PERSONALIZED INCENTIVE PLANS THROUGH EMPLOYEE
PROFILING
Silverio Petruzzellis
C
´
ezanne Software S.p.A. - Via Amendola, 172/C - I70126, Bari, Italy
Oriana Licchelli, Ignazio Palmisano, Giovanni Semeraro
Dipartimento di Informatica - Universit
`
a di Bari - Via E. Orabona, 4 - I70126, Bari, Italy
Valeria Bavaro, Cosimo Palmisano
DIMeG, Politecnico di Bari - Viale Japigia 182, I70100 Bari, Italy
Keywords:
Decision Support Systems, Knowledge Management, Machine Learning, HRM Decentralization.
Abstract:
Total reward management (TRM) is a holistic practice that interprets the growing need in organizations for
involvement and motivation of the workers. It is oriented towards pushing the use of Information Technol-
ogy in supporting the improvement of both organization and people performances, by understanding em-
ployee needs and by designing customized incentives and rewards. Customization is very common in the area
of e-commerce, where application of profiling and recommendation techniques makes it possible to deliver
personalized recommendations for users that explicitly accept the site to store personal information such as
preferences or demographic data. Our work is focused on the application of User Profiling techniques in
the Total Reward Management context. In the Team Advisor project we experimented the analogies Cus-
tomer/Employee, Product, Portfolio/Reward Library and Shop/Employer, in order to provide personalized
reward recommendations to line managers. We found that the adoption of a collaborative software platform
delivering a preliminary reward plan to the managers fosters collaboration and actively supports the delegation
of decision-making.
1 INTRODUCTION
Competitiveness is a key issue for knowledge-
intensive organisations in today’s turbulent global
marketplace. The most industrialised countries can
no longer expect to compete only on efficiency. The
labour cost gap is for the time being unbridgeable, at
least until the market’s natural economic forces in-
crease the labour cost in developing countries such as
India and China. It is a well-documented fact that to
be competitive now depends more than ever on a com-
pany’s capacity for innovation and on the level of its
customer services, both of which attributes depend on
the level of involvement and motivation of the work-
ers. Total reward management (TRM) is a holistic
practice that interprets these growing needs in organi-
sations and offers a new action mode oriented towards
pushing the use of Information Technology in sup-
porting the improvement of people performances, by
understanding employee needs and by designing cus-
tomized incentives and rewards. In a recent study car-
ried out by Watson Wyatt (Watson Wyatt Worldwide,
2000), interesting statistics emerged about the factors
that most commonly influence people’s performance
within a company. According to the study, there are
three factors that stand out from the rest, namely,
trust in senior leadership, chance to use skills on the
job, and competitiveness of rewards. The Human Re-
source (HR) departments that historically looked af-
ter all the personnel development issues are now fac-
ing a new challenge: to reinterpret their role and be-
come a more strategic component for the company, by
contributing in the definition of incentive plans and
professional development. To achieve this goal, they
strongly need the participation of those who better
know the single individuals, working side by side and
coordinating their activities: the line managers. Geo-
graphical dislocation, complexity of roles and the es-
calating average size of the companies, force the HR
departments to involve middle management in provid-
ing the right answers to the pressing demands of the
employees. The line managers on the other hand do
not possess the right background to cope with internal
conflicts, to make objective evaluations and to act as
compensation experts towards their staff. They need
the right support and a good mix of up-to-date infor-
107
Petruzzellis S., Licchelli O., Palmisano I., Semeraro G., Bavaro V. and Palmisano C. (2006).
PERSONALIZED INCENTIVE PLANS THROUGH EMPLOYEE PROFILING.
In Proceedings of the Eighth International Conference on Enterprise Information Systems - AIDSS, pages 107-114
DOI: 10.5220/0002493401070114
Copyright
c
SciTePress
mation in order to apply their judgement in a timely
and effective manner.
2 THE ROLE OF INFORMATION
TECHNOLOGY IN TRM
IT support in this area could be very valuable if
designed in a way that foster collaboration and fa-
cilitate the devolution of some HRM responsibili-
ties to the line managers. Many authoritative man-
agement theories, like Drucker’s (Drucker, 1955),
Mintzberg’s (Mintzberg, 1973), include responsibil-
ities such as ”hiring”, ”motivating” and ”people de-
velopment” as well as other more operational activ-
ities. As Mintzberg sums up; ”He [the manager] de-
fines the milieu in which they [his subordinates] work,
motivates them, probes into their activity to keep them
alert and takes responsibility for hiring, training and
promoting them” ( (Mintzberg, 1973), p96).
