A CBC study lives in the Trade-off Platform. Its header shows the study name and three steps — Design, Analysis and Simulator — each unlocking when the one before it is done.
Design the study — or import yours
The Design step offers two ways in.
Design it here. In Attributes & levels, define what the product is: every attribute needs at least two levels, and you can click any cell to edit it. Then:
- Prohibitions — click a cell to ban two levels from ever appearing together (for example a flavour that is never sold in a bottle).
- Price mapping — give each price level a dollar value; it is used for willingness-to-pay and profit later.
- Design — plan the fieldwork: Respondents, Tasks, Options per task and Survey versions, and whether to include a "None of these" option.
Press Generate design. It can take a few minutes; you are told when "A choice design is ready." You get the design facts (rows, versions, tasks, concepts), downloads of the design file and its report, and a preview of the generated tasks.
Start from a file you already have. Upload the design file you already have. The AI reads the file and proposes how its columns map; a mapping that passes the checks is applied, and you can review or revise it any time. The design is shown back to you in Crowdmines' own format.
Analyse the results
1 · Respondent data. Choose where the choices come from — every number in the simulator comes from this pick:
- Synthetic panel — simulated respondents answer your design, so the whole workflow runs before fielding. The fit scorecard is real; the preferences are invented.
- Fielded results — upload what real respondents chose. Either layout works: one row per respondent × task × option, or one row per respondent (joined onto your design). Choose the File layout — Detect, One row per task, or One row per respondent — then the AI maps the columns for you to review.
2 · Estimation. Hierarchical Bayes fits one set of part-worth utilities per respondent; the defaults work for most studies. Pick a preset, or set MCMC iterations and Burn-in yourself, choose to Run on CPU or GPU, and press Run HB estimation. Estimation keeps running if you close the tab.
The model. When it finishes you see the checks: RLH · fit, Holdout hit rate and Recovery ρ (parameter recovery on a synthetic panel), with counts and an aggregate MNL benchmark beside the model.
Explore the simulator
Open the Simulator. Its rooms are grouped as The model, This market and The people:
Room What it answers Importance Which attributes drive choice, and what each level is worth Counts + MNL The model-free counts and the aggregate benchmark Shares Shares of preference for a shelf you build, with a What-if Optimizer The best configuration to offer Portfolio What a new product takes, and from whom WTP (with Sampling) Willingness to pay for a change Segments How any of this differs by group Each room starts from a question card: set it up, press the button, and the answer folds the card away. Kept runs are compared to a baseline.