AI Marketing Planning: How Our Campaigns Become Tracked Tasks

AI marketing planning works when an agent does the part people keep skipping: noticing that a brand’s task queue is running low, reading what is selling right now, and drafting the next batch of tasks against the campaigns already live. A person reviews every batch before anything moves. That is the system we run across our own brands as of September 2026, and this post walks through how it decides, where it refuses, and what it still cannot do.

We built it because our planning was a mess we could see clearly and still could not fix by trying harder. Several brands, one small team, and a content calendar that got filled in bursts whenever someone remembered. By the time a task was written, it often had nothing to do with what was selling that week.

Why most marketing AI never reaches the plan

Adoption numbers for marketing AI are high, and the planning layer underneath them is mostly unchanged. Tools that draft copy faster do nothing about a calendar nobody updates, campaigns nobody links, or tasks written from last month’s assumptions. The distance between using AI and planning with it is where most marketing teams sit in 2026.

Salesforce’s 10th State of Marketing, published 19 February 2026 from a double-anonymous survey of 4,450 marketing decision-makers, found 75% of marketing organizations have adopted AI. The same survey found 84% acknowledge running generic, one-way campaigns, and 98% hit personalization barriers, mostly from data quality. It is vendor research and self-reported, so read it as direction rather than measurement.

The pressure underneath is people. The CMO Survey’s 35th edition (Duke Fuqua, Deloitte and the American Marketing Association, published 31 March 2026, 308 US marketing leaders surveyed in January 2026) reports that marketing headcount growth slowed by more than half from the prior year’s rate, while AI use in marketing more than doubled in two years. Fewer people are being asked to plan more, with tools that write faster and plan nothing.

Marketing is also where companies point AI first. Eurostat’s 2025 survey of ICT use in enterprises, published December 2025 from a sample of roughly 157,000 EU businesses, found 34.70% of enterprises using AI apply it to marketing or sales, the most common use it records.

What the planning agent reads before it drafts anything

It reads four things: the brand’s current sales and stock picture, the open ideas backlog, the brand’s voice rules, and the dates already committed. Each pass pairs a proven seller with a slow mover, so the plan pulls on demand that already exists and pushes the products that need help. Nothing is drafted from a blank template.

What starts a pass is a shortage, and the calendar has no say in it. The agent checks each brand’s queue of scheduled tasks. A brand with a healthy queue gets skipped. A brand running low gets a fresh planning pass. It works like reordering stock before the shelf is empty, and it puts planning effort where the gap is instead of spreading it evenly across brands that do not need it.

How a task gets linked to a campaign without faking one

A person decides what a campaign is, when it runs and what it is for. The agent links a new task to an active campaign only when the match is exact. If nothing matches, the task stands alone. It is not allowed to invent a campaign so that a task has somewhere to sit.

The queue row that seeded this post promised “campaigns broken into tasks”. That overstates it, and we would rather say so here than let a headline do work the system cannot back up. Campaign design is still human work in our shop.

The same rule covers categories. Every task must land in a pre-approved category and format. When a topic fits none of them, the agent flags the gap for a person instead of stretching the nearest label. A forced fit is harder to catch later than an honest flag today.

Anything touching price or promotion saves as a draft that needs a human sign-off before it goes anywhere. The agent recommends. A person decides.

Calendar-driven planning versus signal-triggered planning

Calendar-driven planning runs on dates and memory, so it drifts away from what is selling. Signal-triggered planning runs when a brand’s queue runs low and drafts from current sales and stock. The table is the before and after in our own marketing, row by row, as of September 2026.

	Calendar-driven (before)	Signal-triggered (now)
When planning runs	Fixed dates, or whenever someone remembers	When a brand’s own queue of tasks runs low
What drafts come from	Whatever was top of mind	Current sales and stock, a proven seller paired with a slow mover
Campaign linking	Manual and inconsistent	Exact match to a live campaign, never invented
Categories	Loose, stretched to fit	Fixed list, mismatches flagged for a person
Review	Whoever had time	Automated gate first, then a person reviews every batch
Who marks it done	A person	A person, still

Every draft passes a gate, and every rejection carries a reason

Before a drafted task becomes usable, it is checked for duplicates, safety and brand-rule fit. A rejected task is neither dropped quietly nor pushed through. It comes back with a stated reason, so the person reviewing the batch can see what the agent tried and why the gate stopped it.

What passes becomes a task card with a status, an assignee, a target date and a format, flagged as AI-generated. Cards always land in a draft or planning state. The agent cannot mark a task in progress or done, and it never edits a card it did not create in that same run.

People review the batch. People move cards to done.

What happens when a data feed goes quiet

If a sales or stock feed is unreachable, the run report says so and the agent falls back to a slower direct pull. It never reads a missing feed as a sign that nothing is selling, which is the failure that would quietly wreck a plan while looking perfectly normal.

Three more guards sit around it. A separate watchdog alerts if the runway check itself is ever missed, so a brand’s queue cannot run dry without someone hearing about it. A backlog guard pauses new drafts once too many sit unreviewed, because a pile of unread drafts is worse than a short queue. And data staleness is printed at the top of every run report, the first thing the reviewer reads.

Where this stops, on purpose

We are not describing a system that plans marketing end to end. Two boundaries are deliberate. Campaigns are a human decision from start to finish. And “done” is a status a person sets by moving a card, never one the agent sets for itself. Both cost us automation we could have built. We kept them because an agent that can invent campaigns or close its own work is much harder to trust when nobody is watching.

The task board those cards land on, with its queue, approvals and performance view, is a separate system and gets its own post. This one stops at the moment a grounded task reaches the board.

What changed is qualitative, and we are not publishing figures for it. Planning moved from dates and memory to a continuous process triggered by a real shortage. Content choices come from sales and stock data instead of whoever had a free hour. Every planning decision leaves a run report someone can audit.

That lines up with the split in the Asana Work Innovation Lab’s State of AI at Work 2025 report, which surveyed 9,236 knowledge workers between February and August 2025: organizations divide roughly evenly between those bolting AI onto broken workflows and those redesigning the workflow around it. We did not want AI drafting bolted onto a calendar that was already broken.

What the equivalent build costs

This is the shape of work our Fractional AI Officer retainer covers: $2,900 a month, cancel monthly. It includes one new automation shipped to production every month, monitoring of everything already live, and a weekly written brief. The comparison points on that page are a full-time AI lead at $180k to $250k plus ramp time, or market retainers at $1k to $10k a month in 2026. A planning agent like this one fits inside a single monthly build: one agent, one defined job, scope written down before work starts.

For a marketing system of ours that runs daily in production, see the inbound machine proof page: a 363-product catalogue with daily automated publishing, as of July 2026.

If you run an agency and want to see where builds like this fit your own service menu, read AI services for marketing agencies. If you are earlier than that and need an order of operations, start with an AI roadmap for small business.

If your marketing plan lives in one person’s head and a spreadsheet that only opens under pressure, write down what is breaking. Start async: send a short written brief, get a written reply within one business day, and never sit through a meeting.