Hierarchical Classification & Routing(tm): A White Paper
Making the Customer Visible (tm)
Hierarchical Classification & Routing(tm): A White Paper
(c)Amacis 2001 Making The Customer Visible(tm) Page 2
Introduction
W
hat happened to that e-mail you sent to your bank regarding your account
balance? Have you ever wondered why the responses to your electronic
requests (such as e-mail, wireless messaging or web forms) are often both late
and irrelevant to the original correspondence?
The answer to those questions is remarkably simple. Most organizations sort
through incoming customer service e-mails manually. While the Internet has
promised customers a seamless means of interacting with companies,
delivering on that promise is challenging. When deluged with customer service
requests, how does an organization ensure that the right person services the
request and how does it integrate the requests with its backend fulfillment
systems?
This issue is particularly problematic for large, global institutions. Imagine that
XYZ Bank is multinational with regional offerings in Brazil, Canada, and the
United States. Not only are there multiple languages to consider (i.e.,
Portuguese, Spanish, French and English), but the regulatory environments for
each country create internal organizations with vastly different needs and
structures as well. While most companies recognize that showing the customer
a single, consistent interface is beneficial, accepting all e-mails though a single
source (e.g. customerservice@xyz.com) will be impossible to administer without
a vast investment in human capital.
Some companies have attempted to curtail this situation. They either have built
"home-grown" systems, or they have implemented an e-mail handling
technology. In either case, the system generally routes incoming e-mails to
customer service representatives who can answer the inquiries. This technique,
called "Skills-Based Routing," was adapted from telephone call center
applications. A set of rules is constructed by classifying existing electronic
requests (e-mail, web forms, mobile messaging, etc) into content
classifications. These classifications are then mapped to the skill sets of the
company's internal agents. Each incoming request is then routed to specific
agents based on this classification scheme.
Hierarchical Classification & Routing(tm): A White Paper
(c)Amacis 2001 Making The Customer Visible(tm) Page 3
Unfortunately, this solution does not work efficiently for the majority of large,
established companies. Most of the early e-mail handling solutions were built
to administer either a relatively small volume of electronic requests, or a small
number of classifications. The means of classifying the request is based on a
flat rule structure that results either from manually composing rules or
generating them using Bayesian Networking techniques. In either instance, this
structure gives each rule equal weighting and evaluates the incoming request
against all rules in aggregate (see figure 1). The final classification is based on
the combined results from tests against the entirety of these rules. While the
means of classifying the incoming request is manageable for smaller volumes
of requests and/or small numbers of classifications, there are several, severe
shortcomings for large, distributed organizations.
Administering a single set of rules across complex organizational structures
where differences might be necessitated by culture, regulation, market
segmentation, or brand specificity is problematic. No single individual is
likely to have the complete knowledge to establish and maintain the entire
rule structure.
Incoming
Request
Rule 1
Rule 2
Rule 3
Rule 4
Rule 5
Rule 6
Rule 7
Rule 8
Rule 9
Rule 10
Outgoing
Classification
Figure 1: Flat Rule
Hierarchical Classification & Routing(tm): A White Paper
(c)Amacis 2001 Making The Customer Visible(tm) Page 4
As business units alter their product offerings or internal structure, a severe
limitation is placed on the organization. Either a centralized administrator
needs to manage the necessary rule changes, thus adding time and reducing
the efficiency of communication, or business-unit level administrators must
be allowed access to the rule set. By allowing multiple administrators to
make alterations to the rules set, the administrators are able to change rules
pertaining not only to their business unit, but also to the other business
units as well. Thus, a potential security risk is added.
The flat rule structure is inefficient where language assignment (e.g., a rule
that says "this is a German language request") is an element of the final
classification. Since all rules are evaluated in aggregate, it is not possible
for the result of one rule to effect the processing of a second. The system
is, therefore, unable to apply a single language dictionary.
A language dictionary is a system device used only in advanced classification
engines that allows a program to handle "word stemming," a process
whereby the dictionary "knows" that variations of a single word (e.g.,
balance, balancing, balances...) have the same root word. Invocation of the
rule dictionary benefits the system by reducing the total number of
variations that must be considered. In addition, storage of the language
information outside of the program, results in a greatly reduced code base.
Since no dictionary may be applied with a flat rule structure, the number of
variations that the system must consider is increased, thus burdening the
system further. No other electronic response management software
currently available handles word stemming.
Hierarchical Classification & Routing(tm): A White Paper
(c)Amacis 2001 Making The Customer Visible(tm) Page 5
Hierarchical Classification & Routing
By contrast, Amacis Visibility's classification engine utilizes one of the most
advanced techniques possible to categorize incoming electronic requests.
