Its important to understand that modern AI is just not a tool for to solve customers problems in a more effective manner. Its not that without modern AI , a you cannot build a good solution to solve a customers problem, but yes it can help the you solve the problem more effectively and may be with even higher speed.
In this lesson, I am going to talk about, which are the problems ML is helping to solve better. Here I am taking an example of a e-commerce website and how AI skilled executive are building for customer success.
Its important to understand, that ML currently following problems
1. Ranking System : Helps the user find the most relevant thing
2. Classification : Figuring out what types of thing/person/being is
3. Prediction : Predict the likely hood of an event
4. Recommender : Giving user things which they may be interested in
5. Clustering : Putting similar things together
6. Anomaly : Finding the uncommon thing
For understanding perspective, I shall be keeping things simple and from a Product Manager standpoint. To solve a problem using ML , two basic things need to be identified
1. Data : Volume and Variety
2. ML Algorithm or Algorithms to be used.
Lets start with e-commerce problem
Ranking System
Problem Statement : User is spending a lot of time on the Product Category page with low conversion rate.
Solution : Ranking of products based on highest purchase probability
Data Sets : Customer Behaviour/Activity, Customer Profile , Catalogue , Product, CTR and more
Learning : Supervised
ML Algorithms : Logistic Regression | Neural Nets | Decision Tree
Metrics : Higher Conversion | Higher Average Order Size
Classification
Problem Statement : Manual Cataloging is not scalable and a time taking effort
Solution : Automated Cataloging based on Image Recognition
Data Sets : Product Tags | Style Tags | Product Images and More
Learning : Supervised
ML Algorithms : CNN (Convolution Neural Networks)
Metrics : Reduced Manual Effort - Low Cost , Less Errors
Prediction Event
Problem : Users are dropping out at Checkout Page
Solution : Give Free Shipping to a user with high purchase probability based on product margins
Data Sets : Customer Behaviour/Activity | Customer Profile | Product Data | Logistics Data
Learning : Supervised | Unsupervised
ML Algorithms : Regression | KNN | Bayesian Clustering
Metrics : Higher Conversion
Recommender
Problem : High Bounce Rate on Product Details Page
Solution : Recommending Similar Items to the user on the product page
Data Sets : Catalog | Sales | Customer Profile
Learning : Supervised | Unsupervised
ML Algorithms : KNN | Matrix Factorisation | Neural Netwok
Metrics : Higher User Engagement | Avg Order Size
Clustering
Problem : Low click through rate of Email Marketing using traditional segmentation
Solution : Feature Identification beyond traditional segmentation
Data Sets : Customer Profile | Customer Sales Data | Customer Behaviour and More
Learning : Supervised | Unsupervised
ML Algorithms : K-Means | Expectation-Maximisation | Hierarchical Clustering
Metrics : Higher User Engagement | Avg Order Size
Anomaly Detection
Problem : Wrong Product Pricing gets displayed on the website , leading to order cancellations, returns
Solution : Notification to the pricing team when products price shows some anomalies
Data Sets : Catalogs Price | Competitor Price | Store Price | Brand Price | Category
Learning : Supervised | Unsupervised
ML Algorithms : Random Forest | Neural Nets | Gaussian Sb
Metrics : Customer Satisfaction| Loss Order
References
Machine Learning Fundamentals
Anomaly Detection for an E-commerce Pricing System :
Learning Ranking Assortment of Products
Machine Learning for Recommender systems — Part 1 (algorithms, evaluation and cold start)
An Introduction to Clustering and different methods of clustering : https://www.analyticsvidhya.com/blog/2016/11/an-introduction-to-clustering-and-different-methods-of-clustering/
コメント