The signal beneath the surface

What we built — and why

Every political campaign, public affairs firm, and polling shop faces the same problem: they are flying mostly blind between data events. A poll conducted three weeks ago. A focus group from two months ago. Intuition filling the gaps in between.

Meanwhile, the districts they're trying to understand are constantly generating signal. Millions of people, every day, reading about the topics they care about, consuming content that reflects their concerns, anxieties, and priorities. That signal has always existed. Until now, no one has organized it into actionable intelligence at the congressional district level.

Undercurrent Data was built to solve that problem. We convert the behavioral signal that flows through the open internet — hundreds of millions of events per day — into a structured intelligence product that tells you what every congressional district is actually thinking, right now.

This is not polling

Polling is a snapshot. It tells you what a sample of people said they believed, on the day they were asked. It is inherently retrospective, and it is increasingly fragile — response rates down, samples harder to validate, methodology disputes shaking confidence in the entire category.

Undercurrent is not a replacement for polling. It is something different: behavioral signal. We are not asking people what they think. We are observing what they read, what content they seek out, and what topics they engage with — across millions of data points, every day, at the congressional district level.

The result is an intelligence layer that complements polling, precedes it, and runs continuously between survey waves. When something shifts in a district, we see it in the signal before it shows up in a poll. That 24-hour intelligence latency is not a feature we describe — it is the entire product.

The data model

The Undercurrent platform is built on a proprietary pipeline that processes behavioral content consumption signals from across the open web. These signals are privacy-compliant, cookieless, and structured to meet the requirements of modern data governance. We do not build or sell individual user profiles. We aggregate signal to the congressional district level.

Each signal is geolocated and attributed to one of every congressional district using a census-validated geography mapping layer. Topic classification runs through a multi-layer NLP pipeline that organizes content into a continuously expanding taxonomy of categories. Sentiment scoring applies positive, neutral, and negative valence at the topic-district level. The result: a continuously updated intelligence layer for every congressional district in the country.

Audience matching against voter files and donor lists is achieved through a probabilistic matching methodology that joins signal audiences to named audiences without requiring individual-level data exchange. Privacy by design, accuracy by architecture.

Who we are built for

Undercurrent is an intelligence layer for political professionals who operate at the highest level of strategy. Pollsters who want behavioral signal to inform their questionnaire design and flag emerging issues between waves. Public affairs firms tracking narrative shifts for clients in regulated industries or political campaigns. Campaign strategists who need to know what their districts care about today — not last quarter. Digital persuasion firms that excel at execution and are looking for the intelligence layer to tell them what to say and to whom.

We are not a polling company. We are not a voter file vendor. We are not an ad platform. We are the intelligence layer that sits beneath all of those — the signal that informs strategy before a dollar is spent on execution.

Founding pilot

Undercurrent is currently accepting a limited number of founding pilot clients. Pricing is scoped to your needs — reach out and we'll put together something that makes sense for your use case.

If you are a pollster, public affairs professional, campaign strategist, or digital political firm and you want to see what behavioral signal looks like applied to your districts and your topics, we want to talk to you.