The Team Advisor project addressed those issues
by designing and implementing a platform for collab-
oration in the complex area of personnel development
and reward planning. Its goal was to enable the line
managers to be in charge of their own development
plans by providing them with a personalized and con-
textualised set of information about their teams, to-
gether with some policy guidelines to follow in the
implementation phase of those plans.
The use of profiling methodologies appeared very
promising in this area. Personalization is very com-
mon in the area of ecommerce, where a user explicitly
accepts the site to store information on her or himself
such as her or his preferences in order to be provided
of personalized recommendations. The more such
a system knows about users the better it can serve
them effectively. There are different styles, and even
philosophies, regarding how to teach computers about
user habits, interests, patterns and preferences. User
modelling simply means ascertaining a few bits of in-
formation about each user, processing that informa-
tion quickly, and providing the results to applications,
all without intruding upon the user’s consciousness.
The final result is the construction of a user model or
a user profile (Kobsa, 1993). The application of pro-
filing methodologies was experimented then to create
personalized incentive plans that could reflect the his-
tory and the habits of the single individuals as well as
balancing the fulfilment of her or his needs with the
company’s global perspective.
3 MARKET ANALYSIS
Our analysis of the actual context of Human Resource
Management (HRM) applications has shown that
most of organisations that already use software so-
lutions for employee administration are now search-
ing for applications which can strengthen their HRM
systems with decision support functionalities that can
simplify their management processes. We did a mar-
ket research that involved the following software so-
lutions: Authoria Advisor
1
, Disc Profiler
2
, Incen-
tive Suite
3
, Etweb Enterprise
4
, Onesis IAS 19
5
,H1
HRMS
6
, IDP-IDipendenti
7
, Minosse
8
.
The market segmentation was analyzed focusing
on those products that appeared to be more compet-
itive in terms of profiling technologies and rewards
management system, which are essential requisites to
create customized reward recommendations (Spitzer,
1996; Busacca, 1998). The obtained results are shown
in a segmentation matrix (Figure 1) based on the two
variables ”variety of rewards offer” (rewards library
on axes Y) and ”focus on profiling techniques” (pro-
filing technology on axes X).
None of the assessed information systems presented
a good mix of both required features. Thus they
are mainly positioned in the lower part of the ma-
trix, manifesting an insufficient level of detail in the
rewards range definition. The analyzed softwares
have in fact been developed essentially for supporting
personnel administration, with low attention to more
complex processes, such as customized reward plan-
ning and assignment. On the other hand, the upper
right part of the matrix represents the ideal place-
ment of the envisioned software application, since
it requires not only a high-level quality in profiling
technologies and in reward library definition, but also
a strong interconnection between the two aspects.
More accurately, this application has been designed
for merging contents from profiling system and el-
ements of reward library and create suggestions for
customized incentives: the purpose is to create a kind
1
Authoria Inc., http://www.authoria.com/
AdvisorSeries.204.aspx
2
Axiom Software Ltd., http://www.
axiomsoftware.com/discprofile.asp
3
Nexcentec Software Inc., http://www.
nexcentec.com
4
ExecuTRACK Software AG, http://www.
executrack.com/en/solutions.html
5
Onesis S.p.A, http://www.onesis.it
6
EBC Consulting Srl, http://www.
ebcconsulting.com/cgi-bin/private/hrms\
area.cgi
7
TPC&Join s.r.l., http://www.idipendenti.
it/demo/demo.asp
8
Mizar Informatica s.r.l., http://www.mizar.it/
prodotti/index.php\#minosse
ICEIS 2006 - ARTIFICIAL INTELLIGENCE AND DECISION SUPPORT SYSTEMS
108
of ”virtual manager” supporting enterprise compensa-
tion and reward policies.
Figure 1: Segmentation Matrix.
4 THE TEAM ADVISOR SYSTEM
Within the Team Advisor project, our work was
focused on the application of the User Profiling
techniques in the Total Reward Management con-
text, where we experimented the following analo-
gies with respect to the on-line market: Cus-
tomer/Employee, Product Portfolio/Reward Library
and Shop/Employer.