Hierarchical Classification & Routing(tm) is an architecture that allows nodes,
which are comprised of sets of rules to be evaluated based on a series of preestablished
dependencies (see figure 2). Since the result of the first node test
determines the subsequent nodes to be evaluated, there are fewer variations to
be evaluated overall in order to determine an end node (i.e., final classification).
In addition language dictionaries may be applied at any juncture in the
"decision tree" and partitioning can be added to the overall "decision tree" to
allow multiple users access to a portion of the node structure.
Incoming
Request
Node 1
Figure 2: Hierarchical
Classification & Routing
Node 2 Node 3
Node 4 Classification 1 Node 5
Classification 4 Classification 3 Classification 6 Classification 5
Classification 2
Hierarchical Classification & Routing(tm): A White Paper
(c)Amacis 2001 Making The Customer Visible(tm) Page 6
In order to establish these dependent relationships, an overall tree is first
constructed using client input and Amacis' experience. Within each node, the
system generates a set of rules by analyzing existing categorized electronic
request data. To return to the XYZ Bank example, the first node might
determine the language that the electronic request is written in (e.g.,
Portuguese, Spanish, French or English), the second, the country of origin
(Canada or United States) and the third, the banking need to be addressed (e.g.,
bank balance inquiry, loan acceptance status).
Individual requests are then routed through the tree; relevant information is
pulled from existing systems, a response is formulated and sent to a Customer
Service Representative; this response is edited as necessary; and it is forwarded
to the customer. Because the tree's nodes determine the rules that are
considered, our example evaluates three rule sets (language, country, and
inquiry) rather than all possible variations. In addition, after evaluating the
language of the electronic request, the Hierarchical Classification & Routing(tm)
architecture is able to assign the language dictionary to the request. For
instance, if the electronic request is sent in English, the dictionary understands
that the word stems for "inquiry," "inquiries," and "inquiring" have the same root
word and may be evaluated equivalently. With support for thirty-five
languages, Amacis' architecture is the only one that supports word stemming.
In addition, the hierarchical structure of the tree enables, through the
implementation of partitioning, multiple administrators to alter portions of the
tree without being given access to change rule sets that don't pertain to them.
If the Brazilian office decides to offer a new product, a designated administrator
would be able to make the necessary changes to the tree in order to
accommodate customer service requests related to this offering. This
administrator would not have access to change the routing that occurs due to
language assignment, country determination, or business offerings in other
countries or business units. By allowing multiple node administrators, the
ability of the structure to adapt to the divergent demands placed upon it by
distributed organizations is greatly increased.
Hierarchical Classification & Routing(tm): A White Paper
(c)Amacis 2001 Making The Customer Visible(tm) Page 7
Conclusion
As the only software company that offers partitioning, Hierarchical
Classification & Routing(tm) and multiple language support as an integral part of
its product offering, the underlying structure of Amacis Visibility's architecture
enables even the largest, most complex and globally distributed organizations:
??To efficiently categorize incoming electronic correspondence in multiple
language settings - Amacis Visibility manages electronic, customer
correspondence in up to 35 languages concurrently.
??To efficiently distribute the administrative burden of rule alteration
throughout the organization - Changes that naturally occur within an
organization, and Amacis Visibility allows a company to have local or
functional experts make changes to the relevant rule sets without
effecting rule sets that don't pertain to them.
For more information about the benefits that your organization can realize from
Amacis Visibility, please e-mail info@amacis.com or contact us at one of the
numbers below.
Amacis North America
Amacis Inc
101 Federal Street
Suite 1600
Boston
MA 02110
T: 1 877 9 AMACIS (Toll Free)
F: 1 617 204 5781
E: info@amacis.com
W: www.amacis.com
Amacis Europe
Amacis Ltd
Station Mill
Station Road
Alresford
Hampshire SO24 9JQ
T: +44 (0)19 6273 8496
F: +44 (0)19 6273 8798
E: info@amacis.com
W: www.amacis.com
Amacis Asia Pacific
Amacis Ltd
2705 Convention Plaza
1 Harbour Road
Wanchai
Hong Kong
T: 852 2829 7404
F: 852 2829 7377
M: 852 9159 7110
E: mike_francis@amacis.com
Amacis R&D
Amacis Ltd
Falcon Road
Belfast
Northern Ireland
BT12 6SJ
T: +44 (0) 28 9087 2000
F: +44 (0) 28 9087 2001
E: info@amacis.com
W: www.amacis.com
(c) Copyright 2001, Amacis Inc.