Sie sind auf Seite 1von 25

S marter Predictions

Dean Abbott
Co-Founder and Chief Data Scientist, SmarterHQ
P resident, Abbott Analyt ics
dabbott@smarterhq.com
Twitter: @deanabb

SmarterHQ - Why

Built For Marketers, By Marketers

40+ years multi-channel retail


25+ years machine learning
20+ years digital marketing

Value Delivered, Clients

Proven time to value


Self funding platform
20x ROI

Thought Leadership, Funding

2015 Sherpa campaign of the year


8 eTail award winning clients
Funded by Battery Ventures & CNB

SmarterHQ - Who

A platform that unites customer intelligence and cross-channel marketing.

Focuses on attainable innovation that grows with retailers.

Delivers on the promise of unifying marketing channels to provide consistent messaging, offers
and promotionsand more.

Places shoppers at the center of the conversation to provide a full understanding of a shoppers
relationship with a brand, from online to in-store.

Enables marketers to take action on this intelligence within their existing marketing partner
investments; from email, to display to onsite personalization.

Data for Customer Intelligence


Customer-centric, Multi-channel
ENTERPRISE LEVEL
BRAND LEVEL

Mobile
App

Website

OTHER BRANDS

Call
Center

In-Store

3rd Party

Amazon

eBay

Web Support

#5306

#2456

Spring Catalog

#9102

#3314

Droid

iOS

Mobile Site

Desktop Site

Customer Intelligence Why It Matters

ENTERPRISE LEVEL
BRAND LEVEL

Mobile
App

Website

OTHER BRANDS

Call
Center

In-Store

3rd Party

Amazon

eBay

Web Support

#5306

#2456

Spring Catalog

#9102

#3314

Droid

iOS

Mobile Site

Desktop Site

WHO IS FRED?
He is a different shopper at every level
Integrating his interactions tells a complete story.
5

Its A Conversation

GOAL
Each interaction, or lack thereof, with a brand is a shopper trying to have a conversation.

WHO
WHO
Who am I talking to?

WHAT
WHAT
What are we talking about?

Customer Intelligence

HOW
HOW
How are we communicating?

Cross-Channel
Marketing
Display

Email

Complete The Conversation

Tuning Abandonment - Conversation

GOAL
Recover a shopper that abandoned their shopping cart .

WHO

WHAT

HOW

Anyone

Recent Visit To Website

Targeted Messaging

Carted Items
Did Not Purchase
Automated Email

Email

WHO IS KRISTEN?
Version 1
What is the right action?
7

Tuning Abandonment - Conversation

GOAL
Recover a shopper that abandoned their shopping cart that was acquired through paid media.

WHO

WHAT

HOW

Guest

First Visit To Website

Targeted Messaging

High Marketing Cost

Carted Items

Targeted Offer

Did Not Purchase


Automated Email

Email

WHO IS KRISTEN?
Version 2
What is the right action?
8

Tuning Abandonment - Conversation

GOAL
Recover a shopper that abandoned their shopping cart that was acquired through paid media.

WHO

WHAT

HOW

Guest

First Visit To Website

Targeted Messaging

High Marketing Cost

Carted Items

Targeted Offer

Did Not Purchase


Automated Email
Onsite Messaging

Onsite

Email

Optimizing Margin The Non-Conversation

GOAL
I want to optimize the value of my High AOV, engaged customer.

WHO

WHAT

HOW

Customer

Browsed Premium Product several

Get out of his way and let him

High Dollar AOV

times in past 3 days

purchase.

Highly Engaged

High predicted propensity to purchase

Save margin and do not make an offer

within the next 3 days

if he doesnt purchase within 3 days,


then make an offer if purchase propensity
still high

Print
Display

Onsite

Email

10

Predictive Models -> Insights

Predicted Next
Product/Category
to Purchase

Cross-Sell /
Up-Sell /
Recommendations

Customer
Attrition Risk

Category Browse
Abandonment

Purchase
Replenishment

Predicted Days to
Next Purchase

Predicted Days to
Next Visit

Customer (Lifetime)
Value

Visit Quality

Engagement (with
Visit Quality Decay)

Marketing
Attribution

Checkout Abandon
Causality

Asset Quality

11

Days To Next Purchase: Single Regression Model

All customers

1day

7days

15 days

30 days
12

Days To Next Purchase: Single Regression Model,


One Customer

Uni-modal
One value per customer

6.8 days +/- 1 day


Single customer

1day

7days

15 days

30 days
13

Days To Next Purchase: Set of Binary Classification Models

3-7 days
2-3 days

Single customer

7-15 days
15-30 days

1 day 3 days 7days

15 days

30 days
14

Days To Next Purchase: Set of Binary Classification Models

3-7 days 7-15 days


2-3 days

1 day 3 days 7days

15-30 days

15 days

Single customer

30 days
15

Days To Next Purchase: Set of Binary Classification Models

3-7 days 7-15 days


2-3 days

1 day 3 days 7days

15-30 days

15 days

Single customer

30 days
16

Models for Customer Intelligence


How Many Do We Need?
ENTERPRISE LEVEL
BRAND LEVEL

Mobile
App

Website

OTHER BRANDS

Call
Center

In-Store

3rd Party

Amazon

eBay

Web Support

#5306

#2456

Spring Catalog

#9102

#3314

Droid

iOS

Mobile Site

Desktop Site

17

What Data Best Predicts Behavior?


hDp://www.kaushik.net/avinash/compeHHve-intelligence-analysis-tools-metrics-reports-
techniques/?utm_campaign=viralheat&utm_medium=social&utm_source=twiDer

18

Data Preparation: Feature Creation

19

Data Preparation: Variable Reduction

20

Data Preparation: Stratified Sampling (for speed only)

21

ModelingLots of Preparation to Get to a Single Model

22

A Quick Aside on Target Shuffling

23

A Quick Aside on Target Shuffling: 500 runs

24

Thank You!!

CUSTOMER INTELLIGENCE UNITES


CROSS-CHANNEL MARKETING