Machine Learning –
Infusing AI into Commerce
Machine Learning –
Infusing AI into Commerce
September 2017
CONFIDENTIAL AND PROPRIETARY. ALL RIGHTS RESERVED.
Introduction
3
Dave Barrowman
VP, Head of Innovation for Skava
• Formerly at Gap Inc., AOL, and Netscape
• Drove Gap Inc.’s mobile and personalization initiatives
• Worked in e-commerce since 1996
CONFIDENTIAL AND PROPRIETARY. ALL RIGHTS RESERVED. 4
CONFIDENTIAL AND PROPRIETARY. ALL RIGHTS RESERVED. 5
CONFIDENTIAL AND PROPRIETARY. ALL RIGHTS RESERVED. 6
CONFIDENTIAL AND PROPRIETARY. ALL RIGHTS RESERVED.
Don’t just take it from me…
7
But much of what we do with machine learning happens beneath
the surface. Machine Learning drives our algorithms for demand
forecasting, product search ranking, product and deals
recommendations, merchandising placements, fraud detection,
translations, and much more. Though less visible, much of the
impact of machine learning will be of this type – quietly but
meaningfully improving core operations.
Jeff Bezos - 2016 Letter to Shareholders
https://www.scribd.com/document/344982413/Amazon-2016-Letter-to-Shareholders
CONFIDENTIAL AND PROPRIETARY. ALL RIGHTS RESERVED. 8
Online
HQ
Stores
CONFIDENTIAL AND PROPRIETARY. ALL RIGHTS RESERVED.
Online: More Efficient…
9
• Make online shopping more efficient
⎼ Fast track to the products most likely to appeal to the customer, with personalized
recommendations, search results, and product ranking
⎼ Smart, personalized shopping lists  “milk” ≠ “milk”
CONFIDENTIAL AND PROPRIETARY. ALL RIGHTS RESERVED.
“milk” ≠ “milk”
10
CONFIDENTIAL AND PROPRIETARY. ALL RIGHTS RESERVED.
Online: More Efficient…
11
• Make online shopping more efficient
⎼ Fast track to the products most likely to appeal to the customer, with personalized
recommendations, search results, and product ranking
⎼ Smart, personalized shopping lists  “milk” ≠ “milk”
⎼ Nudge: Anticipatory UI that emphasizes the appropriate next steps based on customer
behavior
⎼ Mitigate high online return rates by identifying likely returns and offering alternatives
CONFIDENTIAL AND PROPRIETARY. ALL RIGHTS RESERVED.
Online: More Efficient… More Engaging
12
• But it’s not just about efficiency…
• Machine Learning will also make online shopping more engaging
• Truly personalized marketing that will appeal and have emotional impact
⎼ Requires understanding your customer at a deep level…
⎼ And balancing that with their current behavior
CONFIDENTIAL AND PROPRIETARY. ALL RIGHTS RESERVED.
Online: More Efficient… More Engaging
13
• Learning from your social activity
⎼ Understand life stage, interests, activities
⎼ Photos and image recognition can power highly engaging experiences
CONFIDENTIAL AND PROPRIETARY. ALL RIGHTS RESERVED.
Stores: Make all associates as effective as your
best
14
• Empower associates with intelligent tools
⎼ Bring online personalization into the store:
• Recommendations, suggested sells
• Prompt the associate with targeted marketing messages
⎼ Machine vision capabilities to support product suggestions
• Call center agents too!
⎼ Much like store associates, call center agents will take advantage of these capabilities
⎼ While also offloading many interactions to NLP-based chatbots
CONFIDENTIAL AND PROPRIETARY. ALL RIGHTS RESERVED.
Stores: Smart support for associates and customers
15
• Also:
⎼ Smart fitting rooms to enable more efficient use of scarce resources (space and labor)
⎼ Task management / scheduling
⎼ Intelligent assistants for store managers (“Alexa for Store Managers”)
⎼ Autonomous robots for restocking, delivering inventory to customers
• Sensor-ification will enable many of these experiences
⎼ Via better inventory awareness and customer identification
CONFIDENTIAL AND PROPRIETARY. ALL RIGHTS RESERVED.
