Market modeling has gotten complex and highly predictive over the years, with models demonstrating never before seen accuracy. With the introduction of advanced agent-based models to handle a wide range of market scenarios, there are few markets that models cannot adequately represent today.

In most cases models are of one of three types:

  • Retail Shelf Based Models
  • User Configured Product Models
  • Recommendation / Prescription Models

Retail Shelf Based Models represent retail shelves, for instance for fast moving consumer goods. In these cases, consumers choose from one of an available set of predetermined options. These models are run in most cases to:

  • Optimally price a product or product line, including adopting product versioning
  • Predict demand for new product designs, allowing for rapid, low-risk, prototyping
  • Uncover preference structures (utilities) and the importance for existing and new products
  • Identify buyers and competitors for each product which enables advertising to be targeted

While many clients may be familiar with a forerunner of today’s far more sophisticated, accurate, and powerful models known as conjoint analysis in many different forms including choice based conjoint, adaptive conjoint and or even adaptive choice based conjoint, the specific form of the model should be determined by the needs of the client and the analysis. Most often however, market-replicating discrete choice analysis is the preferred method. Compared to conjoint analysis in all of its forms, market replicating discrete choice provides far better matches with the actual market, and as a result accurate outcomes, better understanding of the effects of decisions on market share and profitability, replacing importances, and can be completed within the survey in about a third of the time with a much lower abandonment rate. Please request our white paper on the difference between conjoint and market replicating discrete choice for more information on these differences, based upon our paper to the 2008 ART Forum, which uses the home gaming console market as an example of a retail shelf based market.

User Configured Product Models represent products that buyers can configure, for instance servers in the technology market, as well as markets in which the product delivered is the result of decisions made based upon a menu. These models are sometimes referred to as Menu Based Choice models. These models are run in most cases to:

  • Optimally price one option relative to another, where the options compete against each other, rather than products from other firms
  • Optimally price ancillary goods and services sold in addition to a retail shelf product
  • Optimally bundle goods and services that are complementary with one another

In these models the core question is usually how one feature will complement or compete with other available options or products. Whether the product is an automobile, heavy construction equipment, a server, a meal at McDonalds, a rental car with ancillary products or services, or a the sum product of a retail shopping visit, the appropriate model often is the user configured product or menu based model.

Recommendation / Prescription Models represent a different type of model, focused on the product or configuration of products a professional recommender would select given the specifics of the actual buyer or situation rather than a choice they would make for themselves. In these models the buyer or patient is experimental, not just the product. These models are run in most cases to:

  • Determine the type of patient for which physicians feel a new treatment is appropriate
  • Estimate demand for and cannibalization patterns for new treatments and therapies
  • Test the effectiveness of new claims or advertisements, for instance in a clinical or retail environment

These models are very common in the pharmaceutical industry and are just beginning to make their way into retail and related environments. In these models, it is quite common for both the patient and the drug to be experimental. These models represent an important class of models for select types of clients.

If you came to this site looking for traditional versions of conjoint analysis, the above descriptions of model types should give you a sense of how advanced and powerful the alternatives to simplified conjoint really are.

Making a Choice

While many firms think they have a good conjoint guy the truth is that there are few firms and even fewer marketing scientists capable of providing these types of accurate and optimizable client solutions. Beware of relatively low cost tools from vendors like Sawtooth Software, which seek to dumb-down the process.

As researchers and consultants we all know that we need to pay particular care to make sure the market properly represented, and this does not stop with sampling. Simplified market representations are not sufficient for accurate models. In head to head comparisons our models beat Sawtooth CBC models 17:1 in error rates, and that same model stayed accurate despite major market changes for more than a year, with a median mean absolute error of less than 2% and a correlation with actual market share of 99.8%. This compares to the standard conjoint model median mean absolute errors between 13% and 22% and correlations between averaging just 28%.

The net value of that difference for that model was estimated at between $680 million and $1.1 billion. More commonly, based upon these highly accurate models, provable results will typically add between $10 million and $50 million in improvements over base case, helping to ensure a solid bond between our clients in marketing research, consulting, and advertising firms and their end clients.