week8_all merged pdf

7/13/2019 Week8_All Merged PDF

1/89

Go To Market Strategies:Digital Marketing

David Bell (@davidbnz)

Xinmei Zhang and Yongge Dai Professor


2/89

Digital Marketing is the use of Internet

connected devices to engage a customer

Definition

[Its still marketing*]


3/89

Elements of Digital Marketing

!Social Commerce: reviews, blogs, curatorsnetworks, and the strength of weak ties

!Digital Advertising and Behavioral Targeting: bigdata and micro-level targeting

!Experimentation and A/B Testing: More datacustomer data collected in 2012 than all prioryears


4/89

Evidence on Importance of Brands

!

Intangible assets now account for about 50% ofall assets for non-financial businesses (only20% in 1950)

!About one third of the value in global stockmarkets is attributable to brands

!The top global brands outperform the market;majority of the top twenty stocks in Stocks forthe Long Runare strong brands


5/89

Other New Brands

!www.warbyparker.com

!www.harrys.com

!www.bonobos.com

!www.diapers.com


6/89

Goals and Tactics

!Brand goals: Heart and mind, thinking andfeeling (sometimes action as well)

!Key tactic: Build engagement through real worldevents that can be leveraged up in the virtualworld


7/89

Digital Considerations

![Outstanding value proposition and positioning]

!Authenticity and transparency (all stakeholders)

!Brand personality and humanization

!Infinite life and potential for serendipity


8/89

Ree Drummond runs one ofthe largest blogs in the US

23M monthly pageviews;4.4M unique monthlyvisitors (rivaling The Daily

Beast)

Est $3.5M in annual adrevenue

Book deals, TV show, SonyPictures movie based onRees book (Reese

Witherspoon starring)

Endorsing Land O Lakesbutter, among other brands

Source: New Yorker, Wikipedia, FederatedMedia, Technorati

Organic Celebrity


9/89

Go To Market Strategies:

Digital Marketing

David Bell (@davidbnz)



10/89

Goals and Tactics

!Customer goals: Attract, engage, retain subject

to:

1. Never pay more to acquire than you will

recoup, i.e., CLV > AC

2. CLV needs to incorporate RLV

!Key tactic: Encourage customers to refer andacquire other customers (www.Diapers.com)


11/89

Digital Considerations

![Attractive target customer]

!Monologue to conversation

!Amplification through virtual and real world synergy

!Long tail leverage

![Selection and Treatment]


12/89

Marketing Spend As An Asset


13/89

Food For Thought

!Try to identify a daily status-quo experiencethat is broken

!Go to the website www.warbyparker.comand

think about how this brand executes authenticity,

transparency, and humanization to its various

stakeholders


14/89

Go To Market Strategies:Reputation and Reviews

David Bell



15/89

Overview

!Motivation

!Theory: How Information Affects Market

Efficiency

!Research Examples: Amazon vs. Barnes andNoble and Trip Advisor vs. Expedia


16/89

Motivation


17/89

!It takes many good deeds to build a goodreputation, and only one bad one to lose it (BenFranklin)

!Reputation is an idle and most false imposition;oft got without merit, and lost withoutdeserving (William Shakespeare)

!But apparently > 60% of users read reviewsbefore making a purchase and positive reviewsincrease conversion rates


18/89

Sites To Check

!Lots of sites for

!Vacations, restaurants, cars, movies, andcontractors but also for other things as well

! http://www.zocdoc.com/! http://www.harmonycentral.com/

! http://www.flyertalk.com/

! http://www.epicurious.com/

! http://www.makeupalley.com/

!In addition


19/89

Motivation and Preliminaries

From Chris Dixon in 12/2011

Great startup story. Raised a total of $4.2m in venturecapital, sold to _____ for $210m, and had some

interesting adventures and pivots They started outtrying to aggregate reviews from other websites andwhite label their product. _____ was just a showcasethat accidentally became a destination site. As of today

_____ is an independent public company with amarket cap of $3.5B (now $6.35B)


20/89

Theory: How Information AffectsMarket Efficiency


21/89

Theory

!More information to consumers is almostalways better (economists and policy markersnormally support policies to make more info

available)

!In addition, when information is provided byfirmsit might affect their behavior as well. (How:

For better; for worse?)