In this context, a user profile is meant as the set
of information collected about a user, in order to take
into account her or his needs, wishes, and interests
in granting those benefits that are part of the organ-
isation’s incentive programmes, thus a set of recom-
mendations reflecting those preferences. Rather than
addressing the recommendations to the customer, as
it generally happens in the standard e-commerce ap-
plications, the Team Advisor project experimented a
new form of advice addressed to the line manager,
suggesting the best reward policy for a specific staff
member, depending on a profile built upon both de-
mographic data and behavioural preferences collected
through a commercial Human Resources Manage-
ment portal, without asking to the employee to com-
pile a periodical questionnaire to provide her or his
preferences for potential rewards. As well as in the
e-commerce area, the privacy problem is also to be
considered in the TRM context: many users are dis-
trustful to give private data because they wish to have
their privacy guaranteed and they may also be afraid
of being controlled. One possibility is to handle per-
sonal data for the customer so that she can control/edit
her own user model. In addition, the company must
guarantee not to give personal data to a third party and
in addition should provide suitable technical means
for users to rectify user model items (Kobsa, 2001).
Actually, the core issues are trust and transparency,
because the user wants the companies to respect their
privacy by clearly declaring when they are collecting
the personal data, offering opt-in to data collection
programs rather than opt-out only choices. In fact,
the EU Privacy Directive (European Parliament and
Council of the European Union, 1995) gives users the
right to have access to ”the knowledge of the logic in-
volved in any automatic processing of data concern-
ing the user”. Taking into the right consideration pri-
vacy issues, the suggested metaphor allowed for a
new HRM paradigm in which the line manager acts
as a partner of the HR department. Provided with a
personalized recommendation system, he could eval-
uate individual needs and expectations and try to find
the right balance between them and the organisation’s
rewarding strategy received by the HR department.
5 EMPLOYEE PROFILING
In the Team Advisor project, the Employee Profil-
ing system creates detailed profiles of the co-workers.
The Employee Profiling system is built upon a Per-
sonalization Engine (PE) (Semeraro et al., 2003; Lic-
chelli, 2005) that takes advantage from information
stored in a commercial Human Capital Management
application. The PE component is a Java applica-
tion deployed in Apache Tomcat. The Human Capital
Management application, provided by a global soft-
ware player in the area of Human Resources Man-
agement Systems, is a web application written in Vi-
sual Basic for the Microsoft COM+ architecture. The
integration of the two systems was carried out by a
C# module run in a Microsoft .NET environment,
and was based on RDF/XML-driven interoperability.
The whole integration architecture schema is shown
in Figure 2.
Before starting with the profile creation process, a
preliminary work is needed to establish a formal de-
scription of the features necessary to accomplish the
given task. Starting from these features it is possi-
ble to define the representation language of the entire
learning system. In parallel it is necessary to define a
number of classes that characterize the corresponding
kind of reward to be associated to the various con-
figurations of features. Then, it is possible to create
the employee profiles in two steps: during the first
one a number of sample classification are provided to
the Personalization Engine in order to create a set of
PERSONALIZED INCENTIVE PLANS THROUGH EMPLOYEE PROFILING
109
Figure 2: Employee Profiling System.
classification rules. In the second phase, those rules
are applied, employee by employee, to the whole set
of identifying features. The application then gener-
ates a set of recommendations composed of a list of
rewards the system deemed appropriate for each indi-
vidual, together with a ranking useful to prioritize the
decisions.
5.1 Personalization Engine
The PE component is the part of the system that uses
supervised learning techniques to automatically dis-
cover users’ preferences. It interacts with the WEKA
library (Witten and Frank, 2000) in order to induce
rule sets for classification. Both the input and the out-
put of the PE component is composed of RDF models.
These models contain:
the input to the learning process (the set of posi-
tive and negative examples, as well as the definition
of the learning problem, i.e. the features together
with their possible values, and the name and type
of classification to be learned), the main class for
this being pe:InstanceBag;
the results of the learning process (a set of classifi-
cation rules), represented by the pe:Classifier class;
the input to the classification phase, in which the
learned rules are used to categorize previously un-
seen instances, which consists of a pe:Classifier in-
stance and one or more pe:InstanceBag instance,
together with the related output, which is instance
of pe:ClassificationResult.
