Deep Learning For Retail And Marketing: Practical Use Cases on - - PowerPoint PPT Presentation

GTC, 3/28

Alexander Tsyplikhin, PhD Data Monsters

Deep Learning For Retail And Marketing: Practical Use Cases

• n GPUs in the Cloud

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Alexander Tsyplikhin 15 years in data science, building products. PhD in speech biometrics. Data Monsters Professional services in deep learning. NVIDIA, Amazon, Intel, Cisco, Cray.

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Value of Deep Learning

Value of Deep Learning

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Accuracy, Business Value Amount of Data Deep Learning Classic Machine Learning

Value of Deep Learning

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Automate “on steroids” Increase revenue Reduce costs Improve control

Value of Deep Learning

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

When to Use Deep Learning

Value of Deep Learning

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

• 1. You have lots of data
• 2. High accuracy is valuable

Value of Deep Learning

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Identify Opportunities for DL-based Automation

Value of Deep Learning

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

• 1. How much you can make or save?
• 2. Is the price of error high?
• 3. Can you get a large dataset?

Value of Deep Learning

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Human-In-The-Loop

Value of Deep Learning

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Retail and Marketing

Value of Deep Learning

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Case #1

Customer Satisfaction: Support Automation

#1 Customer Satisfaction

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Operator cost (text) $2,500/mo

Includes hiring, training, PTO, bonuses, rent, software licenses, IT, infrastructure, AP, etc.

6 minutes per request $1.5/request 65 operators 100,000 requests/mo $2,000,000/y

#1 Customer Satisfaction

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction #1 Customer Satisfaction

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction #1 Customer Satisfaction

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction #1 Customer Satisfaction

Dozens, Hundreds, And Thousands Of Intents

Frequency of Requests

Real Statistics for a Bank 5000 conversations, 250 intents

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

FAQ different questions may have the same answer for all customers Personal Answers the same requests may have different answers for different customers Chain QA A sequence of interactions aimed to identify and solve customer’s problem

#1 Customer Satisfaction

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

NVIDIA K80 GPUs in AWS Bidirectional LSTM with dropout Hybrid Code Networks

#1 Customer Satisfaction

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction #1 Customer Satisfaction

61% requests processed 10 times faster Cost reduction 58% = $1,100,000/y Higher customer satisfaction

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction #2 Customer Relevance

Case #2

Customer Relevance: Clickstream-Based Segmentation

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Weak targeting based on profile information or previous purchases Need to increase conversions and sales

#2 Customer Relevance

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction #2 Customer Relevance

Collected all website user actions. Data was a stream of events: Timestamp Session Descriptor Action Descriptor

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction #2 Customer Relevance

https://arxiv.org/abs/1708.01715

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Website Users User Segmentation

#2 Customer Relevance

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

NVIDIA Tesla V100 GPUs in AWS Deep Autoencoder for segmentation Kafka, Spark, Hadoop

#2 Customer Relevance

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction #2 Customer Relevance

Higher customer relevance Higher campaign efficiency Targeted offers A/B test each segment separately

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Case #3

Process Improvement: Analytics On Video

#3 Process Improvement

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

A retail company wanted to analyze and

• ptimize processes in physical

Tracking actions based on CCTV camera videos 100 human analysts to manually mark events

#3 Process Improvement

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction #3 Process Improvement

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Hundreds of hours of unlabelled video footage Different views of objects Obscured by people and other objects

#3 Process Improvement

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

NVIDIA Tesla V100 GPUs in AWS Resnet-based CNN for object detection Domain-specific logics for analytics

#3 Process Improvement

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Cost Reduction Instead of 100 contractors working for 6 months 5 annotators worked for 4 weeks

#3 Process Improvement

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Control Improvement Analytics to improve business processes Real-time insights & anomaly detection

#3 Process Improvement

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

More Use Cases

Security and Analytics

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Security and Analytics

Security and Analytics

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Gather broad information about demographics and behavior patterns in-store. Detect anomalies in people flow and processes. Generate accurate, real-time estimates of all products in a given store (Digital Inventory). Detect suspicious behavior in store.

Security and Analytics

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Customer Experience

Customer Experience

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction Customer Experience

Frictionless commerce. Customer sentiment analysis in support centers and

• n social media.

Product personalization. Customer guidance for digital properties. Predicting problems using social media data.

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Sales and Marketing

Sales and Marketing

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction Sales and Marketing

Recommendations based on views, scrolls, tweets. Smart replies on media rich social media. Customer churn prediction. Dynamic pricing. Demand prediction.

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

AI Acceleration Program

AI Acceleration

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

AI Readiness

Assessment

Ideation Workshop

AI Acceleration

Deep Learning

Training Proof of Concept

End-to-end AI Solution

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

https://www.manning.com/books/deep-learning-with-python

Thursday, 12:30 PM

“Deep Learning With Python” (at the GTC bookstore) by François Chollet, the author of Keras

• No equations, instead, Python

code, accessible to anyone.

• Simultaneously practical &

advanced.

Book Signing At Our Booth

#3 Process Improvement Security and Analytics Customer Experience Sales and Marketing AI Acceleration #2 Customer Relevance Value of Deep Learning #1 Customer Satisfaction

Alexander Tsyplikhin 858.733.0469 alex@datamonsters.com

Thank you!