Month: March 2012

Decisioning and Recommendation Systems

Posted by

0

Often Interaction Management systems are used to implement a customer Best-Next-Action recommendation capability. This is a typical usage in large financial services, insurance and telecommunications providers.

[The common usage is Next-Best-Action rather than Best-Next-Action and I have never really understood why this is. It may be because people find the acronym easier to pronounce, or that they like the acronym NBA better than BNA in that it is shared with the National Basketball Association or the Net Book Agreement. I have always preferred Best-Next-Action as Next-Best sounds like second-best.]

In discussions about Best-Next-Action systems there is often a conflict between the worlds of risk and marketing. Broadly speaking decisioning is typically found in the world of credit risk, fraud and compliance. This is typically a request to ensure that some business process can proceed based on all of the known context. In contrast a recommendation system is typically a marketing led operation where there is no single answer required or designated authority required but instead a set of best-next-action recommendations. Particularly in online  environments many recommendations are consumed by a web site in a single session or even page.

Also these two domains have different ways of configuring their content which is naturally led by the “process flow” single decision thinking for decisioning and the many possible outcomes world of marketing. It is common to find graphical process modelling paradigms in risk based decisioning and rule-based ranking paradigms in marketing recommendation systems.

These ar predominantly focussed on marketing recommendation systems, however with our broad definition of customer-centric marketing it is possible to include risk decisioning in marketing recommendation engines, but not the other way around. The main focus of how to operate recommendation style systems changes to how do I map my customer strategy into such a system. How do I know what to do for each type of customer in each and every circumstance and how do I specify what is the most important of all of the possible actions that I could take. Development and implementation of the Customer Strategy and Prioritisation are key topics in Interaction Management.

The (Customer) Context

Posted by

0

The Context, often called the Customer Context is the name for the key device that contains all of the known, imputed, calculated, predicted and derived information about a customer at the current point of an interaction.

The context typically includes as much customer data as it is possible to know and is required for interaction management prior to the interaction. This typically includes all information about products held, services used, purchases paid, payments, complaint history and crucially all other interaction history, whether this has been controlled by the current interaction management tool or not.

However the context of the interaction also includes non-customer information such as the date, hour of day and the physical location of the interaction. Indeed the physical location can lead to a huge number of new imputed informaton that may be useful to the interaction such as distance to store, distance from home etc.

Thirdly the context of the interaction needs to include the channel and the cpabilities of the channel. Is this a manned or unmanned channel? Does the channel have the ability to capture more detailed context. In online channels this might involve web tacking behaviour on public or private sites, or it might be mouse location or eye tracking information.

In a manned channel the agents capabililty and skills also forms parts of the the context of the interaction. An agent database of skills provides information that could be critical to the management of the interaction.

Use of the customer context:

  • In defining eligibility criteria for recommendations
  • Customer governance rules
  • As an input to real-time scoring
  • Ranking and prioritisation of recommendations
  • Personalization of messages through merge fields
  • Selecting variable content
  • Assisting with the downstream fulfilment process by avoiding replication of data
  • Reporting about the success of interaction management

Channels also often have the capability to capture extra information which could not be known prior to the interaction. This might be as simple as the stated reason for call given to the agent and cpatured through a sdrop down menu. It could be the customer tone of voice captured through sophsiticated voice analysis software or it may be the agents interpretation of that through the use of simple happy, neutral and sad smiley faces. In any case we have to cater for the unknown.