Tiffany Neill, CFRE

1
Mining For Gold Using Modeling to Find Hidden
Treasure

• Presented by
• Tiffany Neill, CFRE
• Amy Sukol, CFRE
• Lisa Drane

2
Todays Structure

• 5 Techniques
• In house to out of house
• Varying expense
• For Each Technique
• A case study
• Using the technique
• Results
• Question Answer

And as the tenth bullet states
Jessica Hagy, Indexed, December 2006
3
Before we begin What do we mean by modeling?

• Using characteristics about people in
  combination to find those most likely to behave
  in a certain manner.

4
Before we begin What can modeling do?

• Help identify groups that may be responsive to
  being treated differently
• In return you can
• Raise More money
• Mail More Cost Effectively
• Invest more in Donors with a higher Lifetime
  value
• Improve on retention and upgrades

5
Before we begin What cant modeling do?

• Replace testing and best practices
• Fix problems that dont exist
• Make up for a weak case for support

6
The Techniques

• Segmenting beyond RFM
• Using your merge/purge
• Using other internal lists
• External data and mail
• External data and telefundraising

In house
Need Help
7
Beyond Recency, Frequency and Gifts Simple
Modeling With Big Results
8
Selecting donors based on things other than
recency, frequency and gift amounts

• Two Challenges
• Organization 1 mailed an annual calendar
  fundraiser to 75,000 people with declining net
  income
• Organization 2 needed to raise 100,000 for a
  building campaign without reducing unrestricted
  revenue

9
How were additional factors used in segmentation?

• Organization 1 added characteristics to
  segmentation
• Has the person responded to a prior calendar?
• Has the person responded to another premium?
• Have they been given the opportunity to give to a
  premium?
• Created a non-premium mailing to send to others
• Selected far fewer and different people than RFM

10
What were the results?
11
Additional Considerations

• People receiving the regular appeal were offered
  a calendar as a back end premium (additional cost
  6,000)
• The next year people who requested the back end
  calendar were included in the main calendar
  mailing

12
A Second Example

• Organization 2 split data in new ways
• Added value of donors to segmentation
• Added past behavior in special funding campaigns
• Added deep lapsed, high dollar and other
  categories not normally selected for appeals
• Mailed more overall with a different split

13
What is value?

• Looked at total giving over a persons life and
  how it related to the investment made on
  acquisition
• To simplify selection used classes of donors
  based on what package they had been acquired on
  (i.e. control used from 1988 1994 etc.)
• Factored in non-mail gifts people had made
  (planned gifts, major gifts, event gifts)

14
What were the results?
15
Additional Considerations

• Acknowledgements and the database had to be
  segmented to respect the gift source
• The restricted gift was not factored in when
  later selecting last gift or highest gift since
  it was a different request
• Messaging is still key!
• This worked because the unrestricted offer was
  very compelling

16
Merge/Purge Youre Already Modeling and May Not
Know It!
17
Your merge/purge can tell you a lot about your
donors!

• Who is still giving?
• Are you their only cause or one of many?
• How does their other giving effect how they give
  to you?
• What about your suppression file?

18
How one organization used merge/purge to improve
appeals

• Mid-size social service organization with
  underperforming donor file.
• As part of the merge, the house file and
  suppression file were run against 30 outside
  lists.
• Appeal segmentation was based on the number of
  hits against the outside lists.

19
What was revealed in the merge

• List Segment of File Response Av. Gift
  CTRAD
• Donor Only 39 3.43 42.83 0.17
• Donor 1 hit 22 4.08 34.70
  0.17
• Donor 2 hits 28 5.09 32.40
  0.15
• Suppression Hits 11 6.53
  49.27 0.08

20
What did the model reveal?

• List Segment of File Response Av. Gift
  CTRAD
• Donor Only 39 3.43 42.83
  0.17
• Those giving only to this organization had the
  lowest response rates but highest average gifts.
  This group can be targeted for upgrade efforts as
  their most loyal donors.

21
What else did we learn?
List Segment of File Response Av. Gift
CTRAD Donor 1 hit 25 4.08
34.70 0.17 Donor 2 hits 32
5.09 32.40 0.15 More than half of
their donors are actively giving to other
organizations. This means higher response rates
but lower average gifts and more competition in
the mail.
22
Other things Learned

• List Segment of File Response Av. Gift
  CTRAD
• Suppression Hits 11 6.53
  49.27 0.08
• Check the donors on your suppression file! If
  they have preferences that are more than three
  years old, you could be suppressing potentially
  responsive donors such as these.

23
Using merge/purge to determine regional
preferences

• Most regional organizations believe that closer
  is better.
• Most national organizations assume that point of
  service doesnt matter.
• Both may be wrong!

24
How one group improved regional performance

• Small New York cultural organization.
• Insisted on limiting zip selects to those
  immediately surrounding the organization.
• This severely limited their list universe.

