How to Win an Election. Just Win.

Kansas had not elected a Democrat to the Senate in almost a century. On 11/03/2026 it did. Adam Hamilton beat Roger Marshall, and Kansas was one of seven Republican held Senate seats that changed hands in a single night.

Nobody won Kansas with a good television spot in October. That race was winnable because somebody decided it was worth contesting while the ratings still called it safe, found a candidate who could carry it, and worked out what Kansans were angry about before the other side did. Winning is a sequence of decisions like that, taken early, in order, with incomplete information.

Campaigns lose for reasons that repeat. They spend money talking to people whose minds are closed. They argue about the thing the candidate cares about instead of the thing the voters care about. They buy their information in September, when every answer it could give them is already too late to use. None of that is bad luck.

How you actually win an election

Find the people who can move

In any race the electorate splits three ways. Most people have decided and will not change. A slice is genuinely open. Another slice agrees with you already and will only show up if something makes them. Three populations, three different jobs.

The expensive mistake is treating them as one audience. A campaign that buys broad reach to persuade is paying to talk to a majority of people who are not listening, and the waste shows up later as a flat trend line that somebody will describe as a tough environment.

Work out who is actually persuadable in your race before you spend anything. In a school board race that might be two thousand households. In a Senate race it might be several hundred thousand people. The number changes by two orders of magnitude. The discipline does not change at all.

Turnout and persuasion are two budgets

Persuasion changes a mind. Turnout changes a behavior. They run on different lists, different messages, different channels and different calendars, and they compete for the same money.

A campaign that cannot tell you what share of its budget is going to each one does not have a strategy. It has a list of purchases. Set the split on purpose, write it down, and move it when the evidence says to, instead of letting whoever is most persuasive in the room move it for you.

The right split is race specific. A municipal election in an odd year, with turnout in the teens, is almost entirely a turnout problem. A Senate race where both sides are saturating the state tilts toward persuasion, because nearly everyone who can be pulled to the polls is already being pulled by somebody.

Start from what voters already want

Most campaigns start from what the candidate wants to say. That is backwards, and it is very hard to argue anybody out of, because the candidate is in the room and the voters are not.

The question that produces wins is narrower and less comfortable. What are people here already angry about, already worried about, already arguing about at work? You are not creating demand. You are attaching a candidate to demand that is already sitting there.

In 2026 the recurring answers were prices, corruption and the war in Iran. PBS tied the result to all three. After the votes came in, Vice President JD Vance said the war was the obvious reason Republicans lost, and the President said that analysis was largely true while pointing at a late economic message that had gone soft. Set aside who is right. Notice the shape of the claim. Everyone involved is describing demand that existed in the electorate before either side wrote a word of copy.

Timing beats volume

A finding in February is worth more than the same finding in October. Not slightly more. Categorically more, because in February you can still act on it.

In February you can change who you recruit, which seats you contest, what the candidate spends the spring saying, who you hire, what you build. In October you can change the creative on an ad and adjust a buy. Identical information, a fraction of the value.

What a finding can still change, by the month it arrives

Eight campaign decisions, and whether each one is still open when the research lands.

This is the most reliable way money gets wasted in American politics. Research gets bought late, when the budget exists and the race feels real, and by then it can only confirm or deny decisions that were made months ago by people working off instinct.

2026 was the most expensive midterm cycle on record. The seats that turned did not turn because somebody found another million dollars in the last fortnight. They turned on choices made eighteen months earlier about where to compete and who to run.

What Upset Research does for a campaign

Four services. Here is what each one is for and when it is worth buying.

The window each of the four services is for

Months before election day, running from twenty four months out on the left to the final week on the right.

Modeling: who to talk to

Modeling answers the first question. Out of everyone on the file in your district, who is worth your time, and in what order? We build a model of the electorate you will actually face, estimate who is open to persuasion, who needs a reason to turn out and who you should leave alone, and hand back a ranked list that your field program and your media buyer can both work from. One list, one set of assumptions, so the two halves of the campaign stop contradicting each other.

Insights: what they actually think

Insights is how we find out what people in your race are thinking, in their own words, at a scale you can act on. Which two or three issues are live in this particular place. What language moves people and what language makes them flinch. Where the candidate's biography helps and where it costs. We do not call this a poll or a survey, because it is not a probability sample and describing it as one would be dishonest. It tells you what is in people's heads, which is the input you are missing, and it tells you how confident you should be.

Strategy: what the campaign does on Monday

Strategy turns the first two into something a manager can execute at nine o'clock on Monday morning. Which audiences, which message to each, which weeks, which channels, the real split between persuasion and turnout, and the specific things that have to be true for the plan to be working. Written down and dated, so that in six weeks you can tell whether it worked instead of arguing about whether it worked.