The main functionalities are then:
Phase name Input Output
Rule building pe:InstanceBag pe:Classifier
instance instance
Classification pe:InstanceBag pe:ClassificationResult
(unlabelled)
pe:Classifier
5.2 User Profile Generation Process
In the context of TRM, the problem of learning em-
ployee’s preferences can be cast to the problem of in-
ducing general concepts from examples labelled as
members (or non-members) of the concepts them-
selves. Given a finite set of categories of rewards
C = {c
1
,c
2
, ···c
n
}, the task consists in learning the
target concept T
i
”employees interested in the cate-
gory c
i
”. In the training phase, each individual set of
features represents a positive example regarding the
categories the employee is interested in and a negative
example regarding those she or he dislikes. To this
goal we chose an operational description of the tar-
get concept T
i
, using a collection of rules that match
ICEIS 2006 - ARTIFICIAL INTELLIGENCE AND DECISION SUPPORT SYSTEMS
110
Figure 3: Personalization Engine learning process.
against the features describing a user in order to de-
cide if she or he is a member of T
i
. The subset of
the instances chosen to train the learning system has
then to be labelled by a domain expert, that classi-
fies each instance as member or non-member of each
category. Stated that, in the context of TRM, the do-
main expert belongs to the HR department, in order
to facilitate the definition of the training sets we intro-
duced a pre-classification process interesting popula-
tion samples instead of single individuals. The expert
was then asked to classify groups of employees shar-
ing the same subset of characteristics, and to decide
weather they belong to a certain category or not.
There are 4 rules extracted for class Company Car:
if Seniority <= 7.0
and Job Class = Employee-CONS then
Class: yes
else if Seniority > 7.0 then
Class: Estate
else if Job Class = White Collar-SALES then
Class: no
Otherwise Class: Saloon
Figure 4: An example of classification rules for the reward
category “Company Car”.
This approach allowed the generation of training
sets without the direct knowledge of the candidates,
thus taking advantage of a natural clustering habit of
the expert user. The training instances were processed
by the Personalization Engine, which induced a clas-
sification rule set for each reward category by tak-
ing into account the whole set of features at a higher
Figure 5: Employee Profile.
level of detail than the expert user can afford (Fig-
ure 3). Each rule is composed of an antecedent, or
precondition, and a consequent, or conclusion. The
antecedent is a series of tests, like the tests at nodes
in decision trees, while the consequent gives the cat-
egory that applies to instances covered by that rule
(Figure 4). The learning algorithm adopted in the pro-
file generation process is based on PART (Frank and
Witten, 1998), a rule-based learner that produces rules
from pruned partial decision trees, built using C4.5’s
heuristics (Quinlan, 1993). The main advantage of
this method is its simplicity: it produces good rule
sets without any need for global optimizations.
The actual employee profile generation process is
performed by applying this set of rules to the em-
ployee data. For each reward category, the system is
PERSONALIZED INCENTIVE PLANS THROUGH EMPLOYEE PROFILING
111
Figure 6: A list of the most representative data.
able to predict whether the employee is interested in
it or not, specifying the hypothetical satisfaction level
for each. An employee profile (Figure 5), as intended
within the Team Advisor project, is then composed by
two main frames: the frame of employee data, and the
frame of the rewards ”learned” by the system.
6 MANAGER ADVISOR
TeamAdvisor has been designed to provide person-
alized reward recommendations (Kerr, 1995) through
the analysis of user profiles stored in a support data-
base. This allows for a rationalization of the process
of incentives planning, with consistent time savings
in the implementation of decisions. However, we had
to cope with the problem of identifying who is the
best candidate to examine the recommendations pro-
vided and to exploit them. This requires on the one
hand some degree of familiarity with compensation
and rewarding practices, and on the other hand a good
acquaintance with the employee they are addressed
to (Wanless, 2003). An innovative solution could be
to integrate the traditional competencies of the HR de-
partment by delegating to the line managers some of
their activities. Legge (Legge, 1995) underlines that
HRM is ”vested in line management as business man-
agers responsible for coordinating and directing all
resources in the business unit in pursuit of bottom-
line profits”. Therefore evaluation process, employee
development, performance management and rewards
assignation can be considered the best areas for an
effective intervention of the line on HRM. In order
to simplify organizational structure and to human-
ize personnel administration procedures, they seem to
be the most appropriate figures to accomplish these
changes (Renwick, 2003). Moreover, they are sup-
posed to know personally the staff they manage, also
thanks to informal conversations in the workspace, so
that they can add a personal value in examining rec-
ommendations and in choosing which reward to as-
sign. Two factors play against a partial devolution of
HR strategic management to the line manager (Whit-
taker and Marchington, 2003):
They generally have a heavy load of work respon-
sibilities, which should not be increased with tasks
they have no great competence about.
