Conjoint Analysis for modelling pricing strategies
Using Conjoint Analysis to model pricing strategies and develop the most competitive product Respondents can answer this type of question rather quickly. The average respondent answers with high ratings while the bottom half of the scale is largely ignored. This results in sub-par data for statistical analysis: skewed distributions, with typically little differentiation between attributes. Such "self-explicated" importances reveal little about how to build a better portable phone. You can't answer the question how much battery life will buyers trade off for a given decrease in weight. Furthermore, stated importances often don't reflect true values; it may be socially desirable to say price is unimportant - after all, respondents don't want to appear "cheap." Yet, in real world portable phone purchase and in the related subscribing decision, price may become a critical factor. Even though you may not care about the statistical shortcomings of self-explicated data, it is obvious that this exercise can't be very realistic. You'll concur that buyers can't always get the best of everything in the real world; buyers must make difficult trade-offs and concessions. When people are forced to make difficult trade-offs, we learn their true values. Conjoint analysis aims for greater realism, grounds attributes in explicit descriptions, and results in greater discrimination between attribute importances. Conjoint creates a more realistic context. The example below from Sawtooth Software's Adaptive Conjoint Analysis (ACA) illustrates this: Conjoint questions can also be asked otherwise: one product profile at a time (the traditional card sort). The rationale behind paired comparisons is this: people can make finer distinctions when they directly compare objects. For example, if I hand you a 4 lb. rock, take it away, and then hand you a 5 lb. rock, chances are you won't be able to tell me which is heavier. However, if you hold one in each hand, you'll have a much better chance of guessing which weighs more. In conjoint analysis people have to compare objects which are defined on various features. For each feature various specifications (levels) are taken into consideration. For instance battery life: 1 day stand by, 2 days stand by, 3 days stand by. But conjoint analysis is more. For each individual we end up with a unique data set reflecting the value of each feature specification. Based on this data set conjoint analysis enables you to predict choices. Another flavour of conjoint analysis offers even greater realism: Choice-based Conjoint. The example below regarding portable computers shows how Choice-based Conjoint (CBC) approaches the same subject: Choice-Based Conjoint questions closely mimic what buyers do in the real world. Including "none" as an option enhances the realism, and allows those respondents who are not likely to purchase to express their disinterest. Choice-based data reflect choices, not just preferences. If we agree that the ultimate goal of market simulators is to predict choice, then it's only natural that we would value choice-based data. You don't need to know about the underlying orthogonal designs, main-effects assumptions or how utilities are derived. I will probably bore (or even annoy) you if I would tell you. Instead, you should grasp that realistic models result from realistic questioning methods, and be comforted that conjoint is a reliable, time-proven method. The main categories and applications of conjoint analysis 2 The two examples shown before represent the two main categories of conjoint analysis. The Adaptive Conjoint system belongs to the category of rating systems. The adaptive element implies that that the questioning is tuned to each respondent separately. Respondents do not have to evaluate all features at the same time, which helps solve the problem of "information overload". A trade-off with more than four features is hard to rate, as is shown in the following example. Choice-Based Conjoint belongs to the category of discrete choice models. Both methods (rating based and choice based) have shown to produce generally similar results. But they are in no ways substitutes. The strengths or primary advantages of Adaptive Conjoint Analysis are: 1. You can include many features and estimate the relative value of these features. But at the same time it is not wise to burden the respondent with too much information. Theoretically 30 features (or attributes) can be included in a single study. But you should be very careful in doing so, because it is a serious mistake to add non-critical attributes just because the capacity is there. 2. The relative value (utility) for each feature specification is calculated for each individual. This creates a number of opportunities for additional analyses, of which, in my opinion, the most important one is to search for groups of individuals with the same value structure (benefit segments). But you can also do it the other way around, to analyse the value structure within known segments. 3. The interview is respondent specific. Based on the answers given in the beginning, the interview is customised and questions are asked about the features and feature specifications, which are most relevant to each respondent's decision. The two product concepts to be traded off are constructed in such a way that they should be more or less equally preferred, according to the respondent's answers specified before. A difficult choice provides better information for refining the estimate of the individual's value system than an easy choice or an obvious choice does. But there are also a number of limitations, which have to be taken into consideration when deciding which conjoint technique to use. 1. The attributes are measured independent of all others and the ability to measure feature interactions is limited. In situations where interactions are common, such as brand and price, it is not advisable to include these features in the adaptive conjoint analysis or it is better to choose for choice based conjoint. 2. Including many features implies that many parameters have to be estimated, which takes time and consequently money (longer interviews). 