Marketing Got an On-Call Rotation

A broken campaign used to be a Monday problem. Someone spotted the dead landing page on the way into standup, a producer paged the agency, and the fix landed by lunch. The team that caused it wrote a note to themselves and moved on.

That worked when a campaign was something you shipped and then watched. It falls apart when the campaign is a live system: a retargeting pool refreshing every fifteen minutes, a generative answer that reads your product pages overnight, a chatbot quoting last quarter's pricing because nobody updated the source doc.

So marketing teams have started borrowing the furniture from the people next door — SLOs, runbooks, and the kind of blameless incident review engineers have been running for years.

A pager that actually goes off.

Not as metaphor, as practice. The sections below walk through specific cases where that borrowing is already happening, and what each one looks like when a content or campaign team runs it for real.

The Generative Answer Went Stale Overnight

A model-generated answer about your product can go wrong between Friday and Monday without anyone touching the site. A pricing page gets edited, the embedding in a vendor's index doesn't refresh, and the chatbot quotes an old number to a buyer in the demo.

There's nobody to email about that in the traditional sense. The content team owns the source, the SEO team owns the crawl surface, and no one owns the fact that an answer engine is now wrong about you in public.

This is the case that pushes marketing toward an SLO. Pick the handful of answers you care about — pricing, availability, support hours, the product one-liner — and define what "correct" means for each. Measure it on a cadence. When the measurement slips below target, somebody is paged.

The structure is lifted straight from site reliability engineering, and it survives the translation to marketing because the underlying question is the same: what does "working" mean, how will we know, and what do we do when it isn't? An AI digital marketing company that already instruments brand mentions across generative engines is doing SLO work whether they call it that or not.

The Launch Email Went Out With the Wrong Subject Line

Email mistakes are the clearest incident any marketing team runs. Wrong subject line, wrong segment, or a merge tag that rendered as a literal {{first_name}} in two million inboxes.

The clock starts the moment someone notices, and the response is almost entirely procedural: can we stop the send, can we send a correction, who tells support, who tells legal, what do we post.

This is where the runbook earns its keep. A runbook is narrow and tactical — the exact steps to take for one specific kind of failure — and that's what you want when the room is panicking. Write one per common failure mode: bad send, broken checkout link, pricing typo, outage at the ESP.

Each should name the person on-call, the kill switch, and the external comms owner. The person who runs the runbook doesn't need to be the person who wrote the campaign, which is the whole point.

• Trigger. The one-sentence description of the failure this runbook addresses. If the responder can't match the symptom in under thirty seconds, the runbook is too broad.
• Immediate actions. The first three to five steps, in order, that stop the bleeding. Pause the automation before you debate the correction email.
• Comms owners. Who tells support, who tells the account team, who approves anything that goes to customers. Names, not roles.
• Rollback and verify. What "resolved" looks like, and how the on-call confirms it before standing down.

The Review Has to Be Blameless or People Hide the Next One

Every one of the cases above produces an artifact marketing teams historically have not written: the postmortem. Google's SRE practice defines it as a written record of the incident, its impact, the actions taken, the contributing causes, and the follow-ups to prevent recurrence. The discipline that makes it useful is blamelessness — the review focuses on process and technology, and assumes the people involved were acting on the best information they had at the time.

That part is harder for marketing than it sounds. Campaign work is personal in a way that infrastructure work often isn't. Someone wrote the subject line. Someone picked the influencer.

If the review turns into a search for who to fault, the next incident gets quietly cleaned up and never written down, and the team loses the one thing that would have prevented the one after that.

The reviews that change behavior read like engineering docs: timeline, impact, contributing causes, action items with owners and dates. No adjectives. No defendants.

None of this makes marketing into SRE. The point of borrowing is to stop treating live marketing systems as if they were still campaigns that end on a date. The pager is already going off. Deciding who answers it is the work.