2. Rick Erwin
President, Targeting,
Experian Marketing
Services
Joyce D’Addio
Digital Marketing Director
Sallie Mae, Upromise
Jennifer Astle
Sr. Director, Channel Mktg
Sallie Mae, Upromise
Your presenters
Bob Baer
Director of Online
Acquisition
Sallie Mae, Upromise
3. Agenda
• What’s old is new
• Who is Upromise?
• Upromise’s business challenges
• Why Experian?
• Modeling and optimization
• Display campaigns
• Campaign KPIs
• Results
4. Then Now
List Audience
Service Bureau DMP/Onboarding Partner
Printing & Letter shop DSP
Direct mail Digital Media Impressions
What’s old is new
5. Marketers: People:
Delivering
• the right offers
• at the right time
• to the right person
Searching
• what they want
• when they want it
• where ever it may be
Direct
Marketing
Digital
Marketing
7. Upromise by Sallie Mae
• Upromise is a simple, yet powerful cash back rewards
program where members can turn everyday spending
into money for college.
• 15MM+ members; $700MM+ in member savings to date
• Mission: “To help people save, plan and pay for college”
• Member composition:
• Parents of high school students
• Parents of young children
• Students
• Free to join
8. How members can save
• Earn 5% cash back on online purchases with 900+ retailers & travel
agencies
• Earn cash back when dining, getting groceries, or filling up your tank at
specified retailers
• Earn 1% on every purchase using a Upromise credit card
What members can do with their earnings
• Keep in Upromise account
• Transfer Upromise savings into a 529 plan
• Use Upromise savings to pay down a
Sallie Mae student loan
• Transfer Upromise savings into Sallie Mae high-yield savings account
Upromise Products
9. Upromise Business Model
How does Upromise make money?
• When a member transacts with a Upromise partner, the
partner funds the member’s savings and Upromise revenue
Who is Upromise’s ideal customer?
• Joins and stays engaged with the rewards program
• Adopts 1 or more financial products
11. GoalCurrent status
Find the right customers @ the
right time who will transact
consistently and engage with 1 or
more Financial Products
Many people signing up
for free service, but not
transacting regularly
Revenue increased by
150%
Year over year
Key Challenge
12. From casting a wide net… To honing in on the ideal target
Low Engaged Members
Grocery
Affiliates
Prospect
Email
Key Question
• How do we find our ideal target and increase the odds that the prospect will
become an engaged member?
• “Ideal” Upromise member
- Transacts frequently with our partners
- Eligible for financial products
- Skews toward higher credit quality with above average levels of
disposable income
13. Database
Marketing
Concepts
• Apply database marketing concepts and strategies to online display
• Powerful targeting capabilities available by combining traditional
marketing strategies and new technology
• Advertise to your ideal prospect where they are, not where you think they might
be
Traditional Offline Marketing Goes Digital
15. Why Experian?
• Confidence in data quality and analytical capabilities
• Breadth of data sources
• Reach of ad network
• Similar privacy and compliance standard
17. Define High Value Customer within
Sallie Mae CRM1
Apply to ConsumerView and identify 200-300
variables that best define audience2
Create model with these variables.
Rank entire US population3
Target top deciles online
4
Display Modeling Process with Experian
18. Best
customer
Find more
of them
Identify and
profile your
best customer
Getting Started
• Use key criteria for defining
“best customer”
• Create model using
Upromise’s existing
customer base
• Build lookalike models
to find other people who
look like Upromise’s
best customers
19. Model Objectives
Target individuals with the following attributes:
• Prospects with a high potential to be active
Online Mall purchasers
• Prospects with a high potential to develop into
Financial Product consumers
• Customers who have been enrolled since 2010
and have had online mall purchase since 2011
Max
(score 1,
score 2)
Sallie Mae loan
holder look-alike
model score
Credit Card
look-alike
model score
Online
look-alike
model
20. Online Mall Purchasers – Model
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
PercentofOLMCustomers
Percent of Prospects
Model Lift: Online Mall customers vs ConsumerView
Baseline Online Mall Customer model
38%
34%
72% of likely best customers are
found in top 30% of ranked data
22. Define best customer,
create models, define
audiences
1
Campaign Execution
and optimization
4Set Segment Objectives:
What does success
look like?
2
Campaign design:
creative, test plans,
media plans
3
Display Campaign Process
23. Define best customer,
create models, define
audiences
1
Defining Your Universe
1. Define your best customers
– Revenue
– Secondary product adoption
– Method and timing of acquisition
2. Create the Model
– What attributes separate your best customers from the
rest?
