Survey research that ships models, not raw responses.

Context

Campaign clients need what a poll implies, a propensity model of who’s persuadable, not a spreadsheet of raw responses. And the research has to be trustworthy: one response per real voter, no ballot-stuffing, respondent identities protected.

What we built

We built a survey-research platform with a public research face and an authenticated response system underneath. Survey instruments are versioned; each invitation carries a single-completion JWT token keyed to a subset of the voter file, so a response is one real person, once. A Mailgun send-and-warming pipeline with webhook ingestion and unsubscribe handling drives distribution. Compartmentalization is enforced by design: the deliverable is a propensity model, not the raw responses. The modeling and the respondent data stay separated.

The value

• Clients get the actionable artifact, a propensity model of who’s movable, instead of raw data they’d have to model themselves.
• Single-completion tokens keyed to the voter file make the research defensible: one response per real voter, no stuffing.
• Respondent data is compartmentalized from the client deliverable by design, a privacy posture most polling operations don’t build in.
• A complete pipeline, not a survey form: versioned instruments, tokenized capture, warmed distribution, webhook ingestion.

Client name and identifying details withheld by design. Every figure here is drawn from the system we actually delivered.