HQ: “Autopilot” to simplify & optimize business
tasks
16
• Machine Learning will reduce cognitive load in business
tools, enabling more efficient and better decisions
⎼ More efficient isn’t much good if the decisions are worse!
• It will be about supporting the experts to deliver better,
more predictable results across business tasks
including:
⎼ Product design
⎼ Planning and allocation
⎼ Catalog attribution & categorization
⎼ Pricing and discounting
CONFIDENTIAL AND PROPRIETARY. ALL RIGHTS RESERVED.
Challenges
17
• Currently, Machine Learning systems in retail are a patchwork
⎼ Various vendors include ML-based features in their services
⎼ Custom data science environments are expensive and difficult to maintain
• Encourages architectures with limited leverage
⎼ Lots of duplication and limited ability to share across systems
• Bold prediction: in medium-term there will be increasing emphasis on
common foundations
⎼ So that many Machine Learning services can leverage the same data and algorithms
CONFIDENTIAL AND PROPRIETARY. ALL RIGHTS RESERVED.
Two caveats!
18
You’re dead without:
1. Well-defined business goals
2. Lots of data
CONFIDENTIAL AND PROPRIETARY. ALL RIGHTS RESERVED.
Let’s imagine the near future of retail, end to end…
19
• Product designs and assortment determined with guidance from predictive models
• Product buy and inventory allocation supported by intelligent systems
• Online setup defaults provided by the PIM, with recommended price and categorization
• Discounts and markdowns automatically suggested
• Customer visits driven by truly personalized marketing
• Online experience with recommendations augmented with tailored UI based on customer behavior
• High online return rates mitigated by identifying likely returns and offering alternatives
• Store associates supported by smart tools to drive incremental purchases with personalized service
• Customers interact with brand and product via conversational and vision-based interfaces in store
• Robots shuttle inventory around the store to support restocking and customers in fitting rooms
• Call center agent tools augmented with NLP-based systems to predict/respond to customer needs
Thank You
skava.com
20

Machine Learning: Infusing AI into Commerce

  • 1.
  • 2.
    Machine Learning – InfusingAI into Commerce September 2017
  • 3.
    CONFIDENTIAL AND PROPRIETARY.ALL RIGHTS RESERVED. Introduction 3 Dave Barrowman VP, Head of Innovation for Skava • Formerly at Gap Inc., AOL, and Netscape • Drove Gap Inc.’s mobile and personalization initiatives • Worked in e-commerce since 1996
  • 4.
    CONFIDENTIAL AND PROPRIETARY.ALL RIGHTS RESERVED. 4
  • 5.
    CONFIDENTIAL AND PROPRIETARY.ALL RIGHTS RESERVED. 5
  • 6.
    CONFIDENTIAL AND PROPRIETARY.ALL RIGHTS RESERVED. 6
  • 7.
    CONFIDENTIAL AND PROPRIETARY.ALL RIGHTS RESERVED. Don’t just take it from me… 7 But much of what we do with machine learning happens beneath the surface. Machine Learning drives our algorithms for demand forecasting, product search ranking, product and deals recommendations, merchandising placements, fraud detection, translations, and much more. Though less visible, much of the impact of machine learning will be of this type – quietly but meaningfully improving core operations. Jeff Bezos - 2016 Letter to Shareholders https://www.scribd.com/document/344982413/Amazon-2016-Letter-to-Shareholders
  • 8.
    CONFIDENTIAL AND PROPRIETARY.ALL RIGHTS RESERVED. 8 Online HQ Stores
  • 9.
    CONFIDENTIAL AND PROPRIETARY.ALL RIGHTS RESERVED. Online: More Efficient… 9 • Make online shopping more efficient ⎼ Fast track to the products most likely to appeal to the customer, with personalized recommendations, search results, and product ranking ⎼ Smart, personalized shopping lists  “milk” ≠ “milk”
  • 10.
    CONFIDENTIAL AND PROPRIETARY.ALL RIGHTS RESERVED. “milk” ≠ “milk” 10
  • 11.