Because of their dimensions, many large-sized en-
terprises make it very difficult to build human re-
lations that go beyond formal employer-employee
relationships.
The answer to these objections can be found in the
philosophy of the proposed IT solution: the adoption
of a collaborative software platform delivering a pre-
liminary reward plan to the managers, makes the de-
volution process a real thing, and actively supports
decision-making. Line managers can then add their
personal judgement coming from personal acquain-
tance storing from the information provided by sys-
tem as a result of the analysis of objective data stored
in the company database. To substantiate these reflec-
tions with a use case, we created a demo data set rep-
resenting a US firm working in the banking/financial
industry, and did a full run of the Team Advisor sys-
ICEIS 2006 - ARTIFICIAL INTELLIGENCE AND DECISION SUPPORT SYSTEMS
112
Figure 7: Selected rewards into the organizational incentive
plan.
tem within the company reward assignment process.
An HR domain expert sets up the employee pro-
file generation process in two phases. First, she se-
lects a list of the most representative data about iden-
tity, professional path and socio-demographical back-
ground of the employees. Anagraphical and socio-
demographical data closely pertain to human dimen-
sion of the employee and are the same for every com-
pany. They can be collected and updated through peri-
odical on line surveys. On the contrary, data about job
conditions, although generally available in all com-
panies’ databases, are context-sensitive, and may be
more or less relevant according to the external envi-
ronment and to the company policy. In figure 6 it is
shown, for a demonstrative purpose, a partial list of
the attributes enclosed in the employee-profile. This
can be properly customised according to the needs of
each company.
Second, she selects the rewards listed into the orga-
nizational incentive plan, shown in Figure 7, to build
the Team Advisor Reward Library. For each item a
domain is also specified, made up either of yes/no la-
bels or of multivalued labels (i.e. saloon, estate, city
car, as possible values for the item ”company car”).
The reward library provided in the example is created
following a recent study by OD&M (OD&M Consult-
ing, 2005), and is given as a basic customizable set.
A user profile generation process is then started, and
it eventually produces the employee profiles as de-
scribed above (see Section 5.2).
Take now a manager of our sample enterprise, Fred
Sidel from the Atlanta Branch, receiving recommen-
dations about potential rewards to be assigned to his
team, through a dedicated section of his digital dash-
board. The system provides him with a detailed view
of the recommendations for each of his thirteen co-
workers, as shown in Figure 8.
For each recommendation one single label is speci-
fied, that is the domain value for which the recom-
mender has computed the highest popularity rating;
the goal of recommendations is to inform the line
manager about the preferred rewards and their adher-
ence with the related employee profile. It’s up to the
manager then to decide whether to use such informa-
tion, including the reward recommendation into the
incentive plan for each employee, or ignoring it and
adopting an autonomous evaluation method.
7 CONCLUSION AND FUTURE
WORKS
The work carried out in the Team Advisor project sug-
gested that the application of profiling and recommen-
dation techniques in the context of the incentive plan-
ning processes it is not only possible but also effective
in providing support to the decision-making. We ob-
served how the recommendations could act as a valu-
able starting point for the managers who need support
in an activity they are not used to perform, by provid-
ing a means for unleashing unexpressed and undis-
covered connections among the various elements that
govern the reward assignation process. In addition,
the recommendation system allows for the distribu-
tion of the knowledge of the HR domain expert by
providing timely and accurate information about the
preferences of the employees, not forgetting the over-
all company perspective. In this context, the judgment
of the managers can be applied to refine the sugges-
tions and to convert them into an effective develop-
ment plan.
