This blog post series was inspired by a series of events & interactions that have been happening for the last few weeks. The synopsis of the events have been haunting & following me to provide an answer to my colleagues & peers.
Event 1 : At a well audienced seminar I was asked, “What does digital mean to you?” to which I replied “It means the Now Economy. If you’re hungry you want food now from the Internet. If your client plans to meet you, they want the itinerary now. If your kids are bored at the Costco, they want Peppa Pig videos to render now”.
Event 2 : With the newly launched product owner at Adobe’s enterprise customer asks – “Sagar, initial reviews & feedback have been great and we have had a successful launch. But people dont seem to buy the product. We are seeing a huge number of hits to the page and added to the cart. But nothing beyond that”
Event 3 : The Head of eCommerce at a national retailer when asked about their personalization initiatives – “Our personalization programs lead to conversion only when there are discounts offered. Offers with no discounts do not move the needle towards conversion”
The solution that comes first to your mind is a data-driven & segment based next best offer engine which would show offers based on user actions. A quick Google of next best offer would attest to this solution. But the solution is not dynamic -it is still driven a set of Boolean rules on a data set defined within each segment.
Here’s a different solution – teaching the next best offer engine about type of offers lead to conversion and let it build models on the fly that try different offers to learn from them. The machine’s definitely not taking your job away! Because it needs your help in identifying what’s good and bad for business , take corrective measures in case of mistakes and stay on course to your quarterly goals.
I am not a data scientist (yet) to define a complete working solution. But the objective of this blog series is to follow best practices in machine learning , data sciences and make the most of your Adobe Analytics implementation. I am aim to end up with a redefined next best offer engine that is dynamic, scalable and able to demonstrate positive ROI for the business.
According to Yufeng Guo’s 7 steps to machine learning as documented in his bestselling book “Hands-On Machine Learning with Scikit-Learn & TensorFlow” ; the first step is to frame the problem. If you have read all of the above details, then you would agree with the frame of business & technical problem I have defined below:
Business Problem : Given a user can be known or unknown ; logged in or logged out; on a mobile device or desktop; we would want to determine the next best offer to be presented to the user. The offer can range from free shipping to % discount on the total price, applied on the checkout stages. The end of goal of the next best offer engine would be to drive offers that lead to conversion and a positive margin for the business.
With the business problem defined and solidified, the next steps towards building the machine learning models are detailed in individual blog posts below :
6.Validate Business Objectives
7. Go Live