US20070260626A1

Movatterモバイル変換

Info

Publication number: US20070260626A1
Application number: US11/417,118
Authority: US
Inventors: Claudia Reisz
Original assignee: Individual
Current assignee: International Business Machines Corp
Priority date: 2006-05-04
Filing date: 2006-05-04
Publication date: 2007-11-08
Also published as: US8332407B2; US20090055244A1

Abstract

A method and system for customer-choice-based bundling of product options collects data from previous orders about customer component choices, computes a pairwise distance between any pair of components that capture how much the probability of a choice pair P(a.b) deviates from the expected probability under the null hypothesis of independence P(a)*P(b), and clusters the components. The methodology can be implemented as instructions implemented in a computer readable medium. In this way, the need for a method to permit bundles of product options to be configured through the use of business processes reflecting choices based on the preferences of customers rather than the preferences of product designers is fulfilled.

Description

BACKGROUND OF THE INVENTION

1. Field of the Invention

The present invention relates to the field of producing and marketing non-standardized goods.

2. Background Description

Mass-produced goods tend to be highly standardized in order to maximize manufacturing efficiencies. Those efficiencies include, but are not limited to, cost savings from long production runs and the reduction of so-called “set up time” required to change an assembly line over from one configuration to another. Some goods, however, are not as standardized; instead, for certain goods, each item may be configured to meet the requirements of its intended purchaser. In classic “one off” goods, such as bespoke suits or commissioned furniture, unique one-of-a-kind items may be designed on the basis of customer interviews conducted by the individual craftsman who will make the item. In general, the prevalence of unique “one off” products appears to have been in continual decline (relative to the number of highly standardized, mass-produced products) since the development and large-scale implementation of mass production. This is in part attributable to the fact that a high degree of customization undercuts efficiencies that come from the use of mass production techniques.

The persistence of non-standardized goods may be explained by the fact that, even though a high degree of standardization may contribute to efficient mass production, some types of products do not lend themselves to complete standardization. By way of example and not limitation, trucks are typically sold with a broad range of options. In the United States, truck options may even include the availability of alternative engines from different engine manufacturers for use on a single model of truck. This is done to accommodate customer preferences which may be based on efficiency considerations relevant to the customer's business. For example, a customer that has already acquired tools and technical expertise for the purpose of maintaining a particular make of truck engine may not consider the acquisition of a new truck unless it has the same make of engine. (In the United States, not all truck engine manufacturers make trucks and not all truck manufacturers make engines.) Similarly, a customer that typically uses a truck for long-haul purposes is likely to require optional “sleeper cab” facilities that a customer using the same type of truck for local hauling would not be willing to pay for. In addition, requirements for safety and handling options relevant to various types of road and weather conditions may differ from customer to customer.

Thus, notwithstanding the efficiencies available from mass production techniques, there are still products (including, but not limited to, trucks) which must be customized, at least to an extent, because of customer-specific requirements that differ from end user to end user.

In order to optimize the tradeoffs between maximizing production efficiencies while at the same time making products that meet the individualized requirements of particular customers, manufacturers have developed techniques of combining options into bundles so that batches of similarly customized products may be made together, rather than making each customized product individually. Existing approaches to developing such bundles, however, have been driven by the choices of product designers and have not afforded a systematic way of incorporating customer preference data. Product designers do not appear to have had access to an effective way to group product options into customer-choice-based bundles.

SUMMARY OF THE INVENTION

There is thus a need for a way to permit product option bundles to be configured through the use of business processes reflecting choices based on the preferences of customers rather than the preferences of product designers. The present invention makes it possible to identify groups of components that exhibit strong customer-choice interdependencies. Sets of optional product components are combined into bundles, with each bundle (e.g., an all-weather package) comprising a group of possible options found to be significantly interdependent in terms of customer choice data. Bundling choices across multiple product components simplifies the product configuration process—both for the customer and for the manufacturer—and can also translate into increased production efficiency by increased standardization.

Starting from existing data on customer configuration choices (e.g., data recorded of previous order), in one embodiment, the present invention uses a distance measure (chi-square, mutual information, etc.) that can capture the notion of interdependence between any two pairs of components with any number of options. To assess the interdependence of choices, we compute from the customer data all probabilities of customers choosing a particular option P(O)(e.g., probability of ordering a RED truck) and all probabilities of pairs of options P(Oj,Ok) (e.g., probability of ordering a RED truck with GREEN upholstery). The independence of two components (e.g., truck color and upholstery color) is measured by comparing the actual probability of all choice pairs P(a.b) for the two components with the expected probability of observing this choice pair if they were totally unrelated P(a)*P(b). There are various mathematical ways of comparing those two probabilities (difference, ratio, . . . ) and combining them for all possible combinations of options for a pair of components (e.g., the sum of squared differences leading to a chi-squared).

Using any of the possible methods to estimate the pairwise independence, the invention constructs a distance matrix that for any two components contains the numerical value for the degree of independence. The present invention employs a clustering approach that iteratively groups pairs of component groups together if they exhibit strong choice interdependence as revealed by the distance measure. The resulting component groups are identified as potential component bundles.

