Consumers can contribute valuable guiding information to the process of new product development (NPD) in almost every step of the way. By reviewing  academic literature in the areas of NPD and marketing research or by browsing the products and services of major marketing research firms worldwide, one may find an assortment of research methods designed for providing information from a consumer perspective to support product development decisions.

A suite of methods and models developed and organized by a team of researchers at MIT (Cambridge, MA) seems to incorporate the most essential and contemporary  ingredients of a comprehensive programme for NPD research. It is comprehensive in the sense that one may find in it a method most suitable for each of three core stages of an NPD process:

• Generating ideas for the purpose of a new product (i.e., what consumer needs or desires the product will answer) and the approach taken to achieve that;
• Selecting attributes and features to be included in the product based on what consumers value more and value less (e.g., “must have”, “nice to have”, and “better without” features);
• Testing product prototypes, models or concepts that already reached an advanced state of their design.

That research programme is titled the ‘Virtual Customer Initiative’. The methodological approaches may not be new in principle but they have been modified and adapted to be fit for the technology as well as the lifestyle of the 21st Century. The data gathering interface is web-based, that is, the interface with consumers is transported to the virtual world of the Web. The programme further offers new techniques for gathering data on the web that take advantage of and adapt to particular properties of the web environment.

From another perspective, prospect customers or users of a product can be introduced to products in a virtual form before a physical model item has been produced. Particularly in a pretest or test stage, consumers-respondents can see a visual design of a candidate product, possibly rotate its image to be viewed from different angles, without the need yet to produce a physical demo. This can save a considerable amount of time and money for the product development (PD) team.

Conducting NPD research in virtual settings has several attractions. Consumers spend more and more time on the Web, they become more accustomed to the conventions and styles of working with the Internet, and consumers can access the study questionnaire from home or work without arriving to a central facility or be visited by an interviewer. The latter advantage implies potentially greater convenience and ease for respondents and less cost in logistic effort, time and money for researchers and PD teams. There are however some limitations: Consumer panels from which samples are drawn for studies are often still not sufficiently representative of the target populations; without supervision respondents can freely abandon the questionnaire at anytime; and, a self-administered questionnaire (SAQ) must be clear and easy to understand its instructions and informational prompts without guidance or assistance from an interviewer.

Dahan and Hauser (2002) classify the six methods in the suite along two dimensions: products described as “feature-based” or “integrated concepts”, and using “fixed design” vs. “adaptive design” (three levels).

• A feature-based method manipulates and measures the values of features composing a product whereas a method of integrated concepts is concerned with preferences for the whole products as given.
• Adaptive designs for constructing products are recognized as computationally more sophisticated designs compared with fixed designs — adaptive designs are flexible and dynamic,  capable of altering the product stimuli for each customer-respondent in accordance with his or her previous responses. The goal is to produce as small as possible a set of products, and thus a shorter questionnaire, for each respondent, while maintaining a sufficiently efficient design for estimating model parameters (e.g., feature part-worth values).

Dahan and Hauser recommend the use of adaptive algorithms in order to decrease burden on consumers-respondents and increase the likelihood that they complete a shorter questionnaire. Nonetheless, they add that interfaces also have to be interesting and engaging so as to attract and persuade the respondents to stay on to the end of the questionnaire.

I chose to focus below on four of the methods:

In Web-Based Conjoint Analysis (feature-based, fixed design), as since the inception of the CA methodology nearly 40 years ago, respondents are introduced to hypothetical product concepts described as profiles of the attributes or features under consideration by the PD team. A respondent is asked to rank-order or rate the full profiles while trading-off levels from the different attributes composing the product. The set of products is constructed in a fixed experimental design, that is, the set is determined in advance and is presented to all respondents. With a web-based application, researchers may include in addition to verbal descriptions also pictorial illustrations of product attributes and apply interactive displays that improve the communication and flow of the conjoint task.