3. The impact of critical factors can be underestimated. As the number of features included increases the impact of a critical feature can be dampened. The consequence is that the preference for a product with many advanced product features combined with a high price will be overestimated. In these situations it is advisable to follow the dual conjoint-approach. This implies that for instance all technical features are included in a first conjoint and in the second conjoint, technical concepts (combinations of technical features) are traded off with price and brand. The strengths and primary advantages of Choice-Based Conjoint analysis are: 1. The choice task is more realistic. As in the real world, respondents can decline to purchase by choosing "none". The simple task also mimics what people do virtually every day, evaluating a set of products and selecting one. 2. The interview can be short. Because the choices are pooled over a group of respondents, the information to be obtained per individual may be limited. A second advantage of the limited number of questions needed is that one can repeat the choice tasks later in the interview, for instance to measure the impact of additional information. 3. Because the analysis is on group level it is possible to measure two way interactions, provided the size of the group is large enough. These interactions are particularly useful in brand equity research in which brand specific price sensitivities are of concern. The Choice-Based Conjoint disadvantages: 1. Individual preferences or utilities cannot be determined, because the analysis is on group level. This disadvantage can partly be overcome by segmenting the individuals, based on choices made or based on other characteristics, and then estimate utilities per segment. 2. The products can be defined on maximum of six features and, consequently, analysis is limited to six features. But even with six features the choice tasks are rather complex, and respondents may simplify the task by focusing on a few features only, which leads to less precise estimates. The advantages and disadvantages of both conjoint techniques partly compensate each other and that is why the techniques are complementary. Therefore it is often advisable to use them in combination. The use of conjoint in cellular pricing research When conducting a cellular pricing survey at SKIM we present respondents for instance with varying subscription alternatives and ask them which they would choose. Each subscription alternative is defined by the provider (PTT or Libertel) and by the additional services (a standard package and the advanced package). Each subscription alternative is presented at different prices throughout the interview. The prices are linked with the alternatives, so the advanced package is combined with a range of higher prices than the standard package. The percentage of times respondents choose each subscription alternative at different prices reveals preference and price sensitivity for subscription alternatives. Compelling demand curves result when we plot these "wins" by price point and connect them with smooth lines, as shown below for the two providers and their two hypothetical packages.  What we see is that at equal prices the advanced packages are preferred over the standard packages and that the chance PTT is chosen is always higher than the chance that Libertel is chosen. Which implies that the brand equity of PTT allows them to ask a higher subscription rate. In this hypothetical example the premium PTT can ask for her standard package is NLG 14 (this is the difference in the subscription rate at an equal probability of choice of 20%). But more important, we can simulate various strategies. Let us assume that the actual situation is that subscription rates of PTT and Libertel are the same (NLG 30 for the standard package and NLG 45 for the advanced package). If PTT lowers the standard package subscription rate with NLG 5.- and Libertel follows (scenario 2) we can see that PTT will loose share, because the slope of the price curve is less steep than the slope of the curves from Libertel. Other actions and reactions may also be simulated.  From the chart we may also conclude that there seem to be two separate segments. The advanced segment does not react to a price reduction of the standard segment (scenario 1,2). However when the prices of the advanced segment are reduced the standard segment is cannibalised (scenario 3,4). Although very appealing and effective it is advisable not to stick to this type of analysis but to look also behind the data. Analysing the choices the respondents made we find that some respondents always choose the advanced package and others always the standard package. Which explains the results of the simulations. But also within the segments we see that some respondents always choose PTT, some always choose Libertel and only 50% selected both depending on price and package.  So in addition to the overall analysis and simulation it is wise to identify segments and analyse and simulate per segment. In the example above we deliberately limited price to the subscription rate. When simulating the impact of strategies it is essential to make the choice task for the respondents not too complex and to make it as realistic as possible. However, the number of pricing alternatives is almost endless and it might well be that subscription rate is a pricing element of minor importance. In that case you should trace first what pricing elements trigger the response. The technique used is the same but the application is different.  