An exit poll is conducted after a voter exits the voting booth. It's conducted by a private organization (usually either a news organization or someone working in collaboration with a news organization) and polls people to find out how they voted. The exit poll is voluntary.

Organizations can then categorize that info based on age, gender, race, area where they voted, and other details. News organizations can then use that info (along with a bunch of other data, including polls conducted leading up to the day of the election) to extrapolate who will win an election in a given area. Typically, despite being somewhat limited in their scope (not everyone at every polling location nationwide is polled), the exit polls are usually reflective of the actual election polls.

Campaign organizers for the next election can also use the data to help figure out their strategies for the next election. For a general example (I came up with it off the top of my head), "We failed to gain the aged 60+ black male vote in this state. We need to study how to appeal to them better in the next election."

Fun Fact: The actual official votes actually take days to count. So these and other types of election polls really help news organizations predict the results even just a few hours after the election polls close, and they're rarely wrong. Sometimes, they're even able to call an election the minute the polls in that area close*. These news organizations often each crunch their own numbers, too, so they don't necessarily all rely on each other's data.

*I should note that each state has its own rules about how and when they release election results. Often, to avoid influencing voters who haven't voted yet, they won't release results (including results from early voting) until polls in the entire state have closed. This is usually the case with news organizations announcing their predictions, too. That's why some news organizations are able to immediately predict some races as soon as the polls close.