You run a campaign in a general election. You have 10 field teams to deploy across three tactics. Where do you put them to win the most votes? Spend all ten, then run the campaign.
Persuading the undecided middle barely moved anything. Across 49 real field experiments, the average effect of campaign contact on vote choice in a general election is close to zero (Kalla and Broockman, 2018). The persuadable middle is far thinner than campaigns assume.
Turning out your base was your biggest lever, up to a point: social pressure ("your neighbours will know if you voted") is the largest reliable get-out-the-vote effect on record (Gerber, Green and Larimer, 2008), but your base is finite, so it runs out. Deep canvassing genuinely works, long, listening-led conversations durably shift attitudes (Broockman and Kalla, 2016), but it is slow and expensive, so it cannot scale to carry a whole campaign.
The lesson: separate turnout from choice, and treat any "big persuadable middle" as a claim to prove, not assume. Most of what gets sold as persuasion is really turning out people already on your side.