Choose a study for your decision

Test the decision before committing the budget. Subconscious connects controlled experiments on synthetic populations with models of choice, price, product configuration, and competitive response. Start with the action you can take, then choose the study that can distinguish its effect.

The eight decision families below follow the Subconscious business research catalogue. A business workflow can combine an experiment, analysis of a fitted model, and customer-supplied market inputs. The method names are not interchangeable API settings; use the execution compatibility section before configuring a run.

Eight decisions, eight answer shapes

Business decision	Study method	Answer to deliver
What should we charge? Can we take an increase?	Discrete choice with price	Willingness to pay, a calibrated demand curve, and the price that optimizes the stated revenue or contribution objective.
Which features and claims actually earn their keep?	MaxDiff / best–worst	Ranked priorities and differences by segment. For effects of feature levels on product choice, use a conjoint design.
What should the product or pack actually be?	Build-your-own + adaptive choice	A preferred feasible configuration, the trade-offs behind it, and its cost to serve.
Why should a retailer give us the shelf?	Category shelf simulation	Source of volume: incremental category demand versus switching from existing products.
How many will they buy, and how often?	Volumetric choice	Unit forecasts over a stated period, with basket size and repeat-purchase assumptions.
Which message or concept wins, and why?	Concept test + complete ranking	Ranked concepts, the measured response differences, and separately labeled respondent explanations.
Who do we go after first?	Preference-based segmentation	Segments with different preferences, their estimated size and value, and the action that changes choice in each.
What if a competitor cuts price next quarter?	Market simulator on the fitted model	Conditional share, revenue, and margin under specified competitive scenarios.

These are the deliverables to agree in the study brief. A ranking does not automatically provide causal effects, and a fitted choice model does not supply market size, costs, distribution, or repeat purchases on its own.

Bring the inputs that make the answer useful

Decision family	Inputs beyond the common population and study brief
Price	Currency and billing period; realistic tested prices; current offer and competitors; a meaningful outside option; costs and market calibration for revenue or contribution forecasts.
Priorities	A finite list of distinct features or claims; the best–worst task and exposure plan; the segments to compare. Attribute importance from a conjoint is a different statistic from a MaxDiff score.
Product / pack	Feasible attribute levels, prohibited combinations, manufacturing or service constraints, and costs. Record which questions adapt and the rule that selects the next task.
Shelf	Current assortment, proposed additions/removals, availability, an outside option, and a category-demand baseline. Switching within a shelf alone cannot establish category growth.
Volume	Quantity and frequency measures, period, eligible market size, availability, and evidence for adoption and repeat behavior. A probabilistic choice response is not a quantity measurement.
Message / concept	Fixed stimuli, evaluation statements or choice tasks, presentation order controls, ranking rule, and the population. A generated reason is explanatory material, not an identified psychological mechanism.
Segments	Respondent-level preference data; the segmentation method; minimum group sizes; stability checks; external weights and value inputs for market sizing. A sample cluster's proportion is not automatically a market proportion.
Competition	A fitted model; complete competitor configurations using tested levels; baseline and changed scenarios; costs, market size, and distribution assumptions where financial outputs are required.

For every family, record the intervention, comparison, outcome, decision rule, and evidence that would change your mind. Use the study brief and population guide.

Execution compatibility

Choose an entry point by the measured task, not by the business label. The published REST and MCP contracts define what an integration can submit today.

Path	Supported configuration	How to use it
Controlled choice experiment	experiment_type: "conjoint"; response_type: "discrete" or "probabilistic"; reviewed attributes and levels; explicit population.	Use Holodeck, REST, or an MCP draft. Use this for price, product trade-offs, and experimentally varied messages.
Concept evaluation	experiment_type: "concept_testing"; concept description/stimuli and evaluation statements with labels.	Use the experiment request schema or a fully specified MCP launch. MCP draft creation/revision currently covers conjoint. A concept rating task does not itself produce a complete ranking or conjoint posterior.
Analysis of a completed choice experiment	The run's available respondent-level artifacts, exact mapped attributes/levels, and supported segment labels.	Use Analytics Studio or the customer analytics tools. Inspect metadata first; an aggregate-only concept result cannot satisfy a posterior-analysis request.
Dedicated best–worst, adaptive, shelf, volume, or complete-ranking study	A study-specific task, response record, estimator, and agreed deliverables. These method names are not public experiment_type values.	Send the business brief to support@subconscious.ai to confirm the configuration and delivery path before launch. Do not substitute a generic conjoint or concept request and label it a different method.

Options that change the interpretation

• Supplied designs: pre_cooked_attributes_and_levels_lookup fixes the submitted attribute levels. Set null_levels: false to avoid adding a Not available level. This is separate from add_neither_option, which controls the choice-task opt-out.
• Response format: discrete records a choice; probabilistic requests a probability allocation. Neither setting creates a dedicated best–worst, complete-ranking, or quantity/frequency instrument.
• Population: select a supported population mode before adding modeled traits. A modeled trait enriches respondents; it is not verified behavioral recruitment. See Design a population.
• Repeatability: retain the actual design, population settings, seeds, model, run identity, and artifacts. Use reproducible run controls for paired comparisons. Repeating a configuration is different from preserving accuracy across time.
• Uncertainty: distinguish a confidence interval on an estimated effect, variation across respondent utilities, and uncertainty in a calibrated forecast. The request's confidence_level is a configuration setting, not a measured guarantee about the result.

Read the analysis contract

get_analytics_metadata supplies exact labels for the run. get_market_share compares complete product configurations using the fitted respondent utilities; its shares are conditional on the submitted set. get_clusters_or_segments finds preference patterns under its sample-size constraints.

Two names deserve particular care: get_posterior_distribution summarizes respondent utilities with empirical intervals, not population confidence intervals or probability-point causal effects. The legacy get_willingness_to_pay tool returns categorical utility ratios, not currency willingness to pay. A monetary premium requires a valid signed price slope in currency units and an appropriate uncertainty calculation. Read the actual returned scale before naming a business metric.

Turn a scenario into a prospective forecast

A choice model lets you change an offer in the model before changing it in the market. That is the starting point for prospective decision analysis.

To forecast units, revenue, or contribution, connect the modeled response to a declared market size and time period, availability, adoption/choice calibration, quantity and repeat behavior, and costs as applicable. Validate that connection against relevant held-out human or market evidence. Carry uncertainty from both the experiment and these additional inputs into the result.

For a competitor-price decision, preserve the baseline assortment, change the competitor's price within the modeled range, and compare the resulting shares. Add market and cost inputs before presenting a revenue or margin forecast. If a new competitor introduces an untested attribute or changes the decision context, revise the study instead of extrapolating silently.

This lets one fitted model inform multiple decisions while making the conditions behind each answer inspectable. See Causal and prospective analytics for the difference between observing behavior, estimating an intervention, and forecasting its business consequences.

Start with the decision you face

Choose a row, complete its inputs, and name the outcome and decision threshold. Then design the study, review its execution path, and retain the result with its assumptions and validation evidence. For a specialized study, include the chosen method and required deliverables in your support request.