Conjoint Analysis

---

name: conjoint-analysis description: > This skill should be used for conjoint analysis or feature/price trade-off studies in product and pricing work — "run a conjoint", "conjoint analysis", "part-worth utilities", "feature trade-offs", "which attributes matter most", "willingness to pay by feature", "design a conjoint survey", "attribute importance". It generates a balanced profile design and estimates part-worths and WTP with a real regression. metadata: version: "0.1.0"

Conjoint Analysis

Actually run conjoint instead of describing it. Use when a strategy needs to know which attributes drive choice and what each is worth (template §7 Product Offering and §8 Pricing).

Why this skill exists

Conjoint is a true capability gap: a language model can explain the method but cannot build a balanced fractional-factorial design or run the part-worth regression — the source GTM report this suite is modeled on even punted to "estimated" results. The bundled script does both deterministically.

Step 1 — Design the study

List attributes and their levels (include price as an attribute with 2–4 points), then generate the profiles to show respondents:

python ${CLAUDE_PLUGIN_ROOT}/skills/conjoint-analysis/scripts/conjoint.py \
  design --attributes attrs.json --out design.csv

attrs.json:

{"Performance": ["Low", "High"],
 "GPS": ["No", "Yes"],
 "Durability": ["Standard", "Extra"],
 "Price": ["39.99", "54.99", "69.99"]}

The script reports the full factorial size and the minimum profiles needed for estimability, then either shows the full factorial (if small) or searches for a balanced, near-orthogonal fractional design so respondents rate a manageable set. Don't hand-pick profiles — unbalanced designs make part-worths uninterpretable.

Step 2 — Collect ratings

Field the design (panel, customer survey, or sales-assisted). Put the average rating or share-of-preference for each profile in the rating column of the CSV. If you genuinely cannot field a study, you may seed estimated ratings from prior research — but label the output "estimated, pending fielding," exactly as a rigorous report would.

Step 3 — Estimate part-worths and WTP

python ${CLAUDE_PLUGIN_ROOT}/skills/conjoint-analysis/scripts/conjoint.py \
  estimate --responses responses.csv --price-attr Price

Returns:

• Part-worth utilities per level (reference level = 0).
• Attribute importance — each attribute's share of the total utility range (what buyers actually weight).
• Willingness-to-pay — when --price-attr is numeric, the dollar value of each level, derived from the price coefficient. (If the price coefficient comes out non-negative, the script warns — the data is suspect.)

Choosing other methods (know when conjoint is overkill)

• MaxDiff when you only need to rank many items by importance (no price).
• Van Westendorp / Gabor-Granger when you only need a price range, not feature trade-offs. (These are simpler and don't need this script.)
• Full conjoint when feature and price trade-offs both matter — that's this skill.

Output back into the strategy

Use the winning attribute configuration to define the offering (§7) and the WTP to anchor price (§8). If two target segments value attributes very differently, that's the signal to split into segment-specific variants. Keep the design and regression output in the appendix.