Big lists, wrong answers: why 100 good picks beat 1,000 invitations
In the autumn of 1936, The Literary Digest ran the biggest poll America had ever seen.
On this page5 sections
The magazine mailed ballots to 10 million people and asked who they’d vote for in November: President Franklin Roosevelt or his Republican challenger, Alf Landon. Its staff addressed more than a quarter of a million envelopes by hand every day. The names came from phone books, club rosters, city directories and lists of car owners.
About 2.4 million ballots came back. On 31 October, the Digest printed the final count: Landon 1,293,669, Roosevelt 972,897. The figures, it told readers, were “neither weighted, adjusted nor interpreted.” It also reminded them of its record: “So far, we have been right in every Poll.”
Three days later, Roosevelt won more than 60% of the vote and every state except Maine and Vermont. He took 523 electoral votes, and Landon took 8.
A young pollster named George Gallup had called the winner with a sample of about 50,000 people, roughly 1 for every 48 ballots the Digest counted. His number wasn’t perfect either, since he had Roosevelt on about 56%. But he had the right man, and the Digest didn’t.
Within 2 years the magazine was gone, and Time took over its subscribers.
The list decided the answer
People who study polls still argue about exactly what went wrong. Gallup blamed the list. In the middle of the Depression, people with phones and cars were better off than most voters, and better-off voters leaned toward Landon.
Later researchers found a second problem. Only about 1 in 4 people sent a ballot back, and the people who bothered weren’t like the people who didn’t.
Both explanations say the same thing. The Digest’s problem was never how many people it asked.
It was who.
Here’s the part that stings. The Digest’s ballot also asked how each person had voted in 1932.
In 2017, the statisticians Sharon Lohr and Michael Brick went back to the returns. Weighting the answers by that one question, they found, would have put Roosevelt ahead in the Electoral College.
The magazine had the fix in its own mailbag. It chose to print the raw count.
That’s hard to see from the inside, because the size felt like the point. A count of 2.4 million answers is something you can print on a cover, while Gallup’s 50,000 sounded like a rounding error. The bigger number felt like proof, and it was, but only of how many stamps the Digest could afford.
A big list doesn’t cancel out a bad pick. It repeats it.
A directory is not a list of buyers
Forty-two years later, someone made the same mistake with a new kind of mail.
In the first week of May 1978, a marketer at Digital Equipment Corporation named Gary Thuerk wanted people to come to product demos in Los Angeles and San Mateo. He had every West Coast address in the printed ARPANET directory typed in, added some customers elsewhere, and sent one message to all of them.
The mail program could only hold 320 addresses in the header. The rest spilled into the body, so people opened it to find a wall of names before they reached the pitch.
The complaints came quickly. The major in charge of the network at the Defense Communications Agency called it “a flagrant violation” of what the network was for, and other users piled in. Today it’s often called the first spam.
Both stories start in the same place: a directory. It was a phone book in 1936 and a network directory in 1978.
A directory tells you who can be reached. It doesn’t tell you who wants to hear from you.
The same mistake, sent from your profile
Founders who still sell face the same choice on LinkedIn every week, and LinkedIn’s search is a bigger directory than anyone in either story could have pictured.
The volume plan goes like this. You pull 1,000 people from a search filter: say, every head of sales at a software company with 11 to 50 staff. You write one note that fits all of them, send it and count the replies.
The judgment plan is slower. You start from one customer you’d like 10 more of. Then you look for people who have that customer’s problem right now and have shown it in public, in a post, a job ad or a funding round.
You invite 100 of them in a week, each for a reason you could write in one sentence. Anyone you can’t explain, you skip.
The volume plan looks better on a dashboard, and a published average shows why that’s misleading. Belkins studied 15.1 million outreach touchpoints and found that 18.7% of invited prospects accepted. Run 1,000 invitations through that average and more than 800 are never accepted.
Those 800 aren’t a blank, though. Each one got an invitation with your name and face on it, and a line written for nobody in particular. None of them said yes.
LinkedIn keeps count too. It lists invitations that are ignored or marked as spam among the reasons it restricts an account’s invitations. So the unanswered ones aren’t free; they’re charged to the profile that sent them.
The Digest’s worst moment wasn’t the wrong answer. It was trusting the answer because of how many ballots stood behind it.
Volume does the same thing to a founder. A thousand sends and a small reply rate feel like data, and they are, but the data describes your list, not your market.
Why picking feels worse and works better
Picking is uncomfortable because you can’t watch it happen. A sending dashboard climbs all afternoon, but nobody sees the moment a person looks at a name and decides not to invite them.
The work that decides the result is the work nobody counts.
That’s why the number I’d watch first isn’t how many you sent. It’s how many you skipped, and why. If 40 of 100 people fall out for the same reason, that reason belongs in the search, and next week’s list starts better.
At GTME, a person decides who gets each invitation before anything is sent. On every plan, your operator, a GTM engineer on our team, picks who to contact, and from our Signal plan up the pick can start from a buying signal.
We cap invitations at 100 a week per profile. LinkedIn doesn’t publish a weekly number, so the cap is our choice, and at that pace the slow part of the week is finding people worth inviting. That’s how it should be.
A cloud-based sending tool with daily limits does the clicking, and careful picking doesn’t change what LinkedIn allows. Its User Agreement doesn’t permit tools like that, and we say so plainly in how we run your account.
The signals we look for when we pick are in the LinkedIn buying signals we act on, and the ones we ignore. The whole method, from the first pick to the booked meeting, is in LinkedIn outreach for founders, start to finish.
A smaller list you can defend
Volume has its place. If the market is huge and each sale is small, a broad list can pay for itself. GTME is built for founders in the other market: a few thousand possible buyers, in a market small enough for people to know each other.
None of this makes 100 a special number. If your whole market is 300 people, the right week might be 20 invitations. The number should come out of the picking, not the other way round.
You probably hold the Digest’s missing question already. You know which customers bought from you, and why. That’s the weight to put on every name before it goes on the list.
There’s a simple test for any size. Could you say, out loud, why each person on your list is on it?
If the honest answer for most of them is “they matched the filter,” you’re running the Digest’s poll. You’ll get a big number back, and it will be about your list.
The Digest counted ballots, and Gallup chose people. Four years later, only one of them was still taking polls.
Sources
All checked on 28 September 2026.
- 01The Literary Digest, final poll returns, 31 October 1936.
- 02Lohr and Brick, Roosevelt Predicted to Win, Statistics, Politics and Policy, 2017.
- 03Wikipedia, The Literary Digest.
- 04ProQuest, the 1936 Literary Digest poll, 7 September 2016.
- 05Brad Templeton, the DEC message of 1978 and the replies to it.
- 06Belkins, LinkedIn outreach study, updated 29 June 2026.
- 07LinkedIn Help, Types of restrictions for sending invitations.
- 08LinkedIn, User Agreement, section 8.2.