25
What the merge revealed

• Zip/Area of Mailing of Gifts
• Org. Zip 8 15
• Close-in Zips 69 56
• NY Suburbs 9 22
• NJ and CT 2 8

26
How was the knowledge used

• Mail only donors in the suburbs and others
  states - No!
• Expand zip select to include donors further
  from the organization Yes!
• Most of all merge/purge is an effective
  tool with which to gain this
  information.

27
Modeling Tools in the Database Next Door
28
Look for Hidden Assets Identifying and
Cultivating Internal Prospects

• Lapsed Donors
• Non-direct Mail Donors
• Volunteers
• Former Participants/Patients
• Anyone else living on your database!

29
Psst what about those other donors?

• Small Regional senior services provider
• Donor base was 50 direct mail acquired and 50
  non-direct mail acquired
• How can we cultivate these non-direct mail
  acquired donors?

30
How those others performed
Mail Segment Response Average
Gift DM in Appeal 9.84 41.74 Non-DM
in Appeal 0.88 64.10
31
But wait
Mail Segment Response Average Gift
Non-DM in Appeal 0.88 64.10 Non-DM in
Acquisition 4.43 46.13
32
What did they learn?

• Non-direct mail acquired donors made excellent
  direct mail prospects!
• Over the life of the program, direct mail has
  been used to introduce an additional channel of
  giving to these donors
• This has lowered the cost of acquisition and
  increased the lifetime value of the non-direct
  mail donors!

33
Using Outside Services for Modeling to Improve
Mail Performance
34
Three Studies

• Using outside data to Improve Acquisition
• Using outside data to Improve Retention
• Using data to Improve Upgrades

35
First, A Note of Caution

• The Best things in Life are Free
• but data costs.
• MUCH LESS THAN IT USED TO!
• Frequently, you will need to give to get
  consider the ethics and internal policies
• Good things come to those who wait
• Build extra time into your schedules

36
External Modeling and Acquisition

• The problem Acquiring too many low dollar
  donors
• 15 of new donors acquired through acquisition
  gave under 5.00
• Creative package treatment helped a bit, but they
  werent upgrading or renewing
• Needed to keep them out of data in the first
  place

37
What they did

• Used the Target Analysis cooperative database
  before mailing acquisition
• Data base contains information on 190 million
  households Enhanced with data from 500
  non-profits with 70 million unique donors and 1.5
  billion individual donations
• Sent post-merge, continuation lists to Target, so
  donors who give under 5 habitually to other
  lists could be identified segmented separately
  in the acquisition
• Almost 5 of prospects (across 35 lists) were
  identified as low dollar donors.

38
What were the results?

• Didnt eliminate the optimized names in the
  first mailing to test the process it worked!
  Next time, eliminated those folks and are working
  on strategies to work with higher dollar donors

39
External Modeling and Retention

• The problem RFM selection on large lapsed file
  had declining results
• It was requiring an investment to reactivate
  people who had not made gifts in the last 48
  months
• Same Target Analysis Coop (as in previous model)
  had helped some
• Had a full linear regression model done on lapsed
  names

40
What they Did

• Sent all data to a third party vendor (selected
  through an RFP) who built a predictive model
  based on more than 50 variables
• RFM part of it
• Time between gifts
• Packages on which people were acquired and gave
  before lapsing
• Upgrade amounts and time between upgrades
• Compared to known reinstatement people to find
  similarities

41
What were the results?

• Tested the records selected by the model against
  a control group not modeled.
• Model which looked for most cost effective
  reinstatement clearly factored gift amounts in
  highly
• When model redone, weighed response rate more

42
External Modeling and Upgrades

• The problem an organization had a donor willing
  to match any gift of 250 or more within a 3
  month period.
• Using traditional RFM analysis, the group only
  identified 2,714 prospects who were appropriate
• Used an outside firm to model entire database to
  find additional people with the capacity, the
  inclination and donor behavior similar to those
  who were selected

43
What were the results?

• Identified another 830 people with very likely
  possibility of giving at that level and 4,400
  likely possibility.
• Generated an additional 100,000 for the
  organization

44
Using Outside Services for Modeling to Improve
Calling Performance
45
DIRECT TAGS and TARGETED TELEMARKETING LIST
MANAGEMENT
The Direct Tag System can help to increase
response rates and net revenue by identifying the
best prospects for phoning.
Problem TM Results declining each year, when
need is to raise more or maintain results from
the previous year.
Solution Find creative ways to mine the data
and identify the most responsive TM records
46
Each membership/donor record is tagged with a
unique code based exclusively on prior
telemarketing behavior.
Allows us to overlay with typically lower
performing segments, honing in on responsive
prospects that may otherwise be excluded from
calling -- enabling more complex and strategic
list management decisions to be made.
Original TAGS
Direct Tag Rankings are determined by
responsiveness (in relation to campaign goals)
for every campaign conducted to date for the past
two years.
T1GO T4GO T6GU T2GO T5GO T7GU T3GO T5GU T8
47
BENEFITS 1) Helps increase response rates and
net revenue by identifying the best prospects for
calling. 2) Allows DAM to call more effectively
into typically lower performing segments by
analyzing the tags within the other established
segments. Peeling away the prospects that we have
flagged as not TM responsiveand honing in on
responsive prospects that would have otherwise
been suppressed because they were part of a low
performing segment. 3) Also, Direct Tag ratios
can be general indicator of the overall
difficulty of a campaign, and assist with
determining messaging, strategic caller style,
and approaches to list management.
More
48