    CONFIDENTIAL AND PROPRIETARY.ALL RIGHTS RESERVED. Online: More Efficient… 11 • Make online shopping more efficient ⎼ Fast track to the products most likely to appeal to the customer, with personalized recommendations, search results, and product ranking ⎼ Smart, personalized shopping lists  “milk” ≠ “milk” ⎼ Nudge: Anticipatory UI that emphasizes the appropriate next steps based on customer behavior ⎼ Mitigate high online return rates by identifying likely returns and offering alternatives
  • 12.
    CONFIDENTIAL AND PROPRIETARY.ALL RIGHTS RESERVED. Online: More Efficient… More Engaging 12 • But it’s not just about efficiency… • Machine Learning will also make online shopping more engaging • Truly personalized marketing that will appeal and have emotional impact ⎼ Requires understanding your customer at a deep level… ⎼ And balancing that with their current behavior
  • 13.
    CONFIDENTIAL AND PROPRIETARY.ALL RIGHTS RESERVED. Online: More Efficient… More Engaging 13 • Learning from your social activity ⎼ Understand life stage, interests, activities ⎼ Photos and image recognition can power highly engaging experiences
  • 14.
    CONFIDENTIAL AND PROPRIETARY.ALL RIGHTS RESERVED. Stores: Make all associates as effective as your best 14 • Empower associates with intelligent tools ⎼ Bring online personalization into the store: • Recommendations, suggested sells • Prompt the associate with targeted marketing messages ⎼ Machine vision capabilities to support product suggestions • Call center agents too! ⎼ Much like store associates, call center agents will take advantage of these capabilities ⎼ While also offloading many interactions to NLP-based chatbots
  • 15.
    CONFIDENTIAL AND PROPRIETARY.ALL RIGHTS RESERVED. Stores: Smart support for associates and customers 15 • Also: ⎼ Smart fitting rooms to enable more efficient use of scarce resources (space and labor) ⎼ Task management / scheduling ⎼ Intelligent assistants for store managers (“Alexa for Store Managers”) ⎼ Autonomous robots for restocking, delivering inventory to customers • Sensor-ification will enable many of these experiences ⎼ Via better inventory awareness and customer identification
  • 16.
    CONFIDENTIAL AND PROPRIETARY.ALL RIGHTS RESERVED. HQ: “Autopilot” to simplify & optimize business tasks 16 • Machine Learning will reduce cognitive load in business tools, enabling more efficient and better decisions ⎼ More efficient isn’t much good if the decisions are worse! • It will be about supporting the experts to deliver better, more predictable results across business tasks including: ⎼ Product design ⎼ Planning and allocation ⎼ Catalog attribution & categorization ⎼ Pricing and discounting
  • 17.
    CONFIDENTIAL AND PROPRIETARY.ALL RIGHTS RESERVED. Challenges 17 • Currently, Machine Learning systems in retail are a patchwork ⎼ Various vendors include ML-based features in their services ⎼ Custom data science environments are expensive and difficult to maintain • Encourages architectures with limited leverage ⎼ Lots of duplication and limited ability to share across systems • Bold prediction: in medium-term there will be increasing emphasis on common foundations ⎼ So that many Machine Learning services can leverage the same data and algorithms
  • 18.
    CONFIDENTIAL AND PROPRIETARY.ALL RIGHTS RESERVED. Two caveats! 18 You’re dead without: 1. Well-defined business goals 2. Lots of data
  • 19.
    CONFIDENTIAL AND PROPRIETARY.ALL RIGHTS RESERVED. Let’s imagine the near future of retail, end to end… 19 • Product designs and assortment determined with guidance from predictive models • Product buy and inventory allocation supported by intelligent systems • Online setup defaults provided by the PIM, with recommended price and categorization • Discounts and markdowns automatically suggested • Customer visits driven by truly personalized marketing • Online experience with recommendations augmented with tailored UI based on customer behavior • High online return rates mitigated by identifying likely returns and offering alternatives • Store associates supported by smart tools to drive incremental purchases with personalized service • Customers interact with brand and product via conversational and vision-based interfaces in store • Robots shuttle inventory around the store to support restocking and customers in fitting rooms • Call center agent tools augmented with NLP-based systems to predict/respond to customer needs
  • 20.