The FastPace Adaptive Conjoint (more formally: Fast Polyhedral Adaptive Conjoint Estimation — feature-based, adaptive design) is an important and impressive recent development aimed at constructing ever smaller sets of product profiles, customised for each respondent. An advanced mathematical algorithm relatively quickly reduces the space of all possible feature combinations into a smaller set based on answers from earlier steps. The method promises to create smaller adaptive designs than achieved in the veteran Adaptive Conjoint Analysis (ACA) by Sawtooth Software Inc..  Apparently, Dahan and Hauser highlight FastPace by introducing it as the dominant approach to adaptive conjoint designs. But FastPace is not always better than ACA: FastPace has been shown to be superior particularly when it uses prior measures of attribute importance (i.e., additional questions) as ACA does, and there is a major concern of respondent wear out. Notably, the relatively new method of Sawtooth Software of Adaptive Choice-Based Conjoint (ACBC) for choice data is founded on the method of FastPace. They retain a prior set of questions before the adaptive conjoint task, and its combination of attribute-based questions and screening choice questions (an elimination phase of unacceptable products) creates a process that seems more intuitive and natural to consumers than that used in the older ACA.

Adaptive conjoint designs are beneficial for feature-based studies with more than 8 attributes. Because of the complexity of a method and interviewing procedure such as ACBC, it is advisable to consider carefully to what degree it is essential, not an overkill for the problem at hand, and especially see to it that implications of the model assumptions and limitations due to the adaptive process are well understood.

User Design (feature-based, intermediate adaptive design) works like a product configurator — it allows a respondent-product user to choose any feature from a list of available features (e.g., car gearbox: manual), drag and drop it in another list of his or her preferred product features. As a feature is added to his self-designed product the total price is updated. If the respondent regrets, he can return a feature to the availability list. And when he reaches a satisfying design and no longer wishes to make changes, he is asked for the likelihood of purchasing the designed product. It is a feature-based method with a moderate level of adaptation. This method  is advantageous particularly when there are many features to be considered, and furthermore, if there are potential interactions between attributes that need to be considered (estimating interactions in conjoint studies can have a substantial effect on the size of the design). The task is engaging because the participating customer  learns his preferences as he tries out features and builds a product to his liking. This method may be used for a preliminary exploration of preferences for plausible features before a conjoint study. Since in User Design each customer-respondent constructs only a single “ideal” product, this method is more limited when making predictions of preference shares by simulation than conjoint methods.

Virtual Concept Testing (integrated concept, fixed design) is concerned with whole product models that are already fully configured. But there are holistic aspects of a candidate product concept — its design, style and appearance — that need to be tested before a product can be approved. These holistic aspects are matters of impression and appeal that are difficult to breakdown into technical or functional features. Each product in a set is represented primarily by its brand name and a visual image (i.e., identifying the concept), and a price tag. It works similar to a conjoint study but with only two attributes: concept and price. Any additional information on specific attributes, such as ratings of performance, are pre-determined for each concept. Only prices may vary in a controlled manner. In a web-based application there is excellent opportunity to make the display of concepts more engaging and realistic with the use of rich media. Dahan and Hauser refer to an earlier study showing that preferences measured with this method are highly consistent with concept tests based on physical prototypes.

A sensible programme of NPD research reveals itself: User Design provides an initial but broad glance at configurations of features customers would like to find in the proposed type of product ; followed by Web-Based Conjoint Analysis or FastPace Adaptive Conjoint to measure more rigorously preferences for hypothetical product profiles, and estimate the values of attributes; and finally complemented with a Virtual Concept Test to examine how a candidate product model designed by the PD team fares against competing products at a target price.

For more information on these and other methods, I encourage interested readers to visit the website of ‘Virtual Customer Initiative’. You will find in the site brief explanations of each method, more detailed and technical information in published papers and white papers, demonstrations and open-source programme codes to download.

Ron Ventura, Ph.D. (Marketing)

The Virtual Customer, Ely Dahan and John R.  Hauser, 2002, The Journal of Product Innovation Management, 19, pp. 332-353