting the ROI of a particular venture in the future. Apart from this, few other benefits of PA are:

1) Predictive Search Techniques: If the site search can be made intelligent enough to predict what the customer wishes to purchase, then that can help in propelling sales.

2) Recommendations and Promotions: It is upto predictive analytics to analyze the browsing patterns of the customer and recommend and promote similar products which will help trigger interest in the minds of the customer.

3) Revenue forecasting: As marketing campaigns undergo change year on year and so do products, it becomes difficult for a company to forecast the revenues going to be generated the next year. PA helps in overcoming this difficulty.

4) Fraud Minimization: PA can help in analysis of low credit chargeback rates by understanding consumer behavior and sales of products and removing those products which are prone to fraud. The PA models help in detecting potential fraud even before the customer completes the transaction resulting in lower fraud rates.

5) Price Management: PA can be incorporated to analyze pricing trends in correlation with sales information available in order to come up with the right prices at the right time. A PA model can be used to analyze historical data of purchases made, types of products bought, amount paid etc. to arrive at the right price which a customer would be willing to pay in order to purchase that certain product. An example is given below of Amazon which charges different prices for the same product of different colors.

6) Supply chain management: PA can help the company effectively store whatever is expected to be in heavy demand in the market soon by helping the company forecast customer demand. Thus, the entire supply chain process right from planning, forecasting, sourcing, order fulfillment, delivery and returns can be completed effectively without any glitches. The inventory can be managed well leading to lesser cases of out-of-stock items or late deliveries.

7) Location Based Marketing: PA can help the company market its products depending upon the geographic region where the site is being accessed from, by the customer. Thus, it can be anything ranging from marketing in a local language to marketing on mobile channels.

8) Business Intelligence: Last but not the least, PA can help in business intelligence by capturing the customer’s purchasing trends and identifying what the customers like. An app provided by Custora uses PA to help increase customer lifetime value (CLV).

The predictive CLV can be calculated in the following way:

The following graph provides the other benefits of Predictive Analytics.

Source: quora 

Some metrics which need to be taken into consideration in predictive analytics are explained in a research paper by Kamal Nain Chopra:

 Server Session or Visit, which involves a collection of user clicks to a single Web server during a user session.

 Conversion Rate, which gives the number of completers, divided by the number of starters for any online activity that is more than one logical step in length, like Purchase or Downloading a research article.

 Attrition, which is a measure of the people successfully converted but not retained for converting again

 Frequency, which measures the activity generated by a visitor on a web site in terms of the time between visits, in the unit of days between visits

 Recency, which is the number of days passed after the last visit/purchase.

Other examples of good predictors can be:

 Recency+ personal income in order to determine how much the customer would be willing and able to purchase in the future.

 Attrition + low response to suggested ads determines what interests the customer and what doesn’t.

The different analytics approaches that can be undertaken by a company are as follows:

I. Collaborative filtering enables e-tailers to recommend new products and concludes how an individual will behave depending on how other similar individuals with same or similar characteristics have behaved in the past. E.g.:- Recommending matching colored shrug to a person who has recently purchased a pink colored sleeveless top by analyzing similar purchases made by other consumers in the past

II. Clustering Algorithms help in clustering customers into different groups. E.g.:- Customers who like extremely high priced clothes and customers who like everyday fashion clothes.

III. Propensity models help in predicting how a specific customer will behave because of the ability to handle enormous amounts of internal and external data.

IV. Uplift Models help determine if an investment is worth the effort. The uplift model helps determine the difference between customers who are likely to purchase a piece of cloth with or without any discounts and helps the e-tailer offer discounts only to those customers willing to purchase the piece of cloth only if discount is offered.

Predictive analytics is being utilized by many large retailers by offering specific loyalty programmes, targeting special customer segments etc. There are many companies like Retalon, custora and sas which give a demo on how predictive analysis can help save money for the retailers.

An instance of good utilization of predictive analytics that can be considered is that of an electronic retailer in US which wanted to understand ROI of its program so that decisions regarding the future spend on marketing could be taken wisely. A test and control group was designed for pre and post program measurement though ANOVA. ROI of the program was calculated by considering costs incurred and revenue generated by the program. The analysis yielded the results that only two out of four regions were profitable. Thus, the retailer only concentrated its marketing spend on the other two regions which yielded profits. Some other applications of PA by e-tailers include the following:

Source: sempreanalytics

Thus, it can be easily inferred that predictive analytics is more than merely gazing into the crystal ball and deals with lot of data, logic and statistics in order to provide the business with meaningful insights. Predictive analytics is going to change the way e-tailers attract new customers, retain acquired ones and maintain their competitive edge in the era of cut-throat rivalry. It is time e-tailers pulled up their socks and became analytics savvy alongwith being tech savvy.

This article has been authored by Deepti Sri Kocherlakota from SIBM Bangalore


Modeling and Technical Analysis of Electronics Commerce and Predictive Analytics. (2014). Journal of Internet Banking and Commerce , 1-10.


If you are interested in writing articles for us, Submit Here