– What behaviors should the model predict?
3. Define your Audiences
– Life stage
– Product adoption
– Self-selected (how they use your product or service)
24. Set Segment Objectives:
What does success
look like?
2
Segmentation and Defining Objectives
1. Segment definitions should be aligned with
goals & products
2. Segment definitions need to consider size –
the larger the segment, the better yield
optimization opportunities
3. Multiple segment campaigns generally
outperform single segment campaigns
4. CPA & CPC objectives are preferred over CTR
and eCPM
25. Campaign design:
creative, test plans,
media plans
3
Campaign Design
1. Creative
– Align creative with segments
– Reinforce messaging and creative in organic channels
2. Select Ad Networks
– Consider reach and traffic volume
– NAI/DAA standards
3. Impression Frequency and Lookback Windows
– Test # of repetitions required for conversion
– Determine appropriate look back window for counting clicks
and view-thrus
26. Home Value
Online ShopperActive Credit Cards Presence of Children
Geo-locationAge & Income
Data Selects
Data Models
Execution and Optimization
Now that we know who else in ConsumerView
would be a good Upromise member, we use
Experian’s display capabilities to find them
Campaign Execution
and optimization
4
29. MES = Median Credit Score
For acquired members not exposed to
Experian campaign: Lower credit scores
For acquired members exposed to
Experian campaign: Higher credit scores
Results
31. Next Steps
1. Engage our customers through the channel(s)
they prefer and/or transact on
2. Implement a mobile strategy
3. Drive for cross-channel optimization
Rick presentsAudience is the new ListDMP is the new Service BureauDSP is the new Printing and Lettershop (this is a little nuanced and needs some explaining… in essence its where the data gets matched to the creative and readied for delivery to consumer) Ad Networks are the new Delivery System (I’m hesitant to say Postal Service but this is where the ad gets delivered to the consumer) There is also the potential to point out that the role that DMPs play in marketing audiences for use in the ecosystem is similar to the role that List Managers have played in the offline model
Jen to present
Jen to present
Jen to present
Jen to present
Jen to presentAs I said at the beginning, many people would probably to be very happy to acquire new customers for a free service. But, as we went through our business model, you saw that people need to spend money in order to save money. With a free service, we have problems finding people who will transact consistently with Upromise partners. Upromise case study:Successful at acquiring a high volume of low-cost customers but they were not engaging in the service – ROI-negativeWilling to acquire higher-cost, quality customers but struggled to find enough of themExperian’s modeling and display solutions helped solve this problemWho should Upromise target?How can Upromise find them?Current statusLots of people signing up for free service, but not transacting regularlyNeedFind customers who will engage consistently in the Upromise service (customers with high disposable income)
Jen to present
Bob to presentApplying Database Marketing Concepts and Strategies to Online DisplayPowerful targeting capabilities Advertise where your prospects are, not where you think they might beApplying Database Marketing Concepts and Strategies to Online Display Direct Mail has been using these concepts for decades, cookie matching, addressable media have provide the tools to do this onlinePowerful targeting capabilities We can take the lookalike model, identify the cookies that we want to target, and show them Upromise display banners as they browse the webTarget people, not sites Significantly broader reach than searchAdvertise where your prospects are, not where you think they might be Pure demographic targeting can be unreliable, or at the least, unpredictable Advanced targeting allows you to increase both reliability and predictability
Bob to presentNotes from our conversationPreviously had to wait months to get/see engagement (campaign success)Upromise now knows right away by seeing activity and engagementProvides proof that we are targeting the right people
Jen to presentConfidence in data quality and analytical capabilitiesUse their data and analytical services for internal custom modelsKnew what to expect from a data quality and service level standpointBreadth of data sources Subsidiaries, segmentations, summarized credit statisticsReach of ad networkTop 2 Display Advertiser 2011Over 140M cookies linked to our US Consumer database of 120m householdsWe currently partner with 4 of the top 5 publishers and dozens of digital partnersSimilar privacy and compliance standardsWe all know how fun legal and compliance can be in pushing through new initiatives
Joyce to present
Joyce to presentUse key criteria for defining “best customer”Create model using Upromise’s existing customer baseBuild lookalike models to find other people who look like Upromise’s best customersConsumerView has thousands of variables per person235 million AmericansDeciles – split evenly in groups of 10. or 10%
Joyce to present
Joyce to presentOn the last bullet – mention the reason for 2010 & 2011 was due to acquisition strategy changes. Customer booked prior to that time look different than after. It also gives us recency of data and customer performance within the same economic cycle
Joyce to presentThe model is a powerful tool to find Upromise Online Mall Prospects among 116 million ConsumerView households72% of likely active Upromise online mall customers are found in top 30% of ranked dataChart demonstrates the lift performance of the Online Mall shopper model. These data are from the “holdout” data set...it’s a completely separate sample from the data the model was built on, but it’s drawn from the same population. This is a standard method of validating model results to make sure that the results are reliable. The curve represents the cumulative percentage of customers (vertical axis) found within the cumulative percent of the ConsumerView population (horizontal axis). The straight diagonal line is the baseline—if we didn’t use any models and made random selections from the population, we would expect to find 50% percent of customers at 50% percent of the population, 10% of customers at 10% of pop., etc. It’s used as a null hypothesis ---how the customers would be distributed if the model added no predictive power at all. Visually, then, the area between the curved line and the straight, diagonal baseline represents the lift performance of the model. If the model were not effective, we would see little or no separation between the curve and the baseline, or we would see the curve cross back and forth across the baseline.
