Concept: New Models for Shopping on Amazon 1

Easy Replenishment via Machine Learning

As the lead designer of a strategic high-profile product, I participated in high-level cross-company proofs of concept that were presented to leadership. This is some of the concept work for hyperefficient grocery shopping workflows.

Grocery and household shopping is a necessary and frequent chore. A key challenge then is how to make that regular shopping trip as quick, efficient and delightful as possible.

PROBLEM
• Long lists (average cart size = 25-30). 
• Limited time to shop. 
• Customers may not discover anything beyond basic or familiar items, given the time & effort to do even a basic shopping trip.
• First-time customers may be overwhelmed or discouraged by putting together their first cart.

SOLUTION
Use what we know about the customer to make their first shopping trip fun and easy, then continue to learn from them and everything else we know to make it smarter and better every time they shop.


Patent & earlier work on theme

In 2012, I filed a U.S. Patent application along with some colleagues on an earlier predictive shopping concept: "Time-Based Item Recommendations for Scheduled Delivery Orders" (13/709569). I also worked on the patent described in this article: "Amazon Patents the Milkman" (Geekwire, 2/8/2013). 


What customer signals do we use to power machine learning?

Amazon has an abundance of customer shopping data, both provided by individual customers and that which can be extrapolated from their cohorts (others with similar shopping patterns). In addition, consumables vendors (e.g. Proctor & Gamble) also have a solid customer research data that can be fed into the machine learning mechanism. 

easy-replen-via-machine-learning.png

Workflow 1: Quicklist Model

Our customer testing and other research showed that the one of the biggest barriers to customers shopping for groceries online was the significant effort needed to covert their real-world list ("milk, eggs, cereal") to specific items — i.e. searching, digging through options, etc. People also make their shopping lists on the go: they don't stand in the coffeeshop line and jot down "Kellogg's Corn Flakes, 72 oz" on their mobile phones — they write "cereal". 

(Some of this generic-to-specific disambiguation logic can also be found in my work on Amazon Dash.)

This workflow lets a customer make their grocery list in a natural real-language way and then uses the smarts of machine-learning to quickly get them to a cart full of specific products. Customers are smoothly led through their list, with high-relevance products presented in order of relevance for each list item, allowing them to move through their long lists in a speedy and game-like way. This allows them to checkout quickly ("look, I finished our shopping in under a minute!") and perhaps use that time they saved to discover new items and recommendations. 


Workflow 2: Quickshop Model

This model works through batch-adding bundles of items, which is ideally suited to new customers especially. For long-time customers, the bundles can be comprised of their most common sets of items that they buy together. The most basic model allows the customer to simply duplicate a past order. 

It can also work as a game-like discovery mechanism for seasonal or other specialized browsing, e.g. School Lunches, Summer BBQ, etc. The flexible workflow has many potential uses. 

easy-replen-quickshopmodel.png