← Research library

2027 planning framework: mapping your marketing org to agents and humans

AI-assisted research. This note was prepared with the assistance of AI tools. It may contain errors and is not professional advice. See our Legal Disclosures.

Annual planning usually centers on goals, budgets, and campaigns. As agents take on a larger share of marketing work, a new question belongs in the plan: for each major body of work, what will people do, and what will agents do?

Without an explicit answer, plans assume capacity that does not exist, or ignore capacity that does. A plan that counts a content agent as free output will miss the reviewer time it needs. A plan that ignores the agent entirely may hire for work that is already covered. This post sets out a five-step framework for mapping your marketing organization's work to people and agents for 2027.

Step 1: List the work, not the roles

Start with the major recurring work your team will do next year, grouped by function: demand generation, content, lifecycle, product marketing, operations, analytics, and events. Break each function into concrete activities, such as account research, campaign planning, first drafts, review, reporting, and optimization.

Listing work rather than roles avoids assumptions about who does what. A role is a bundle of activities that made sense for a human team, and agents rarely take over a whole bundle. They take over pieces of it, and planning at the activity level shows which pieces move and which stay.

Aim for a list the team can actually discuss. We recommend 20 to 40 activities for a first pass. Fewer than that and the categories hide real differences, while more than that turns the exercise into an inventory nobody finishes.

Step 2: Classify each activity

For each activity, decide the target division of labor for 2027. We use four levels, which map loosely to the autonomy ladder.

  • Human-led: people do the work, and agents may assist with research or drafts.
  • Agent-assisted: agents produce first versions, and people review and finish.
  • Agent-led with oversight: agents do the work within guardrails, and people review samples and exceptions.
  • Agent-run: agents handle routine cases autonomously, and people manage the system.

Base classifications on evidence from this year. An activity that has run in shadow mode or as live assistance, with tracked quality, has earned a move up a level. Activities without evidence should not jump straight to agent-led or agent-run, however promising the demo looked.

Record the evidence next to each classification. One line is enough, such as "edit rate tracked since July, stable" or, just as usefully, "no evidence yet, pilot planned for Q1". That line is what makes the plan defensible when someone asks why an activity sits where it does.

Step 3: Estimate capacity shifts

For activities moving toward agents, estimate how much human time shifts and where it goes. Freed time should be reinvested deliberately, or it disappears into whatever is loudest that week. We recommend four destinations for it.

  • Strategy and creative work: the judgment-heavy work agents do not do well
  • Review and quality assurance: work that grows with agent output
  • Agent operations: monitoring, improving, and governing agents
  • Relationships: time with customers, sales, and other stakeholders

Be realistic about the first year. Review and operations work can absorb much of the time saved, especially while a team is still learning what to check.

Consider a hypothetical lifecycle team that spends 30 hours a week writing and building email variants. If an agent-assisted model halves the drafting time, 15 hours shift. A plan that assumes all 15 hours return as new capacity is optimistic. A more careful plan might assign six of those hours to review, three to agent operations, and the remaining six to new work, then revisit the split after a quarter of evidence.

Step 4: Identify new roles and skills

Look for work that has no clear owner today. Agent operations, evaluation, knowledge base maintenance, and data contracts are typical examples. These tasks tend to land on whoever set up the first agent, which works for a pilot and fails for a program.

Decide whether each becomes a new role, part of an existing role, or a shared responsibility with a named lead. Then plan training.

Reviewing agent output well is a skill, and so is writing instructions an agent can follow. Neither comes automatically, so plan time to build both through practice on real work, with feedback from someone more experienced. Our post on budgeting for 2027 covers how to fund this work alongside model and tool spend.

Step 5: Sequence the changes

Not everything should change in January. Sequence shifts across the year, avoid your busiest periods, and attach an evidence checkpoint to each move. An activity moves from agent-assisted to agent-led only after it meets defined criteria.

Good criteria are specific and measurable. For example, you might require that edit rate stay below a threshold you set for eight consecutive weeks. You might also require no high-severity incidents in that window, and the agreement of the activity's owner. Write the criteria before the change starts, so a single good week does not make the decision.

Sequencing also limits risk. If three activities shift in the same month and quality drops, it is hard to tell which shift caused it. We suggest one significant shift per function per quarter as a default for a first year.

Communicate clearly

Changes to who does what affect people's jobs. Share the plan openly, explain the reasoning, and involve the people doing the work in designing the new arrangements. They know which activities are routine and which only look routine from outside.

Be direct about what is uncertain. If you do not yet know how a role will change, say so, and say when you expect to know. We expect teams that design changes with their practitioners to adopt agents with fewer problems, because the people closest to the work spot failure modes that planners miss.

What we are watching

The year-end report is the natural input to this exercise. Its evidence on quality, cost, and risk should decide which activities move up a level, and our year-end measurement plan describes how to gather it.

The open question we are studying is how the mapping changes once agents coordinate with each other rather than with a single person. A plan built activity by activity may undercount the work that sits between activities, such as handoffs and reconciliation. We expect that seam to become the next planning problem.

How to apply this: list your team's 20 to 40 most significant recurring activities and classify each as human-led, agent-assisted, agent-led with oversight, or agent-run for 2027, based on this year's evidence. Note where freed time will go and which checkpoint each shift must pass.

Related research

Put this research to work

Marketix turns these patterns into software: agents, automated workflows, and a hub that keeps people in control. Sign up for early access.