How to run a Van Westendorp analysis

The user types something like:

• "What's the right price range for Product X?"
• "Run a Van Westendorp on the pricing data."
• "How sensitive are customers to the price of our new subscription?"

The platform recognises the intent and selects Van Westendorp automatically — the user doesn't need to name the method if their question is clear enough.

The chat asks the user to confirm which columns in their dataset correspond to the four PSM questions:

• Too cheap
• Cheap / bargain
• Expensive
• Too expensive

If the column names are close enough (e.g. Q12_too_cheap), the platform suggests mappings automatically and asks the user to confirm. If the names are ambiguous, it asks the user to pick from a list.

The chat asks whether the user wants to filter the data before running the analysis. Examples:

• "Only respondents from the US"
• "Exclude anyone who didn't complete the survey"
• "Only people who saw Concept B"

The platform validates that the filter leaves enough responses (≥ 30). If not, it warns the user and suggests widening the filter.

The chat asks whether the user wants to split the analysis by a categorical variable:

• "Split by region"
• "Compare age groups"

If set, the analysis runs independently per group and results are returned side by side.

The chat shows a final confirmation screen summarising:

• Which columns are mapped to which PSM questions
• Any active filters
• The subgroup variable (if set)
• The expected sample size

The user confirms, and the analysis runs.

Within seconds, the chat returns:

• A PSM curve chart — four cumulative curves with the four intersection price points marked
• A price points table — PMC, OPP, IPP, PME with their values
• Summary statistics — sample size, mean, median
• A written interpretation — a plain-language summary of the findings

If subgroups were requested, each group gets its own chart, table, and interpretation, plus a comparison summary.

The user can immediately ask follow-up questions in the same conversation:

• "Now split that by gender"
• "What if we exclude the bottom 10% of income?"
• "Run the same thing for Product Y"

No re-uploading, no re-explaining context — the chat remembers the dataset and prior analysis.