What two and a half years of replies taught us about vote by mail

Christine Fano ran our Montville program with her personal cell number in every message she sent. Not a shortcode. Her actual phone.

That phone told us things no survey could. People wrote back that they never got a ballot. People wrote back that they'd mailed it two weeks ago and wanted to know why we were still asking. People wrote back confused about a deadline that had been printed on the envelope the whole time. Christine answered every one of them, hourly, and we changed the next send based on what came in.

If you want to know why that program won a Reed Award, it isn't the sequencing. It's that somebody was on the other end of the phone.

What the inbox taught us

Two things, and they've shaped how we build every vote-by-mail program since.

The first is that the voter who decides your race usually doesn't have an opinion yet. She isn't hostile and she isn't yours. She's undecided, and she's going to stop being undecided in the next four weeks whether or not you participate in that. The campaigns that treat the undecided bucket as a problem to be surveyed rather than a conversation to be in are the ones reading concession speeches.

The second is that a ballot sitting on a kitchen counter is not a persuasion problem. It's a logistics problem wearing a persuasion costume. Somewhere in Montville there was always a voter who agreed with us completely, intended to vote, and had a ballot in a drawer under a stack of mail. No message about the candidate was going to move that ballot. A message about the drawer would.

Those two ideas sound like they're in tension. They aren't. They're the two halves of a program, and almost every campaign builds one and skips the other.

The research caught up

Cross Screen Media published a persuasion texting test in September that put numbers on the first half.

The finding that matters is in their ballot position chart, credited to Battleground Impact and Public Opinion Strategies. In the market that received texts, the undecided share fell from 27% to 13%, and those voters broke to the Republican, 40% to 55%, with the Democrat essentially flat. In the control market that got no texts, the undecided share fell a nearly identical amount, 34% to 21%, and broke the other direction entirely, Democrat 27% to 41%.

Same collapse. Opposite direction. The texted market gained 16 points of margin while the control market lost 15.

Worth saying plainly that the two markets didn't start identical, and one paired-market test is not a body of literature. But a 31-point spread is not noise, and it's the cleanest published confirmation I've seen of something we'd already concluded from watching a phone: undecided voters aren't waiting to be persuaded so much as waiting to hear from somebody. Whoever shows up gets them.

Most of the industry is resharing the ad recall number from that study. Recall is the vanity metric. The ballot chart is the invoice.

What the study doesn't cover

It stops at preference, which is where survey research has to stop.

We get paid for the part after that. A voter you moved 15 points who never returns her ballot counts exactly the same on election night as a voter who never heard your name. That gap is the second half of the program, and it's the half that produced our numbers.

Montville

When we started, Montville had fewer than 380 active Republican vote-by-mail voters, and most were dormant. They'd requested a ballot at some point and then stopped returning them. Deadline confusion was routine, and blast messaging had already been tried without much to show for it.

We built more than ten distinct messages, each tied to a stage: requesting a ballot, waiting on delivery, sitting on one that arrived, running out of time to return it. A voter who got her ballot yesterday received a different message from a voter who'd been holding one for nine days. That sounds obvious written down. Very few programs do it, because it requires knowing where every voter sits on any given morning, and that means live data rather than a list you bought in August.

Messages went out in the voice of actual candidates asking directly rather than in the voice of a committee. And the program ran across multiple cycles over two and a half years, which meant the list learned something every time instead of starting cold each spring.

The final numbers: 1,043 ballots submitted, a 76% return rate against a 73% Morris County Republican average, and roughly 175% growth from where we started.

The comparison nobody puts in a case study

Here's the part most firms would cut, and cutting it is exactly why nobody believes case studies anymore.

Township	Return rate	Ballots returned
Hanover	80%	508
Jefferson	78%	458
Montville	76%	1,043
Chatham	71%	429

We came third of four on return rate. Hanover beat us by four points and earned every bit of it.

Now look at the second column. Montville returned more than twice the volume of the next closest township, and did it while building the universe up from fewer than 380 active voters.

Return rate is the flattering number, and it's the easy one, because rate climbs when your universe is small and already engaged. Shrink your list to your hundred most reliable voters and you can post a 90% return rate that means nothing. Volume and growth are the hard numbers. Nobody has ever won a seat on a percentage.

If a vendor shows you a rate without a denominator, ask for the denominator.

The same idea at statewide scale

The governor's primary ran under worse conditions: a low-turnout summer primary, national political fatigue, very little paid media supporting vote-by-mail promotion, and a ballot universe whose status changed daily.

Working with Grassroots Targeting and RumbleUp, we built it around a daily data pipeline. Ballot return data came in every morning, which let us re-cut the audience every morning. Somebody who returned a ballot Tuesday morning was out of the chase Tuesday afternoon, so nobody got nagged about a ballot they'd already mailed. That sounds like a courtesy. It's actually a budget decision, because every message sent to a voter who already returned is money that bought nothing and irritation that cost something.

Jack finished the primary with 316,283 votes, 67.8% of the total, up from 49.46% in 2021. Grassroots modeled 462,363 total votes; the actual came in at 468,888, a miss of under 1.5%.

He lost the general in November. I'm not going to pretend otherwise or bury it in a footnote. It doesn't change my read on the turnout program, which did what it was built to do in an environment designed to work against it. Those are two separate questions and this industry should be more willing to say so out loud.

What carried across both

One is a township with a thousand ballots. The other is a statewide primary approaching half a million votes. Four things were true in both.

The data was live. Both programs changed who got what based on what happened the previous day. A ballot universe built in September is already wrong by the second week of October.

Every message had a job. Ten messages that each do one thing will beat twenty that all say "vote." Every send existed for a reason, and every send stopped going to people it no longer applied to.

Somebody answered. This is the one almost everyone skips, because it doesn't scale neatly and it isn't fun to staff. When a voter texts back to say her ballot never arrived, your whole program comes down to whether a person replies.

And both ran long enough to compound. Montville took two and a half years to look like an overnight result. A program you start in October has no history, and a list that's never been worked doesn't know anything about the people on it.

That last one is the uncomfortable one, because it's the one you can't buy in the final month.

Thirty-three days

Early voting is opening across the country through October, and New Jersey's in-person early voting runs October 24 through November 1. Every day your chase isn't running is another day ballots sit unopened under stacks of mail.

The research says your undecideds are going to decide soon, and that whoever is talking to them gets to influence which way. Our inbox says the ones who already decided still need help getting the envelope back. Both things are true at once, and a program that only does one of them is going to lose to a program that does both.

If your texting program right now is a handful of scheduled sends and some optimism, we should talk this week.

And if you're reading this and it isn't your campaign that needs it, it's probably somebody you know. Send them this. If you refer a campaign to us and they sign, we pay referral fees, or we'll make a contribution to a committee you name instead, whichever is cleaner on your end.

Get your program built: campaignengine.io/contact-us