!Information asymmetry (moral hazard andadverse selection)


22/89

Data*

!In December 1997 LA County passed anordinance requiring restaurants to display ratingcards (came to NYC 10 years later)

!Panel data set with:

!All inspections of all restaurants

!Quarterly revenue (via tax data)!No. of people admitted to hospital

! *Jin and Leslie The Effect of Information on Product Quality


23/89

Hypothesized Effects

!The rating cards will help mitigate informationasymmetry

!

If this happens then

!Demand at good places should go up

!Demand at bad places should go down

!Restaurants themselves might improvequality (if the benefit exceeds the costs)

!Note too, that ratings reduce search costs for

consumers


24/89

!Demand: Sales at restaurants with an A wentup 5.7%; Bs went up .7%, Cs went down 1%

!Objective quality went up: Hospitaladmissions for food borne illnesses weredown 13% in LA, but up 3.2% in the rest of

CA!

Findings


25/89

!Review systems change behavior!

!Ideally, reviews should be objective and

verifiable

!But this is not always the case (see nextexamples)

Principles


26/89

! On average, reviews are positive (4.14 at Amazon and4.45 at BN)

! One-star reviews are pretty rare (7%, 3%) and five-starreviews are abundant (57%, 67%); Amazon reviews are

longer

! Key Findings

! Consumer WOM affects sales (helps Amazon?)

! Better reviews on a site increase sales at that site! The negative effect of a one-star review > positive

effect of a five-star review

! The text of reviews are actually read

Reviews on Amazon


27/89

Specifically

A few years ago, Mr. Rutherford, then in his mid-30s, hadanother flash of illumination about how scarcity opens thedoor to opportunity.

Suddenly it hit him. Instead of trying to cajole others toreview a clients work, why not cut out the middleman and

write the review himself? Then it would say exactly whatthe client wanted that it was a terrific book. A shatteringnovel. A classic memoir.


28/89

And

In the fall of 2010, Mr. Rutherford started a Web site,GettingBookReviews.com. At first, he advertised that hewould review a book for $99. But some clients wanted achorus proclaiming their excellence. So, for $499, Mr.

Rutherford would do 20 online reviews. A few peopleneeded a whole orchestra. For $999, he would do 50.

There were immediate complaints in online forums that theservice was violating the sacred arms-length relationshipbetween reviewer and author. But there were also orders,a lot of them. Before he knew it, he was taking in $28,000a month.


29/89

Then

It turns out all those fake-sounding reviews on Amazon.comprobably are. Bing Liu, a data-mining expert tells the NewYork Times that about one in three online reviews are fake.The reason: there's a lot of money in fake reviews,

according to an excellent expos by the New York Times'sDavid Streitfield.

Read more: http://read.bi/OiuMh0


30/89

! Reviews could be helpful

! But, authenticity is a concern (data-mining for the truth?)

! What about an alternative approach

!Approach and Key Findings

! Examine the distribution of reviews (histograms)

! Net gains should be highest for independent hotelswith single-unit owners

! Such hotels have more five-star reviews on TA

! Neighbors of such hotels have more one-star reviewson TA!

Reviews on Trip Advisor


31/89


Lead UsersProduct Life Cycle

David Bell



32/89


33/89


Influence and How Information Spreads

David Bell



34/89

Overview

!Controversial Study on Obesity

!Elements of Networks

!Elements of Neighborhoods

!Four Research Studies on Influence andContagion through Networks and Neighborhoods


35/89

Four Research Studies

!Neighborhood: Internet retailing (diffusion at

www.netgrocer.com)

!Network: Influence among users at a socialnetworking site

!Network: Prescribing behavior of physicians in

Los Angeles

!Neighborhood: Internet fashion retailing (diffusion

at www.bonobos.com)