MORE BENEFITS and interesting FACTS
4) DAM can rank lapsed donors who will respond
to a phone call, reactivating these donors in the
most economical way possible. 5) Additional gift
campaigns are much more targeted by focusing on
TM responsive donors. 6) Requires approval and
sign-off from each client to participate. 7)
Future of the Direct Tag program? Include
demographic data as well. 8) A work in progress.
Constantly gathering more Direct Tag data with
every campaign conducted!
49
WORK IN PROGRESS? WHATS CHANGED?
Findings from working with Direct Tags over time
In some cases, too many records fell in the
lowest direct tag segments. Solution Developed
a second, more focused system, where leads are
more evenly dispersed.
Second incarnation of the Direct Tags
50
POAOCONO POAOCONU POAOCUNO POAOCUNU POAUCONO POAUC
ONU POAUCUNO POAUCUNU PUAOCONO PUAOCONU PUAOCUNO P
UAOCUNU PUAUCONO PUAUCONU PUAUCUNO PUAUCUNU NA
No one ever claimed data mining wouldnt be
complicated!

• - New 20 level code that groups records into
  smaller rankings, based on the four main goals of
  any campaign
• Pledge Rate Average Gift Credit Card -
  Answer Rate
• Offers another list management tool, allowing us
  to hone in on a specific area of need for a
  campaign.
• (For example, if Pledge Rate is the main priority
  we can overlay these tags with other segments to
  find the best PR producing pockets of leads.)

51
(No Transcript)
52
MORE
POAOCONO POAOCONU POAOCUNO POAOCUNU POAUCONO POAUC
ONU POAUCUNO POAUCUNU PUAOCONO PUAOCONU PUAOCUNO P
UAOCUNU PUAUCONO PUAUCONU PUAUCUNO PUAUCUNU NA

• We can identify pockets of donors who will not
  only perform above or below a certain goal, but
  also those donors who are most likely to pick up
  the phone!
• We are continuing to test both systems and apply
  overlay suppressions to a wide variety of
  segments including

1. Low Dollar Donors 2. Single Year Donors 3. Non
Appeal Responsive Donors 4. Deep Lapsed Donors 5.
Acquisition Campaigns
53
THE PROOF IS IN THE PUDDING
Client A Challenge Declining results on a
yearly summer appeal campaign. Since, the purpose
of the campaign is bottom line revenue, it was
imperative to try to maintain or exceed results
from year to year but that was not happening.
DPC Overview 2005 14.03 2006 17.13 2007
16.52 (and achieved a higher file
penetration!)

• Developed Direct Tags and applied them to 2006
  and 2007 appeal campaigns.
• Used the Direct Tags to overlay onto the 2
  primary segments appeal responsive (ARs) and
  non-appeal responsive (NARs).
• 3.10 and 2.49 increase in DPC from 2005!

54
THE PROOF IS IN THE PUDDING
Client A Challenge - continued
Increased performance and penetration on
Non-Appeal Responsive donors (NARs), an
increasingly challenging segment.
DPC 2005 7.91 2006 12.16 2007 9.42
PENETRATION 2005 19.1 2006 24.9 2007
33.2
PLEDGES 2005 576 2006 1281 2007 1296
PLEDGED 2005 36,372 2006 85,670 2007
84,648
- Utilized Direct Tag System
Based on these increases, we changed the
description of NARs from Non-Appeal Responsive,
to Yet-To-Be Appeal Responsive!
55
THE PROOF IS IN THE PUDDING
Client A Challenge - continued
T1GO 24.28 T2GO 22.93 T3GO 20.83
T4GO 15.70 T5GO 8.74 T5GU 8.20 T6GU
7.76 T7GU 5.32 T8 10.75
Here are the actual DPC results in each of the
Direct Tags for the NARs on the 2007 Appeal By
penetrating the higher performing Direct Tags
more deeply than T5GU, T6GU and T7GU, we
accomplished our goal of increasing results
compared to 2005.
56
SOME FINAL THOUGHTS
Strategic Data Mining and tagging systems are
becoming an increasingly essential component in
maintaining historical levels of
performance. Donors continue to receive
solicitations by mail and phone from many
organizations competing for their charitable
dollars. Many are becoming increasingly
cautious about giving because of the state of the
current economy and are cutting back to only a
few of their most cherished groups. While tagging
systems provide a list management tool, their
effectiveness is limited by the way in which they
are used The human element in developing list
management strategies remains the most important
ingredient!
57
Tiffany Neill tneill_at_lautmandc.com Amy Sukol
asukol_at_lautmandc.com Lisa Drane ldrane_at_dam.com
Slides Available at www.lautmandc.com Click on
Events and Recent Presentations