Classification of inbound contacts according to “contact purpose”

Abstract

To improve the knowledge of its customers, a Company can use data about contacts.
If operational data on different channels are not suitable for Customer Relationship Management (CRM), it becomes necessary to define new dimension analysis.
In this topic we describe a methodology to classify inbound contacts according to “contact purpose”.
We propose a definition of contact purpose suitable for contacts from all channels, so that analysis can be done through channels.
We then describe a procedure to apply this definition to inbound operational data contacts, to obtain automatically classified contact data using a Keyword-Classification algorithm.
The same procedure can be applied to any other classification criterion, and to more than one criterion at once.
At last we suggest how this methodology can be nested in a process flow.

Introduction

To improve the knowledge of its customers, a Company can focus on relationship with customers.

To be suitable for CRM modelling and decisions, the general concept of “relationship” must be instanced to become concrete (definible), measurable and updatable.

A Company usually has available data that can be used to instance relationship: data about contacts between customer and company (inbound and outbound), over different channels.

In fact contacts are:

  • Concrete (known definitions),
  • Measurable: defined quantities measurable over time,
  • (Automatically) updated.

In this topic we focus on inbound contacts and describe a methodology to classify contacts according to “contact purpose”.

Inbound contacts data can be collected  from different channels (i.e. web and mobile site, call center). If data are collected from operational systems (ex. Web site logs), they are probably not suitable for CRM analysis: they are punctual data, specific for channel, with limited information usable for analysis.

Significant analysis could be done if inbound contacts were classified by significant criteria defining new analysis dimensions, that is new classification criteria should be independent from channel.

To define a significant classification criterion for contacts, we think in term of “contact purpose”.

We propose a definition of “contact purpose” suitable for contacts from all channels, so that analysis can be done on contact purpose through channels.

We then describe a procedure to apply this definition to inbound operational data contacts, to obtain automatically classified contact data using a Keyword-Classification algorithm.

The same procedure can be applied to any other classification criterion, and to more than one criterion at once.

Definition of Contact purpose

Classification is defined “a priori”.

OpenP Open or Activate Product
NewRes Research on new products and services
OwnRes Research on own products and services
CloseP Close product
Tran Place transactions
Claim Make a complaint
Info Quantitative Information Request
Support Support Request
Other Other

Procedure for classification

A. Initial manual classification (for initial “learning” of algorithm)

Above definition is fitted on operational contacts, starting from a chosen channel, for example web pages:

  • Analysis of site map and first manual classification of pages addresses: analysis of pages distribution on purpose and analysis of page accesses (a sample) distribution on purpose
  • Review of manual classification

Then the definition is generalized to fit also other channels, for example mobile pages addresses and call center call reasons, until it can be considered optimal.

Next steps B. and C. are strictly related, and need reiteration.

B. Definition of keyword list

From initial manual classification is then derived a keyword list to be used by classification algorithm.

The keywords must be as few and as shorter as possible, to preserve broadness of algorithm. This is very important.

C. Development of a Keyword-Classification algorithm

An algorithm implemented to automatically classify contacts. The algorithm:

  • Is based on keyword lists
  • Associates to contact purpose value defined by keywords, according to: a Positional criterion (for example in web pages right positions weigh more than left positions), a Priority criterion (to manage nested keywords and other critical cases)

Keyword list has to be periodically revised, to improve learning of algorithm.

After some learning steps, algorithm becomes more robust, and it is proven to be quite general.

Generalization

Inbound contacts can be classified according to more than one criterion. Every classification criterion becomes a dimension for analysis on this data. The described algorithm is able to manage one or more criteria, simply treating them as independent dimensions, with independent keyword lists.

Nesting in process flow

Classification process can be nested in process flow in different ways, for example:

  • completely automated: leave ambiguous contacts in ‘other’ class, and periodically review ‘other’ list
  • partially left to human evaluation: automatically classify known contacts, and use the algorithm  to ‘propose’ classification for new contacts, that will be seen and approved or modified by analysts.