Starting from the results achieved in the prototyp-
ical phase, the Team Advisor working group is com-
mitted to perform both a technical and an organiza-
tional evaluation of a commercial version of the soft-
ware developed. For what concerns the technical val-
idation we are interested in checking how the system
is able to replicate the knowledge of the domain ex-
pert and thus to provide coherent recommendations.
On the organizational level it is instead important to
verify how the provided recommendations can con-
tribute to streamline the compensation and rewarding
process of the organisations. To this goal, we’re going
to submit an ad-hoc survey to real users of the system,
chosen among the managers implementing personal-
ized incentive plans. This survey will aim at evaluat-
ing the overall satisfaction as well as the acceptance
level of the recommendations as for their usefulness,
clarity and completeness.
For what concerns possible future software devel-
opments, the group is aiming at embedding new func-
tionalities capable of collecting live user feedbacks on
the recommendations, and use them in order to re-
PERSONALIZED INCENTIVE PLANS THROUGH EMPLOYEE PROFILING
113
Figure 8: Recommendations for an employee.
fine the process of profile generation. This will enable
the system to become a dynamic engine of employee
profiling, thus supporting the organisations in refining
and sharing the knowledge about its human capital.
REFERENCES
Busacca, A. G. (1998). Il customer profiling. In Costruire
Fedelt
`
a. Il Sole 24ore.
Drucker, P. F. (1955). ”management science” and the man-
ager. Management Science, 1(2):115–126.
European Parliament and Council of the European Union
(1995). Directive 95/46/ec of 24 October 1995 on the
protection of individuals with regard to the process-
ing of personal data and on the free movement of such
data. In Official Journal of the European Communi-
ties, page 31.
Frank, E. and Witten, I. (1998). Generating accurate rule
sets without global optimization. In Proceedings of
the 15
th
International Conference on Machine Learn-
ing, pages 144–151. Morgan Kaufmann.
Kerr, S. (1995). An academy classic: On the folly of re-
warding A, while hoping for B. Academy of Manage-
ment Executive, 9(1):7–14.
Kobsa, A. (2001). Tailoring privacy to users’ needs. In Pro-
ceedings of the 8
th
International Conference on User
Modelling. LNAI 2109, pages 303–313. Springer.
Legge, K. (1995). Human Resource Management:
Rhetorics and Realities. Macmillan, Hampshire.
Licchelli, O. (2005). Personalization in Digital Libraries
for Education. Computer Science in the Graduate Di-
vision of the University of Bari, Italy.
Mintzberg, H. (1973). The Nature of Managerial Work.
Prentice-Hall, London.
OD&M Consulting (2005). Rapporto benefits 2005. in col-
laborazione con Lavoro & Carriere - ed. Il Sole 24
Ore.
Quinlan, J. R. (1993). C4.5: Programs for Machine Learn-
ing. Morgan Kaufmann.
Renwick, D. (2003). Line manager involvement in HRM:
an inside view. Employee Relations, 25(3):262–280.
Semeraro, G., Abbattista, F., Degemmis, M., Licchelli, O.,
Lops, P., and Zambetta, F. (2003). Agents, personal-
isation and intelligent applications. In Corchuelo, R.,
Cort
´
es, A. R., and Wrembel, R., editors, Technologies
Supporting Business Solutions, Part IV: Data Analysis
and Knowledge Discovery, Chapter 7, pages 141–160.
Nova Sciences Publishers.
Spitzer, D. R. (1996). Power rewards: Rewards that really
motivate. Management Review, 85(5):45–50.
Wanless, C. (2003). Devolution of HR activity to
the line: does the reality lag behind the rethoric?
http://business.king.ac.uk/research/hrm/wanless.pdf.
Watson Wyatt Worldwide (2000). WorkUSA 2000
- Employee Commitment and the Bottom Line.
http://www.watsonwyatt.com/research/resrender.asp?
id=W-304&page=1.
Whittaker, S. and Marchington, M. (2003). Devolving
HR responsibilities to the line. Threat, Opportunity or
Partnership? Employee Relations, 25(3):245–261.
Witten, I. H. and Frank, E. (2000). Data Mining: Practi-
cal Machine Learning Tools and Techniques with Java
Implementations. Morgan Kaufmann Publishers, San
Francisco.
ICEIS 2006 - ARTIFICIAL INTELLIGENCE AND DECISION SUPPORT SYSTEMS
114