The Three Things That Break During the Final Campaign Sprint (And How to Prevent Them)

Every campaign AI operation that's running correctly in June has the same failure modes under sprint conditions. The technology works. The configuration is right. And then October arrives and things start breaking.

Not because the AI fails. Because the human systems around the AI fail under pressure.

Three things break consistently. None of them require a technology fix. All of them require process decisions made before the sprint, not during it.

1. The Approval Workflow Gets Abandoned

The most common failure mode in deployed AI campaigns: the approval step gets skipped because things are moving too fast.

It usually starts with a reasonable workaround. The candidate is in a two-day travel stretch. The finance director is running on four hours of sleep. A fundraising email is ready to go, and it looks fine, and the approval review that's supposed to happen before it sends just... doesn't. Not this one time.

Then it happens again. And again. And the "required approval" workflow becomes an optional step that runs when someone has bandwidth, which during the sprint is never.

The result is inconsistency in what goes out. Sometimes the emails are reviewed and sharp. Sometimes they're close-enough drafts that went out because the queue was backing up. Donors who are paying attention notice. Staff can't be sure what's sending. The quality control that made the AI operation trustworthy is gone.

How to prevent it:

Lock the approval path before the sprint. Decide now: who has authority to approve standard donor follow-up emails without the candidate's review? Who approves the fundraising email when the finance director is unavailable? What's the maximum time an approved email can sit in the queue before it sends without a second look?

These are process decisions. Make them in August, document them, and make sure the people in the approval chain know what they are. An approval workflow with a clear decision tree runs under pressure. One that depends on whoever has time does not.

2. The Voice Model Drifts Without Anyone Noticing

Voice model drift is slow, which is why it's dangerous. The candidate's communication style evolves through a campaign. The stump speech in October sounds different from the stump speech in March. The donor conversation has a different register in the final sprint than it did in Q2.

The voice model doesn't update itself. If it hasn't been recalibrated since spring, it's generating emails that sound like the candidate did six months ago. The drift is subtle enough that the people reviewing drafts every day don't notice it. But donors who've been in the relationship for the full cycle sometimes do.

The practical problem: draft quality drops. The emails that used to require light editing start requiring heavier editing. The approval cycle slows because more edits are needed. The finance director is spending 30 minutes on a batch that should take 10. Under sprint pressure, the response is usually to reduce the quality bar, not to find the root cause.

How to prevent it:

Schedule a voice model review before August ends. Pull 10 recent AI drafts and have someone who hasn't been reviewing them daily read them cold, ideally the candidate or a senior staff member who knows how the candidate is currently communicating. If the drafts sound off, recalibrate before the sprint fully accelerates.

This takes two to three days. It's worth it before October, not during it.

3. Email Frequency Drops and Doesn't Come Back

Campaign email programs are the first thing to slow down when staff is stretched and the first thing that should maintain frequency under pressure.

The pattern: a busy stretch in September creates a two-week gap in the email program. The gap gets filled with an urgent fundraising appeal. Then another gap. By mid-October, what was a 2x-per-week program is running at once every 10 days, with spikes around FEC deadlines. The list relationship that was built through consistent contact in spring is degrading.

Email frequency is a list relationship variable. A list that's been receiving consistent, on-schedule emails maintains warmth and engagement. A list with erratic frequency (silence, then burst, then silence again) has lower response rates and higher unsubscribes. The impact is cumulative and shows up in October fundraising results that are lower than they should be.

How to prevent it:

The email program needs to be the one thing that doesn't get adjusted when staff is stretched. The mechanism for this is getting the drafts further ahead: having next week's email drafted and in the approval queue this week, so the approval step isn't a bottleneck when things get busy.

This requires the AI drafting pipeline to be running on a consistent weekly input schedule, not a reactive one. If the content feed is generating draft topics from the news cycle, the drafts can be generated before they're urgently needed. The approval queue should always have 3-5 emails ready, not zero.

Why These Break Under Pressure (And Not Before)

These failure modes don't show up in June because June isn't a sprint. There's bandwidth for the approval review. There's time to notice voice drift. The email program runs because no one is juggling seven urgent things.

The sprint removes all of that. Every day has competing priorities. Every week is busier than the last. The systems that run on process discipline survive it. The systems that run on bandwidth do not.

The campaigns that make it to Election Day with their AI operations running correctly are the ones that made the process decisions in August. Locked the approval path. Scheduled the voice model check. Protected the email cadence. None of this is complicated. All of it has to happen before October.

Eric Linder is a former California State Assemblyman (2012-2016) and founder of AutomatedTeams, an AI operations consultancy for political campaigns and advocacy organizations.