A producer of a complex non-standardized good may offer a number of different options for different components. The customer constructs his preferred configuration depending on the intended use of the product by choosing one of the available options for each component. The present invention allows the identification of groups of components that exhibit strong customer-choice inter-dependencies. The configuration and production of complex multi-component products, bundling the choices across multiple product components, simplifies marketing and the customer-configuration process.

Thus, one embodiment of the present invention offers a process of recording customer choices for product components, clustering components based on the customer choices, and bundling components based on clusters. An embodiment of the present invention (as shown, for exemplary purposes, in detailed steps inFIG. 1): collects data from previous orders about customer component choices; computes a pairwise distance between any pair of components to capture how much the probability of a choice pair P(a.b) deviates from the expected probability under the null hypothesis of independence P(a)*P(b) (e.g., the sum of squared differences leading to a chi-squared or mutual information); and clusters the components and (e.g., through a hierarchical clustering as shown, for exemplary purposes, inFIGS. 2 and 3)

BRIEF DESCRIPTION OF THE DRAWINGS

The foregoing and other objects, aspects and advantages will be better understood from the following detailed description of preferred embodiments of the invention with reference to the drawings.

FIG. 1 is a representation of the stepwise outline of the business process of the present invention.

FIG. 2 provides a conceptual overview of hierarchical clustering according to the present invention.

FIG. 3 shows detail of a dendrogram and of a distance matrix according to the present invention.

FIG. 4 shows detail of a dendrogram according to the present invention.

FIG. 5 shows an example of a system implemented with a machine-readable medium according the present invention, in which customer choice data is obtained over a network.

DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT OF THE INVENTION

In a preferred embodiment, a method includes the steps of collecting data from previous orders about customer choices for all components, computing a pair wise distance between any pair of components that capture how much the probability of choice pair P(a,b) deviates from the expected probability under the null hypothesis of independence P(a)*P(b), and clustering the components, including but not limited to the use of a hierarchical clustering as shown inFIG. 1.

A product is defined as a set of choices P={C1, . . . , Cn} where Ck C- {O1, . . . , Ok_max}. Product configurations are observed for m customers P1, . . . , Pm. There is defined a distance D: Cm×Cm→R which captures dependence between the two components based on choices. Examples of such distances are:

- Chi-squared divergence of expected probabilities and actual probabilities,
- Mutual Information, and
- Normalized Mutual Information (accounting for effect of varying k_max).
  Hierarchical clustering of components C is employed, such that minimization criteria for choice of cluster may be applied combining:
- Average distance,
- Min/Max distance, and
- Ward approach minimizing the within cluster distance.

Independent components may then be identified and cluster elements may be allocated to bundles.

Referring now to the drawings, and more particularly toFIG. 1, there is shown a representation of the stepwise outline of the business process according to the invention.Product orders110 go through a process of data collection of choices for eachcomponent120, resulting in customer data ofcomponent choices130. Customer data ofcomponent choices130 then goes through a process of estimation of distances betweenpairs140, resulting in adistance matrix150. Thedistance matrix150 goes through a process ofhierarchical clustering160, resulting in adendrogram170 representing a combination of clusters. Thedendrogram170 then goes through a process ofbundle selection180, resulting in option bundles190.

FIG. 2 gives a conceptual overview of hierarchical clustering. Initially, each of the n product components will be assigned its own cluster and the algorithm will start with this initial set ofn clusters135. The algorithm uses thedistance matrix150 to choose an optimal pair ofclusters165 that may be combined. The algorithm adds then removes the two chosen clusters from the set and adds the new, bigger cluster, finishing once all components are joined into a single cluster210. Each such step of combining two clusters is represented in a tree-like graph known asdendogram170 as shown, for example, inFIGS. 3 and 4.

FIG. 3 shows detail of adendrogram170 consisting of 5 components with interdependence as recoded in adistance matrix150. Detail is shown of thedistance matrix150 where, at step1, each component represents one cluster. The two clusters with minimal distance are c1 and c2 and they get joined in step2. Instep3, c4 and c5 are combined based on the small distance between them, step4 adds c3 to the cluster (c4, c5) and finally all components are joined in step5.

FIG. 4 shows detail of adendrogram170 in which identifies potential component bundles as well as a large set of independent components.

FIG. 5 shows an example of a system according the claimed invention, in which customer choice data is obtained over a network. Acomputer510 has a machine-readable medium520 for providing instructions. Anoperator530 is able to provide input via a keyboard540 ormouse545, and the computer is able to provide output via amonitor550 or aprinter555. The computer is connected to anetwork560 to which is connected adatabase570 from which the computer may obtain wholesale price data. Other data may be obtained fromother databases575a,575b,and575cconnected to thenetwork560.

While the invention has been described in terms of a single preferred embodiment, those skilled in the art will recognize that the invention can be practiced with modification within the spirit and scope of the appended claims.