Realizing AI Effectively for Digital Commerce: The Composite AI Approach

AI has made organizations more agile and adaptable to changes in the future.
AI has made organizations more agile and adaptable to changes in the future.

The pandemic has changed the dynamics of the retail world which has taken on a new avatar with growing smartphone penetration and disruptive technologies, one of which is Composite Artificial Intelligence (leveraging a multiplicity of AI methods in organic harmony with each other). 

Deployment of AI in retail is expected to grow at a CAGR of 34.4% from 2020 to reach $19.9 billion by 2027, according to Meticulous Research. Gartner too had identified Composite AI as No. 1 Hype Cycle trend for 2020 and we think the feature rules in 2021 and beyond.

Why is the retail sector investing heavily in AI? The reason is not difficult to fathom. AI has opened up new opportunities and capabilities, speeded up processes and has made organizations agile and adaptable to changes in the future. To stay ahead of the competition, the retail sector is banking heavily on AI.

[Related: Retaining the New Consumer Through the Power of Technology]

However, there are two key requirements to doing AI well: Firstly, a deep understanding of the sector being served and what we need to address. The key problem of AI is not coming up with the right answer; it is knowing what question to ask.

Secondly, a data-driven approach that carefully identifies the outcomes to optimize for, the inputs to use, the evaluation criteria, and the appropriate technologies to use. Depending on the problem, this may involve a variety of approaches such as rules-based systems, unsupervised and supervised machine learning, natural language processing (NLP), optimization techniques, graph techniques, deep learning etc.

A rigorous and data-driven approach ensures that the models and data flows we design are the ones that lead to truly successful outcomes.

In e-commerce there is no one-size-fits-all for AI deployment. Specific needs of retail and e-commerce have specific technologies. Composite AI harnesses powerful technologies such as deep learning for images and natural text and contextual bandits for nimble and sophisticated optimization. It facilitates rich customization that is controllable or automated as the client desires.

Intelligent Customer Interactions

The benefits of Composite AI are many. It allows retail enterprises to have intelligent customer interactions, leading to personalization of customer experience, cross-selling and up-selling; merchandising; catalog handling; and end-user satisfaction.

Customer journeys make important data and retail thrives on seamless and frictionless customer journeys across channels and Composite AI acts as an enhancer and a facilitator for quick launches. Streaming live product recommendations driven by location, device, contextual understanding from clickstream/NLP chat conversations become a selling point.

Deep Learning for Right Recommendations

The Deep Learning approach delivers right recommendations in e-commerce where collaborative filtering does not succeed in making the right predictions, particularly where products do not have any user interaction at all. There needs to be a deep language model to understand cross-sell associations between products, based on product descriptions, trained by actual purchase events.

This new approach removes constraints linked with traditional recommendations that may not work for retailers with sparse data. It also helps product discovery by capturing user’s preferences through a product’s visual features and textual description.

[See also: AI-Powered Fulfillment and Distribution]

It can train a deep image network for "Complete the Look" recommendations for fashion/apparel items, working on a combination of clothing items and accessories that go well together (based on actual outfits) and then uses this trained model to propose complementary items, with phenomenal results. 

Contextual Bandits Picks Optimal Strategy

The Contextual Bandits approach utilizes available context information to pick the optimal strategy for the page/placement at hand, in real time. It targets different optimization metrics for different stages of the shopping funnel: optimizing click-through-rate (CTR) while browsing items and optimizing revenue during/after purchase.

This is the answer for the recommendation space where there are many layers such as target engagement; inputs from point of sale etc.

Transformative Results from Composite AI

Retailers have benefited by deploying DeepRecs NLP and DeepRecs Visual AI. A Japanese entertainment online retailer with over 2 million products reported that 96 % of the products did not get sufficient view or have purchase history but with DeepRecs NLP (recommendations use text data i.e., product descriptions and products with similar affinities) the products propped up for shoppers. The CTR for the retailer increased +4.99%, while the average order value (AOV) was up by over 6.22% and the revenue per visitor (RPV) grew by over 7.29 %.

A French fashion retailer reported over 19% RPMI and 40% CTR with store-like recommendations coming up on e-commerce, leveraging visual characteristics of a product. This resulted in generating similar product recommendations based on visual similarity.

For transformative results from Composite AI, enterprises and vendors need to be immersed in the business space and have thorough knowledge of the key applications and pain points. A Composite AI approach needs a “composite architecture,” factoring in packaged business capabilities that run atop a flexible data fabric.  Retail needs to have a deep understanding of Deep Learning to harvest rich dividends.

Apu Mishra
Apu Mishra

Apu Mishra leads the machine learning and artificial intelligence team at Algonomy, which pursues technical innovation in data science to improve the retail/shopping experience. Prior to this he lectured at the University of Washington and worked at Intel. He received his PhD in Electrical Engineering from the University of Washington, where he first discovered data science through working with brain-computer interfaces and bio signals. He lives in Seattle.

More on Artificial Intelligence