Bob to present
Bob to presentDefine segments (done by lookalike model)Set KPIs for each segment and campaignDesign the campaignLaunch the campaignSelf-optimizingOriginal slide:Define segmentsSet segment objectivesHigh valueNet new membersUpsell additional productsCampaign DesignMedia selectionCreative developmentSet KIPs, test plan and yield optimizationCampaign ExecutionFlexible flight datesSeasonally adjusted spendWeekly campaign reportingMeasurement reports
Bob to presentDefine segments (done by lookalike model)Set KPIs for each segment and campaignDesign the campaignLaunch the campaignSelf-optimizingOriginal slide:Define segmentsSet segment objectivesHigh valueNet new membersUpsell additional productsCampaign DesignMedia selectionCreative developmentSet KIPs, test plan and yield optimizationCampaign ExecutionFlexible flight datesSeasonally adjusted spendWeekly campaign reportingMeasurement reports
Bob to presentDefine segments (done by lookalike model)Set KPIs for each segment and campaignDesign the campaignLaunch the campaignSelf-optimizingOriginal slide:Define segmentsSet segment objectivesHigh valueNet new membersUpsell additional productsCampaign DesignMedia selectionCreative developmentSet KIPs, test plan and yield optimizationCampaign ExecutionFlexible flight datesSeasonally adjusted spendWeekly campaign reportingMeasurement reports
Bob to presentDefine segments (done by lookalike model)Set KPIs for each segment and campaignDesign the campaignLaunch the campaignSelf-optimizingOriginal slide:Define segmentsSet segment objectivesHigh valueNet new membersUpsell additional productsCampaign DesignMedia selectionCreative developmentSet KIPs, test plan and yield optimizationCampaign ExecutionFlexible flight datesSeasonally adjusted spendWeekly campaign reportingMeasurement reports
Bob to present
Bob to present
Bob to presentNotes from our conversationPreviously had to wait months to get/see engagement (campaign success)Upromise now knows right away by seeing activity and engagementProvides proof that we are targeting the right people
Bob to presentThe model is generating members who are much more likely to be approved for the Upromise credit card.
Bob to presentThe number of people engaging in the online shopping program is nearly double our prior rates so shopper volume has remained the same despite decreasing the overall acquisition volume.The number of shoppers in 2012 and 2013 is the same. However, the spend / shopper from the 2013 shoppers is 60% higher because the new members are modeled after our best online shoppers. Our best existing shoppers had high household income. The Experian lookalike model was able to find more people who had high household income (and other attributes similar to these best existing shoppers). Even though we’re spending more for high quality, this is ROI positive and we will be continuing with this strategy.
Bob to present(Previous slide included) Apply models to other channels – email, mobile, direct mail, site placements. Expand to other products in the portfolio and among other business units Continue to optimize the model Increase lifetime value of customersIncrease frequency of transactionsDrive loyaltyIncrease additional product adoptionCompress purchase cycle time lineBob to presentNext Steps: Score ConsumerView households to make targeted group available for digital display advertising Consider use of scored prospects for other marketing channels (direct mail, e-mail, etc.) Consider use of model for targeted marketing efforts to existing Upromise customers Consider building additional models to select targets for other Sallie Mae products within different marketing channels and populations