Individual level, not aggregate

The average consumer does not exist. Simulation models that work at the individual level capture heterogeneity in preferences and give markedly better forecasts than aggregated models.

Most standard reports from conjoint studies present aggregated results: average utility weights for each attribute level. The problem is that the average hides the real preference structure. If half of consumers strongly prefer red and half strongly prefer blue, the average is "neutral to color" — a result that describes exactly no actual consumer.

Individual-level simulation works with each respondent's unique preference profile. Instead of calculating average utility weights and simulating with them, the model simulates each person's choice separately and then aggregates the results. It captures the full variation in the market.

Reflect always simulates at the individual level. Hierarchical Bayes estimation gives us individual preference profiles that respect both the individual respondent's data and the overall population distribution. That gives stable individual estimates which then drive a realistic simulation of the market.

Key takeaways

The average consumer does not exist, aggregates hide heterogeneity
Individual-level simulation captures the full variation in the market
Hierarchical Bayesian estimation gives stable individual profiles
Aggregated models underestimate the size of niche segments
Individual level gives markedly better forecasts with heterogeneous preferences

Example

A telecom operator simulated three new price plans. Aggregated model: plan B wins with 35% share. Individual-level model: plan B wins in aggregate but reveals a large segment (22%) that strongly prefers plan C. The operator kept both plans and captured 12% more volume than if they had only launched B.

When conjoint works and when it does not

Conjoint works best when the consumer makes conscious trade-offs between clear attributes. It works poorly in low-involvement categories, with habitual behavior, and when price dominates the decision.

First choice vs share of preference

First choice shows what the consumer picks first. Share of preference shows how preference is distributed. Which metric is right depends on the category's purchase behavior, and they often give entirely different answers.

How simulation should adapt to category

The same simulation model does not work in all categories. Purchase process, involvement, repertoire behavior and price sensitivity vary, and the simulation model must reflect that reality.

Reflect simulation model

Our simulation model works at the individual level, is calibrated against observed data, and adapts to the category's purchase behavior. The result is forecasts that hold up, not just in the presentation but in the market.

See related service →

Discuss your simulation models with us

← Back to Simulation that reflects real individuals, not averages

Individual level, not aggregate

Key takeaways

Example

Related articles