Every purchase your customers make involves a choice among alternatives. They weigh a handful of options, consider the factors that matter to them, and pick one -- or walk away entirely. If you want to predict those choices before committing to a product configuration, pricing tier, or go-to-market strategy, you need a discrete choice model that mirrors how real decisions actually work.
Discrete choice models simulate the decision process by presenting individuals with realistic choice scenarios -- sets of alternatives defined by varying levels of key factors like price, features, brand, and service terms -- and observing which option they select. At Meraki Research, we build and apply discrete choice models that go far beyond academic exercises. Our models are calibrated to your market, your competitive set, and your strategic questions, giving you a decision-support tool you can use to simulate "what if" scenarios long after the research is complete.
Our analytics team has built discrete choice models for subjects ranging from consumer packaged goods and healthcare to financial services and transportation. We combine deep methodological expertise with practical business framing so that every model we deliver is a tool your team can actually use, not a statistics lecture they file away. The difference between an academic choice model and a commercially useful one is activation -- and that is where our work begins.
Our Approach: The Choice Architecture Method
Discrete choice modelling sits at the intersection of economics, psychology, and statistics. The theoretical foundation traces back to Daniel McFadden's random utility theory -- the same framework he won the Nobel Prize for. The practical application involves designing choice experiments, fielding them with real people, and estimating mathematical models that quantify how each factor influences the final choice.
We follow a five-phase process we call the Choice Architecture Method.
Phase 1: Decision Mapping and Attribute Identification
Before building any model, we map the actual decision your customers face. What alternatives are they choosing between? What factors drive their preferences? Where does the choice happen -- online, in-store, through an advisor?
We conduct stakeholder interviews and secondary research to identify the attributes and levels that define each alternative in the choice-set. For example, a subway system evaluating fare structures might define alternatives by price, frequency, travel time, and seat availability. A software company evaluating packaging might use feature bundles, support tiers, contract length, and monthly cost. The key is selecting factors that are both strategically relevant and realistically varied in the market.
Phase 2: Experimental Design
The experimental design determines which combinations of attribute levels appear together in each choice-set and how many choice-sets each respondent evaluates. This is where the statistical rigor lives.
We use D-efficient or Bayesian optimal designs that maximize the information extracted from each choice task while keeping the respondent burden manageable. A typical study presents 8 to 12 choice-sets, each containing 3 to 5 alternatives plus a "none" option (because in the real world, people can always walk away). Our designs account for attribute-level balance, orthogonality, and overlap restrictions to ensure the resulting models can cleanly estimate main effects and key interactions.
Phase 3: Fieldwork and Data Collection
We embed the choice experiment within a broader survey that includes screening, context-setting, and attitudinal questions. Each respondent evaluates a series of choice-sets, selecting their preferred alternative from each set. The task is designed to feel natural -- it mimics the kind of choice people make in the real market.
Quality controls flag individuals who speed through the exercise, always pick the cheapest option, or show random response patterns. We remove low-quality records before estimation because a discrete choice model is only as reliable as the choices it is trained on.
Phase 4: Model Estimation and Validation
We estimate multinomial logit, mixed logit (random parameters), or latent class models depending on the complexity of your market and the heterogeneity of preferences across your audience.
- Multinomial logit provides aggregate-level part-worth utilities -- a clean, interpretable baseline.
- Mixed logit captures individual-level variation in preferences, revealing that different people weigh the same factors differently.
- Latent class models identify distinct preference segments within your audience -- groups of individuals who share similar choice patterns.
We validate every model using holdout choice tasks (sets withheld from estimation) to confirm that the model predicts actual choices with high accuracy. A model that fits the estimation data perfectly but fails on holdout data is overfit and unreliable. We do not deliver models that cannot predict.
Phase 5: Market Simulation and Strategic Application
The estimated model becomes a simulation engine. We build an interactive choice simulator that lets your team define any combination of alternatives -- your product at various configurations and price points, plus competitors -- and predict market share, switching behaviour, and revenue impact.
Want to know what happens if you add a premium tier? Change the price by 10%? Drop a feature? Launch against a new competitor? The simulator answers these questions in seconds. This is the function that makes discrete choice models a lasting strategic asset rather than a one-time research report.
We train your team to use the simulator independently so the model continues to generate value long after our engagement ends. Teams that run regular scenario analyses -- before quarterly pricing reviews, product roadmap sessions, or competitive responses -- extract far more return from the original research investment than those who treat the model as a one-time deliverable.
When Discrete Choice Modeling Is the Right Approach
Discrete choice modelling is the right tool when:
- Your question involves trade-offs among multiple factors -- price, features, brand, service terms -- not just a single dimension.
- You need to predict market share or switching behaviour under different competitive scenarios.
- You want a reusable simulation tool that supports ongoing strategic decisions.
- The choice context involves clearly defined alternatives that can be described by a structured set of attributes.
Discrete choice modelling is not the right starting point when you do not yet know the relevant attributes or acceptable price range. In those cases, qualitative research, MaxDiff Analysis, or Van Westendorp should come first. If your question is purely about price -- no feature or brand trade-offs -- Gabor-Granger provides a faster, more focused answer.
We are also honest about complexity. Discrete choice models require careful design, larger sample sizes, and more analytical investment than simpler pricing methods. If your decision does not warrant that level of rigor -- for example, a minor price adjustment to an existing SKU -- a lighter approach may be more appropriate and cost-effective.
What We Deliver
Every discrete choice modelling engagement produces a comprehensive set of tools and outputs:
- Part-Worth Utility Report: The relative importance of each attribute and the utility of each level within each attribute. This tells you exactly which factors drive choice and by how much.
- Market Share Simulator: An interactive, Excel- or web-based tool your team can use to model competitive scenarios, test new configurations, and evaluate pricing moves. This is the deliverable teams use for months (sometimes years) after the study.
- Preference Segmentation: Latent class or cluster-based segments defined by choice patterns, not just demographics. This pairs directly with our Market Segmentation Research practice.
- Sensitivity and Elasticity Analysis: How market share responds to changes in each attribute, with special attention to price sensitivity across segments.
- Strategic Recommendations Deck: A 40-to-60-minute executive presentation that walks your team through the model, the simulator, and the scenarios most relevant to your immediate decisions.
Discrete choice modelling engagements typically run six to ten weeks from kickoff to final deliverables, depending on the number of attributes, the complexity of the design, and sample requirements.
Industry Applications
Consumer Packaged Goods: Optimizing product configurations, pack sizes, and price tiers in competitive shelf sets where multiple factors drive the purchase choice.
Technology and SaaS: Designing subscription bundles, hardware configurations, and service tiers where people trade off features against price and brand.
Healthcare and Pharma: Modeling patient and physician preferences for treatment alternatives defined by efficacy, side effects, dosing, and out-of-pocket cost.
Transportation and Public Policy: Evaluating preferences for transit options -- for example, a subway system modelling ridership under different fare and service configurations.
Financial Services: Optimizing account packages, loan terms, and insurance coverage options where multiple attributes define each alternative.
Retail and Hospitality: Testing loyalty program structures, room or experience configurations, and promotional bundles.