Customers Are Not Widgets: CRM, BPM and Improving the Customer Experience

I’ve been researching software to model the flow of customers through an organization. The thought is to take literally the mantra that “the value of a business is the value of its customer relationships” by modeling the relationships individually and viewing an aggregate result. This is all central to the Client X Client business concept.

The obvious way to model the relationships is to define a sequence of stages and then specify the percentage of customers who graduate from each step to the next. For customers, these stages might be first order, second order, third order, and so on. The technical name for this type of model is a Markov chain, which is defined by www.computeruser.com as “A random process in which the probability that a certain future state will occur depends only on the present or immediately preceding state of the system, and not on the events leading up to the present state.”

Such models are often used in business process management to define the flow of items through a standard process such as an assembly line. In models that are realistic enough to be useful, the chains can incorporate multiple paths, such as detours to repair defects or install options, and the stages themselves may have characteristics such as capacity constraints. Within a CRM system, call center activity is often modeled this way to understand the impact of staff, configuration or rule changes on wait times and costs. So long as the basic Markov condition is met—that every object within a given stage is equally likely to progress to the next stage—these models can work without actually tracing each object individually.

But customers have memories, while most widgets do not. So a customer’s future behavior may depend on how they were treated during a call center interaction, and any model of the long-term customer relationship must take this into account. This requires a different non-Markov modeling technique, in which the past experiences of each customer help to predict the customer’s future behavior. I believe the approach called “agent-based simulation” will allow this, although I haven’t yet found out for sure. But whether or not that turns out the be the proper technique, the important point for now is any attempt to optimize the customer experience must be able to simulate the impact of each interaction on all future interactions, something that a simple Markov chain can never do.