brings unique challenges with respect to available data. Firms
typically collect Customer Relationship Management (CRM)
type data on millions of customers and their transaction records.
Marketing activity is captured in two ways: traditional aggregatelevel
data on “push” mass media (e.g., TV, radio, print, banner
ads) and customer-level data on what was “pulled” by the
customer to enable purchase (e.g., permission-based communication,
coupons claimed, newsletter emails). This combination
should allow companies to profile customers based on their
responsiveness to push marketing as well as their pull behavior.
Potentially, short- and long-term effects of marketing actions in
isolation and combination (multiple touch points) can differ
substantially across customer segments. If so, a segmentation
analysis can generate actionable insights for marketing budget
allocation. Unfortunately, realizing this potential is complicated
by the sheer size of the customer base and the lack of a modeling
framework combining response-based segmentation with longterm
effect estimation. This paper introduces a modeling
approach that enables managers to quantify marketing effectiveness
based on all available data. Our approach combines existing
“best practice” methods of segmentation and long-run effects
modeling to investigate marketing mix effectiveness. Ultimately,
we aim to generate new insights into which marketing actions
yield long-term benefits for the most valuable customer segments
in the digital media space.
We use our framework to study marketing mix effectiveness
in the digital music space. Our data come from the leading
digital media provider in a large European country.2 Our
contribution is threefold. First, we show how online consumer
segments, based on their short-term marketing response, have
substantially different sizes and profiles. Second, we quantify
for each segment the long-term effects of coupons and
advertising media as well as their interactions. In contrast to
empirical generalizations from consumer packaged goods,
heavy users of digital music are less price sensitive than light
users and more responsive to advertising. Third, we show how
marketing actions with insignificant direct sales impact (print)
may still be worthwhile due to their synergy with effective
marketing actions (TV and Internet marketing). The remainder
of this paper is organized as follows. First, we discuss how
previous literature on CRM and long-term marketing effectiveness
may apply differently to digital media products. Next, we
present our data and propose a modeling approach that allows
combining customer-level purchase data using the whole
customer base and customer-level and aggregate-level marketing
mix data. The first modeling step involves segmenting
customers based on observed purchase behavior while accounting
for unobserved heterogeneity using a latent-class
approach. The second step involves persistence modeling to
investigate the short- and long-run effects of marketing in each
segment. We report our results and show that segmenting
instead by an ad hoc approach (such as median or quartile
splits) does not allow uncovering the marketing response of the
most valuable customers. Finally, we discuss how our findings
brings unique challenges with respect to available data. Firmstypically collect Customer Relationship Management (CRM)type data on millions of customers and their transaction records.Marketing activity is captured in two ways: traditional aggregateleveldata on “push” mass media (e.g., TV, radio, print, bannerads) and customer-level data on what was “pulled” by thecustomer to enable purchase (e.g., permission-based communication,coupons claimed, newsletter emails). This combinationshould allow companies to profile customers based on theirresponsiveness to push marketing as well as their pull behavior.Potentially, short- and long-term effects of marketing actions inisolation and combination (multiple touch points) can differsubstantially across customer segments. If so, a segmentationanalysis can generate actionable insights for marketing budgetallocation. Unfortunately, realizing this potential is complicatedby the sheer size of the customer base and the lack of a modelingframework combining response-based segmentation with longtermeffect estimation. This paper introduces a modelingapproach that enables managers to quantify marketing effectivenessbased on all available data. Our approach combines existing“best practice” methods of segmentation and long-run effectsmodeling to investigate marketing mix effectiveness. Ultimately,we aim to generate new insights into which marketing actionsyield long-term benefits for the most valuable customer segmentsin the digital media space.We use our framework to study marketing mix effectivenessin the digital music space. Our data come from the leadingdigital media provider in a large European country.2 Ourcontribution is threefold. First, we show how online consumersegments, based on their short-term marketing response, havesubstantially different sizes and profiles. Second, we quantifyfor each segment the long-term effects of coupons andadvertising media as well as their interactions. In contrast toempirical generalizations from consumer packaged goods,heavy users of digital music are less price sensitive than lightusers and more responsive to advertising. Third, we show howmarketing actions with insignificant direct sales impact (print)may still be worthwhile due to their synergy with effectivemarketing actions (TV and Internet marketing). The remainderof this paper is organized as follows. First, we discuss howprevious literature on CRM and long-term marketing effectivenessmay apply differently to digital media products. Next, wepresent our data and propose a modeling approach that allowscombining customer-level purchase data using the wholecustomer base and customer-level and aggregate-level marketingmix data. The first modeling step involves segmentingcustomers based on observed purchase behavior while accountingfor unobserved heterogeneity using a latent-classapproach. The second step involves persistence modeling toinvestigate the short- and long-run effects of marketing in eachsegment. We report our results and show that segmentinginstead by an ad hoc approach (such as median or quartilesplits) does not allow uncovering the marketing response of themost valuable customers. Finally, we discuss how our findings
การแปล กรุณารอสักครู่..