Bespoke Tech Tools: the thing your race needs that nobody sells

Every race has one problem that no product solves. A county that publishes returns as scanned images. A volunteer operation that needs a tracker nobody makes. Forty jurisdictions of local results that have to be collected and compared before you can tell whether your region is moving. We build that thing for your race, and you keep it. Bespoke Tech Tools exists because the interesting problems in a campaign are usually the ones that are specific to it.

We help candidates win races they normally shouldn't. That means the work is worth most before a race has been rated, in the window where nobody is paying attention and the decisions are still cheap. We work on every race on the American ballot, from school board to president.

What the 2026 midterm just taught us

Democrats won the House and the Senate. Republicans went into election day holding the House 219 to 214 with two vacancies and holding 53 Senate seats. Seven Republican held Senate seats changed hands: Kansas, Iowa, Texas, Ohio, Maine, North Carolina and South Carolina. It is the first time in twenty years that a Republican president has lost both chambers. Three lessons are usable immediately.

The boundary of the real map sits further out than either list said

Look at the states. Kansas had not elected a Democratic senator in almost a century. Texas had not since 1988. South Carolina had not since 1998. None of the three was on a realistic target list two years ago. All three went.

That is a measurement failure before it is anything else. If your ratings said a seat was safe and it was not, every decision built on that rating was wrong, including the ones that looked sensible when they were made. The fix sits upstream of the advertising. It is measuring a wider set of seats, early and cheaply, so the list you spend against resembles the map you are actually on.

A wave has to land on somebody

Seven flips required seven people who had already filed, cleared a primary and built enough of an operation to receive a national shift. Roy Cooper, a former governor, beat former RNC chair Michael Whatley in North Carolina. Sherrod Brown took Ohio from Jon Husted. James Talarico beat Attorney General Ken Paxton in Texas and became the first Democrat to win statewide office there since 1994.

Filing deadlines sit months before a general election and primaries sit months before that, so serious recruitment happens twelve to twenty four months out. Whatever moved the environment in 2026, the capacity to convert it into seats was built in 2024 and 2025 by people who had no idea what this November would look like. That is the part a party controls entirely on its own, and it is the part that gets cut first when budgets are tight.

Money bought reach and reach was not the constraint

Elon Musk spent at least $474 million on the cycle, including $224 million in Texas, where his preferred candidates lost. In Georgia, Rick Jackson put more than $200 million of his own money into a race he lost to Keisha Lance Bottoms. The fifty largest donors gave $1.7 billion to Republican aligned groups against $832 million to Democratic ones, and both chambers went the other way.

The two largest individual checkbooks and the margins in the races they backed

Both of the cycle's biggest personal spenders were aimed at races their side lost.

Money raised from the 50 largest donors, and the chambers each side won

The 50 largest donors put about twice as much into Republican aligned groups. Both chambers went the other way.

Money multiplies a reason to vote for somebody. It does not supply one. A campaign with the reason and half the budget beats a campaign with the budget and no reason, and 2026 is now a very large data point in favor of that proposition.

Why this decides 2028

The presidential cycle has already started in the only sense that matters to an operative. Target lists for 2028 get drafted over the next few months. Recruitment runs through 2027. Which seats get treated as real, which get written off and which get a candidate is being decided right now, off a reading of 2026 that most people will finish long before the detailed data exists.

For the party that won. A majority that contains Kansas and South Carolina is a majority with an exposure problem. Those seats were taken once, in a midterm electorate, against an opponent who will spend two years changing the conditions. A presidential electorate is different in size and in composition, and nothing about 2026 tells you how these particular seats behave inside one. Chuck Schumer is set to become majority leader with members sitting in territory his party has not held in decades, and holding a seat is a different job from winning it. The work now is separating the seats that are structurally competitive going forward from the ones that were a single cycle, and being honest internally about which is which. Treating all of them as permanent is the standard route from a wide majority to a narrow one.

For the party that lost. The argument about why is already running, and it is running on explanations that imply completely different budgets. Vance blamed the war. The President pointed at the late economic message. Both can be partly true and neither tells you where to put next year's money. The version you can act on is narrower. Pull out the target and defense lists you were working from in September, before anybody edits them, and set them against the seats that turned out to be live. Every seat that came off the board without ever appearing on the list is evidence about the forecasting inputs, and that is a cheaper problem to fix than a persuasion problem. Then restart recruitment, because the people who run in 2028 are identified in 2027, and the House leadership question will take up most of the oxygen in the meantime.

For both. The marginal research dollar is now worth more in a seat nobody has rated than in a seat everybody is watching. That was arguable in 2024. After Kansas, Texas and South Carolina it is close to settled. Whichever side finds the edge of the real map first gets to choose the ground in 2028, and the other side spends the cycle reacting.

One more thing, and it applies to every race on the ballot, not just the federal ones. Thousands of contested county, municipal and school board elections will run between now and 2028. The results are public and almost nobody collects them. They are the earliest available read on whether a region is moving, and they arrive in the window when recruitment decisions are still open.