When responding to these choice tasks the situation should be defined first. For instance, it should be specified which type of subscription the respondent has or should take in mind. The primary results indicate which pricing elements contribute most to the acceptance or rejection of an offer. By simulating the choices it is possible to find the combination of pricing elements which is most effective when trying to persuade the potential subscriber to choose you as a provider and at the same time trying to maximise profit or sales. As in the previous example, it is most important to look behind the data and try to find if there are categories of customers that are focused on specific pricing elements, for instance customers who are driven by "off-peak hour costs per minute" or by the "fixed monthly fee (subscription rate)". If there are, one should try to optimise the pricing combination for these segments. Various studies have revealed that after having become a subscriber the sensitivity to the costs per minute changes, especially by the corporate decision makers. Therefore it is important to realise that the conjoint results reflect the value system at a certain period in time. Over time the conjoint results may change. When conducting a conjoint study one should always register a number of identification and background data and we strongly advise to stratify the sample to include various target groups and analyse per target group. To optimise the package and analyse which value added services are preferred most it is advisable to use the Adaptive Conjoint Analysis technique. The reason for this is that there are many services and product features. Combining price with these features in the same conjoint makes sense if you want an idea of the value of these features in money terms. But at the same time you have to be very cautious when using these results in your market simulations. The attractiveness of expensive features rich products or services mostly overestimated. Besides there is the same problem as indicated before: which "pricing element" should be included? When conducting such an adaptive conjoint one should be aware that respondents stick to the features as specified. Example: In an adaptive conjoint for the development of new portable radio phones the feature specification "being reachable all over the country" was directly and implicitly associated with cellular phones. And even the respondents who were mobile within 10 km only were extremely sensitive to the feature because they thought it would be cellular. Because in adaptive conjoint we measure the value of the features for each individual it is possible to analyse if there are groups of individuals who are sensitive to the same feature specifications. Interpreting these combinations of features leads to a better understanding of the needs. In fact one is looking for benefit segments which lead to Value Added Packages for specific target groups. Instead of features one should include these packages in the trade-offs with price in the primary choice tasks. To summarise, conjoint analysis can be used in various ways in product and pricing research. In designing a conjoint pricing study one often has to break the problem down in a hierarchy of conjoint modules. Each module has to be placed in its own context. The results of conjoint studies are appealing but one should realise that additional questions have to be asked and included in the analysis. For instance: to estimate price elasticity one should at least ask additional questions about price awareness and the awareness of the services; about the volume or user habits; about who decides and who pays; etc. Besides it is extremely important that you define the target groups to be included in the analysis. A requirement which is critical and often forgotten is that simulation scenarios have to be defined in advance. When defining the simulation scenarios you always find that the attributes have to be defined slightly differently; some features have been forgotten and some features are the same for all providers and will not change, meaning they are not relevant for the analysis and could have been left out. Conjoint models do not predict market share due to a variety of reasons, including: 1. Conjoint assumes perfect information. In the conjoint interview, respondents are educated about available brands and features. In the real world, obscure brands have less chance of being purchased. Conjoint analysis cannot fully account for differences in awareness or preference developed through advertising and promotion. 2. Conjoint assumes all products are equally available. One brand is as conveniently selected as another in a conjoint interview. Using conjoint and combining it with other techniques, like benefit segmentation, leads to a better understanding of the market and of the customers' needs and choices. It is this better understanding which increases the validity of the forecasts, more than whatever statistical fine-tuning. Summary   Conjoint analysis increases the return on research dollars by providing managers with useful, valid information. Conjoint's realism leads to more accurate results, and provides a strategic tool for quantifying brand equity and relative price sensitivity. To ensure success, researchers must carefully set management's expectations regarding what conjoint can and cannot do.Conjoint analysis can be used in various ways in product and pricing research. In designing a complex conjoint pricing it is advisable to break the problem down in a hierarchy of conjoint modules. Each module has to be placed in its own context. The market simulator is usually the most anticipated deliverable for managers. Don't let this enthusiasm get out of hand. Conjoint simulators are directional indicators that can provide a great deal of information about relative feature importances and preferences for product configurations. Conjoint simulators are not market share predictors. Many other factors such as awareness, distribution, advertising and product life cycles drive market share in the real world. While conjoint models can be fine-tuned to partially account for these elements, we must not let managers believe that adjusted conjoint models can accurately predict volumetric absolutes such as market share.Literature 1 Based on "Helping Managers Understand the Value of Conjoint", by Bryan Orme, Sawtooth Software, Inc. 2 Based on "ACA, CBC, or Both? Effective Strategies for Conjoint Research", by Bryan Orme, Sawtooth Software, Inc., 1996. |
 Applied methodologiesTechnical papers |
#01
Product manager, Pharmaceutical Company
#01 "The team walked out of SKIM's presentation
session with an exact understanding of what
needed to be done to improve the product's
design"
|
