Dataset analysis

Upload a data file and interrogate it in conversation: what the variables are, how the answers distribute, what predicts what, and what it looks like as a chart.

Availability: this runs on a dedicated deployment rather than from the shared integrations list. Talk to your Customer Success Manager about setting it up for your environment.

What it is

A statistical analysis connection for a data file you provide — typically a survey export, in CSV or Excel. The file is uploaded, the integration describes what is in it, and from there the analysis happens as questions rather than as a script.

It covers the standard survey-analysis repertoire: frequencies and cross-tabulations with segment filters, linear and logistic regression for driver analysis, principal component analysis, and charts.

Charts come back as images embedded directly in the conversation, so nothing is loaded from an external address.

Use it for

• “Describe this dataset — what variables are in it?”
• “Give me the frequencies for this question, split by this segment.”
• “Which variables best predict this outcome?”
• “Run a factor analysis and tell me what groups together.”
• “Chart this as a grouped bar chart.”

Tools

Function	Answers
Usage guide	The order to work in, and how to reference variables — the assistant reads this first
Upload link	Where to put the file, when it is not attached to the conversation
Describe dataset	The variable table: what is in the file and what each variable is called
Frequencies	Distributions and cross-tabulations, with segment filters
Regression	Linear and logistic regression for driver analysis
Factor analysis	Principal component analysis over the variables
Visualise	Bar, grouped bar, pie, histogram, scatter or heatmap, as an embedded image

Data and permissions

• Your file, held for the analysis. The upload lives in the integration for the working session; it is not a permanent store and nothing is indexed.
• Charts are embedded images, so the conversation loads nothing from an external address.

Getting started

Either attach the file to the conversation, or upload it and paste back the reference. The usage guide covers both routes.

Limitations

• Statistics, not interpretation. A regression reports an association. Whether it means anything causal is a judgement the method cannot make, and an assistant restating a coefficient as a cause is the failure mode to watch for.
• Garbage in, confident output. The tools will happily run on a badly coded file. Describing the dataset first is not a formality.
• One file at a time. Joining datasets is not part of this.
• Not a permanent store. The uploaded data does not persist as a dataset you can come back to later.