David Lamb, Senior Consultant, Target Analytics DATABASE SCREENING.

David Lamb, Senior Consultant, Target Analytics

DATABASE SCREENING

Why Warren Buffet is not a prospect for you!

Your best prospects are already on your own database

Corollary: Just because someone is wealthy doesn’t mean he’ll give anything to you

Database screening provides a way to filter your databaseWho is more likely to giveWho is more capable of giving

Two basic approachesStatistical modelingList matching

Strange History Of ModelingEarly openness to the technology – late 80sThe disappointment factorDevelopment staff often misunderstood the

purpose of data mining“These people can’t give”“These people won’t give”“Who are these people?”

New improvements in wealth screening resulted in disillusionment with modeling in early 90s

Modeling ReconsideredModeling addresses the “interest” and

“linkage” pieces of the major gift puzzleThrew out the baby with the bath water?

A Mathematical Way To Predict Behavior

Annual givingPlanned giving

BequestsAnnuities Trusts

Major givingGift size

Patient responseMail responsePhone donorsMail donorsElectronic donors

Method Behind The ModelMost donors have certain things in common

Minimum requirements for a modelData qualityData quantityCollection over time

Types of modelsRFM analysisHome grown modelsVendor supplied models

• Giving pattern• Age• Participation• Major• Degree• Location

• Volunteer• Income• Credit use• # of children in the

home• Constituent type• Real estate value

Modeling TrapsNot every variable that seems predictive isSome variables with correlations may not be

“weighty” enough to influence a person’s score

Endogenous variables are caused by the behavior you’re trying to predictYou collect email addresses on those who are

closest to youPresence of an email address is correlated with

givingNot predictive!

Data QuantityWhat size gift is “major”Must have at least 200 examples of gifts in the

last year at a particular level for valid statistics

Don’t include gifts from corps or founds

Gift Level Gift Count % of totalCumulative Count Cumulative %

0 Dollars 109,135 90.85 109,135 90.85

1-49 Dollars 2,497 2.08 111,632 92.93

50-99 Dollars 1,902 1.58 113,534 94.52

100-249 Dollars 3,867 3.22 117,401 97.73

250-499 Dollars 1,153 0.96 118,554 98.69

500-999 Dollars 728 0.61 119,282 99.30

1000-2499 Dollars 582 0.48 119,864 99.79

2500-4999 Dollars 113 0.09 119,977 99.88

5000-9999 Dollars 72 0.06 120,049 99.94

10000+ Dollars 73 0.06 120,122 100.00

One year gift table

RFM AnalysisRecency – when was the most recent gift?

Score 0 if more than 3 years agoScore 1 if 3 years agoScore 2 if 2 years agoScore 3 if 1 year ago or less

Frequency – how consistently has the donor given?Score 0 if none of the last three yearsScore 1 if only one of the last three yearsScore 2 if only two of the last three yearsScore 3 if each of the last three years

Monetary Value (must be customized)Score 0 of largest gift is $0Score 1 if largest gift is $1-$999Score 2 if largest gift is $1,000 – $4,999Score 3 if largest gift is >= $5,000

RFM AnalysisIf a prospect scores >= 6

Top priority for additional research to estimate capacity

Consider the person a high likelihood prospectIf a prospect scores 3 – 5

Second priority for research to estimate capacity

Consider the person a moderate likelihood prospect

If a prospect scores 0-2Do not do additional research unless specific

indicators come to lightConsider the person a low likelihood prospect

Home Grown Model MaterialsSoftware like SPSS or SASStatistical education Data sources or enhancements if you plan

to use info beyond your databaseAge overlayAddress updatesOther datasets (census data, marketing data)

Home Grown Models: Frequency Distribution

Major DonorsVariable

Entire Population

% yes % no % yes % no11 89 Income > $100,000 3 9714 86 Current or past parent 15 8525 75 Live within 50 miles 83 1731 69 Home value > $500,000 14 8640 60 Age > 50 37 6359 41 Attended >= 3 events 22 7866 34 Alumni 70 3084 16 Gave in each of the last three

years 9 91

95 5 Credit is in satisfactory status 90 10

Frequency Distribution ResultsYour best major gift prospects

Close callsAge?Alumni?Satisfactory credit?

Yes = 1, No = 0Have incomes > $100,000 1Live more than 50 miles from your institution 0Have home values > $500,000 1Attended >=3 events 1Gave in each of the last three years 1

4

Problems With Frequency DistributionNot all variables have equal weight

Major DonorsVariable

Entire Population

% yes % no % yes % no11 89 Income > $100,000 3 9714 86 Current or past parent 15 8525 75 Live within 50 miles 83 1731 69 Home value > $500,000 14 8640 60 Age > 50 37 6359 41 Attended >= 3 events 22 7866 34 Alumni 70 3084 16 Gave in each of the last three

years 9 91

95 5 Credit is in satisfactory status 90 10

More Statistical PowerMust run correlations between your

dependent variable (major giving) and all available independent variables

Regression analysis allows you to compare strength of correlation of the variables in relation to each other

Weighted correlations yield a score for each person

Statistical package like SPSS or SAS will facilitate calculations and reporting

Vendor Statistics – 3 OptionsGenericPrescriptiveCustom

Generic ModelStatistical profile of the entire countryAggregate data based on many datasetsNot specific to any one organizationGeo-demographic data is in this category

Claritas Prism Clusters Gold Coast Blueblood Estates Shotguns & Pickups

Equifax Niches Chic Society Diamonds to Go Oodles of Offspring

Prescriptive ModelHybrid of generic and a custom modelIncludes standard variablesIncludes donor-specific giving dataProvides a more targeted analysis than

generic for who gives to an organization like yours

Custom ModelConstructs a profile of giving behavior

unique to your organizationExamines the people in your database who

have done what you’re trying to predictCapability to contributeLikelihood to make a gift

Compares those people to the ones who did not behave the same way.

List Matching – aka Wealth ScreeningAn automated process that matches the names on

your database to those on other databasesPublic company insidersPrivate company owners & officersReal estateBiographical sourcesDonor listsAny list in electronic form

Simple minded, but fastYields the same kind of specific research that

human researchers seek to findInformation returned requires verification

It’s Not A Perfect WorldMatching issues

False matchesThings you can find easily are not picked up

Sources of matching errorSource data are flawed and incompleteSource data are chaoticProgramming issues are not trivial

Modeling issuesGood at describing groups, not individualsEndogenaity Sample size

Comparing the approachesResults Per Record

CostEmphasis

Modeling A score or other indicator about every record on the database

Lower •Prospect Identification

•Inclination & Linkage

List matching

Specific information about a few very capable prospects

Higher •Prospect qualification

•Capacity

What’s Right for Me?

You should do list matching if:You have a well established major gift operationYour constituents are wealthyYou are located in the midst of wealthYou need to qualify people for gifts of at least

$25,000You should do modeling if:

You want to segment your entire databaseYour major gift operation is less developedYour constituency is unlikely to be in listsYou need to improve your annual fund strategyYou need to improve your planned gift strategy

Best Of Both WorldsIdeal approach is to pre-screen the

database with one method, then go deeper with anotherStart with a modelDo list matching with records that score well

On a pre-screened database, 1 in 10 may end up looking like major gift prospects.

If you need 4,000 major gift prospects, screen 40,000 constituents