36/89

Controversial Study


37/89

Please Watch

http://www.youtube.com/watch?v=pJfq-o5nZQ4


38/89

Elements of Networks


39/89

Background

!Pathways through which information, advice,

resources, and support flow between peopleAral 2012)

!Networks can be physical or virtual

!Networks usually exhibit homophily, either in

characteristics of participants, or in preferences


40/89

Background

!A network can be a simple as a dyad, e.g., two

partners, or so complex as to encompasshundreds or thousands of people

!Nodes (people)!Connections (between people)

!Dynamic behavior as ties form and break

!Influence, homophily, confounds

!Constraints (geography, socioeconomics)

![More Backgroundhttp://www.youtube.com/watch?v=2U-

tOghblfE]


41/89

Entering a Network

!Participation in a network is a choice (often

governed by some form of homophily)

!We also decide how manypeople we want toconnect to and how central we want to be

!When our relationships are transitive (our friendsknow each other) then we are deeply embedded


42/89

Separation and Influence

!1960s: Few hundred business people send

letters from Nebraska to Boston

!2002: 98,000 people send emails to unknowntargets

!Six degrees of separation

!Three degrees of influence (decay, instability)


43/89


Elements of Neighborhoods and Examples

David Bell



44/89

Connections

!The unit of analysis could be a zip code or city

block (or even a single individual)the methodis agnostic to the unit of analysis

!Define a neighborhood

!After defining the neighborhood, define thecontiguity relationships among neighbors

!Next, define the activities that take place


45/89

First Order Contiguity

!Imagine a neighborhood of 4 zip codes (there

are several thousand such neighborhoods inthe US)

!Define a 4 x 4 matrix that shows which zip codes

are connected (share a boundary)

!File in the matrix with 1 (connection) or 0 (noconnection

!Normalize if need be


46/89


*Bell and Song Neighborhood Effects and Trial on the Internet


47/89


!Now, lets imagine that we are interested in the

chance that something happens in a focallocation given that nothing has happened yet,

e.g., the chance that someone buys a pair of

Bonobos pants

!Notice that the activity (a purchase of Bonobospants) has already happened in zip codes 1 and

2, but not in zip codes 3 or 4

!So, lets focus on zip code 3


48/89

Beginning of Technical Material


49/89

Zip Code 3

!The relevant row is row 3 of the contiguitymatrix, C; lets call that wzbecause it serves asthe weight

!There is only one vector, Y(t-1), which captures

the cumulative prior activity, i.e., contains 1 ifsomething happened, and 0 otherwise

!wz= 1 (row) x 4 (columns); Yz(t-1) = 4 (row) x 1

(column)

wz =

13, 1

3,0, 1

3!"

#$

Yz(t!1)' = 1,1,0,0[ ]


50/89

Identifying Influence

!So, how much influence is being brought to

bear on zip code 3?

!We can get the answer by multiplying the row

vector of weights and the column vector ofactivities

!The product is 2/3 which is intuitive

[wzY

z(t!1)]


51/89

Identifying Contagion

!Some econometrics (optional)

!By definition, the utility of the maximum utilityperson in zip code zat time twas defined as

Vz(t)which is now augmented as

!Where the !"#!$is a measure of contagion

and the part in [ ] is between 0 and 1

Vz'(t) =V

z(t)+![w

zYz(t"1)]


52/89

End of Technical Material


53/89

Research Studies on Influence and

Contagion in Networks andNeighborhoods


54/89

Study One

!Neighborhood to neighborhood influence in trial

of an Internet retailer www.netgrocer.com


55/89

Aggregate Space-Time Sales Pattern

(Shaded areas: Zip codes and at least1 customer within the cumulative time

period)

Some Motivation


56/89

6 months


57/89

18 months


58/89

30 months


59/89

42 months


60/89

A Closer Look


61/89

Month 1


62/89

Month 2


63/89

Month 3


64/89

Month 4


65/89

Month 5


66/89

Month 6


67/89

Month 6


68/89

Month 45


69/89

Results on Contagion

!In the study, we found that the influence

parameter was positive and statisticallysignificant, even after controlling for all the things

that might affect homophily (e.g., demographics)and other unobserved factors

!The statistical effect had an economic impact:

The average zip code has about 8,700 people

and 5.6 neighboring zip codes if another20,000 neighbors tried the serve the probabilityof focal zip code without customers trying goes

from about 2.7% to 14.0%


70/89


More Examples of Influence

David Bell



71/89

Study Two

!Network: Influence among users at a social

networking site


72/89

Observations*

!ConnectionsWhile we may have numerous

friends only a fraction carry a lot of influence

!

Scale ProblemBecause there are lots offriends and links, inferring who is influential for

whom is difficult

!GoalInfluence-based targeting

! *Trusov, Bodapati, and Bucklin Determining Influential

users in Internet Social Networks


73/89

Data and Approach

!Activity LevelsDevelop a simple metric of

influence based on whether activity of onemember influences another

!Log In ActivityNumber of check ins per day

(but could easily use other metrics)

!LogicUsers check in to consume and createcontent


74/89

Conjectures

!InfluentialIf a member increases usage and

connected others do too, the member isinfluential

!Non-InfluentialIf a member increases usage

or decreases usage and there is no change in

the behavior of connected others, the user is notinfluential


75/89

Findings

!Influence InIndividuals are influenced by

about 1/5 of their friends and 1/3 of users haveno influential friends

!HeterogeneityWide variation but some users

with similar numbers of friends have total

network impacts that differ by a factor of about 8

!RegularityRegular users (higher meanactivity, lower variance) learn about their friends

more quickly


76/89

Influence

![Influence of friend f on user u]


77/89

Influence

![Regression involving influence parameter]


78/89

Implications

!Simple MetricsFriend counts and profile

views are inadequate proxies for influence

!

AdvantageMost of the payoff from the modelcomes from superior identification of the best

customers

!RetentionIf top users defect there is adisproportionate negative effect on the network


79/89


More Examples of Influence

David Bell



80/89

Medical Research Findings


81/89

Medical Research Findings

!Self-reported opinion leadership versus ties

among physicians wij= 1 if physician imentionsphysicianjas a source of influence

!Wharton study of physicians prescribing drugs inLA, NYC, and SF

!Self-reported opinion leaders adopt sooner;

however, knowledge of ties (socio-metricstructure) is more important


82/89

Implications

!The firm found that it was helpful to understand

the network structure as unexpected leadersemerged

!Contagion was at work and very important in the

diffusion process

!Based on the success of the study in the US,The firm is continuing with socio-metric studiesin several cities in China


83/89

Study Four

!Neighborhood: Internet fashion retailing (diffusion

at www.bonobos.com)


84/89

Social Capital (see Bowling Alone)

!Ability of individuals to secure benefits due to

trust, cohesion, and reciprocity; based onfrequency and quality of interaction in a

community

!Higher levels of social capital lead to moreefficient transfer of information

Social Capital


85/89

Important Conditions

!Apparel category has non-digital attributes

!Focus on new trialsconsumers have

incomplete knowledge ex ante

!The product is socially visible

Study Features


86/89

H1

H2

Data


87/89

H1: Social Learning

! Social learning helps resolve the problem of

incomplete consumer knowledge about non-digital

attributesthis increases pre-trial expected utility

! Findings:

! Expected utility evolves with acquisition of signals

! Initial belief underestimates true quality (p < .001)

! No evidence of risk aversion (r = -.001;p = .217)

! Expected utility weakly increasing with signals


88/89

o 2,781 / 5,745 (around 50%)of trials would not havehappened if no additionalinformation about non-digitalattributes were transmitted

through social learning

o More of the later triersaredriven by better informationabout non-digital attributes

that is spread throughcommunication with earliertriers (link to TALC anddiffusion theory)

Implied Economic Value (H1)


89/89

What could / should www.Bonobos.comdo?

!Q: Can you think of a good proxy for offline

social capital?

!Q: How could / have other firms used this

idea (implicitly)?

Implications for the Firm

week8_all merged pdf

Documents