Faster AI execution does not automatically improve marketing operations

A marketing team can produce work faster and still leave the business waiting longer. AI agents, systems that carry out repeatable activities within a larger workflow, can speed up research, reporting, and drafting while putting more pressure on the people who review or use their output. For leaders facing higher delivery expectations with leaner teams, delegation therefore requires clear decision context and human accountability.

This challenge shaped “Meet your new teammate: Building marketing workflows around AI agents” at the November 2026 MarTech Conference. In an audience poll, 51% of respondents said their organizations were actively piloting AI, and none reported avoiding it. The results describe adoption among attendees; the panel’s recommendations addressed how individual experiments and enterprise production deployments can become coordinated ways of working.

Give agents the reasons behind past decisions

Coordination starts with preserving the knowledge needed to delegate work well. When employees leave and vacancies remain unfilled, the remaining staff inherit delivery pressure alongside gaps in organizational memory. A task may still have an owner even after the reasons behind earlier choices have left with a colleague. Producing more cannot recover knowledge that nobody has recorded.

Lisa Shaw, founder of Devon Point Group, addresses this problem by recording the reasoning behind past decisions and adding it to her AI instructions. Her records cover brand subtleties, earlier customer friction, and relevant history, giving agents context they cannot reconstruct from a task request. As the founder of an advisory business, Shaw has a commercial interest in demand for guidance on how organizations work. Her approach makes that knowledge available during execution and review.

Brand language shows how this context helps. An instruction to avoid a particular phrase sets a boundary, while the reason behind it helps explain which other language could cause the same problem. A previous customer disagreement may likewise change what is appropriate in a routine message. Shaw builds these lessons into delegation: “I think the why is just as important as the what,” she said.

With those constraints recorded, Shaw delegates task coordination and the compilation of reporting data while keeping strategy and work that requires personal judgment. “What AI is doing is it’s freeing up my attention so that I can spend it on judgment and relationships,” she explained. For teams that have lost experienced colleagues, the same documentation helps remaining employees supervise output even when they had no part in the original decisions.

To turn that history into an assignment, Ed Poppe, founder and fractional marketing leader at Poppe Marketing, recommends established creative-briefing practices. His marketing advisory work gives him a commercial interest in demand for this kind of management guidance. A brief should identify the deliverable, set brand boundaries, and specify where review must occur. Shaw’s decision history explains why constraints exist; Poppe’s brief turns the assignment and its constraints into instructions an agent and reviewer can use.

Okoone experts
LET'S TALK!

A project in mind?
Schedule a 30-minute meeting with us.

Senior experts helping you move faster across product, engineering, cloud & AI.

Please enter a valid business email address.

Fix the workflow before accelerating individual tasks

Clear instructions help an agent complete its assignment, but that assignment still sits inside a larger process. Sean O’Connell, head of customer success at Screendragon, distinguishes the workflow, the complete route through which work moves, from the agents performing repeatable activities within it. His customer-success role gives him a commercial interest in customers succeeding with the approaches he recommends. He warns that an agent can finish its step successfully while sending the next team more work than it can handle.

When the next team lacks capacity, faster execution moves the delay downstream. Local productivity can rise while overall throughput, the amount of work the complete process can finish, remains constrained. O’Connell therefore recommends finding the bottleneck and defining the desired result before choosing an agent’s role. “Start with the process, not the technology,” he said.

Examining the process also exposes problems that speed could magnify. Stacey Ackerman, the independent consultant and fractional CMO who moderated the discussion, warned that automating a defective manual process can intensify existing friction. Her consulting work gives her a commercial interest in demand for marketing and operational advice. Her warning concerns defects in how work moves, while O’Connell’s concerns a mismatch in capacity between stages; either can limit the benefit of faster execution.

Because these problems cross team boundaries, Poppe recommends showing a functioning pilot to partners in sales, operations, and creative teams. A working example gives hesitant colleagues something concrete to examine and shows how the change affects their responsibilities. Sharing practices can also connect individual experimenters with enterprises already running agents in production. As Poppe put it, “So learn and share.”

A pilot gives that collaboration a defined scope: choose one genuine bottleneck, set the limits of agent responsibility, and test the change with affected colleagues. The test must account for the capacity needed to document context, supervise agents, and review output, alongside the time freed from execution. Shaw’s experience shows the possibility of redirecting attention toward judgment and relationships, but each team needs to assess that result within its own workflow. Expansion should depend on evidence that the change frees up time the people responsible for the outcome can use.

Keep human ownership closest to customer consequences

Setting a pilot’s limits means deciding where people must remain involved. Shaw uses the consequences for the customer as her guide: “The closer a task gets to the customer, the more it needs a human in the loop.” Here, human involvement means having a person review or decide at the relevant stage. A technically simple message can therefore need close attention when the relationship is sensitive.

Within those limits, agents can retrieve data, format assets, and assemble initial drafts for a person to evaluate. Research, reporting, and task coordination are also candidates for delegation when instructions provide the necessary context and review requirements. Each assignment needs a clear boundary between preparation and the human decision that follows. Existing briefing practices offer a place to record that boundary before work begins.

This boundary matters especially for executive communications, sensitive renewals, and messages requiring genuine empathy. These tasks require someone to consider the relationship, its history, and the likely effect of the communication, which a polished draft alone cannot settle. Human involvement at these points protects customer relationships and the organization’s reputation. However much preparation an agent performs, that judgment remains necessary.

To keep responsibility clear, O’Connell reserves final approval for human leaders. The level of involvement varies with the task’s consequences, while a person remains answerable for the result. Task difficulty is therefore only one consideration in delegation. Leaders also need to identify who must exercise judgment and who will answer for what happens next.

Accountability requires continuing evaluation

A person responsible for approval needs a way to spot output that falls below expectations. Shaw warns that agents can produce confident, polished material that needs close scrutiny and can gradually drift from brand requirements. Initial training therefore needs continuing guidance. Editorial review must assess relevance and persuasiveness for the intended audience, since presentation quality alone cannot establish either.

To help reviewers assess the work, O’Connell proposes an evaluation layer: a second AI model that checks the agent’s output against defined brand criteria. The process starts by establishing those criteria, then uses the second model to score the work and flag items for human attention. Brand boundaries and review expectations recorded in the brief can inform the checks. This gives automated assessment an explicit basis and directs reviewers toward work that needs attention.

Directing attention helps only if the approval decision remains clear. A score can help a reviewer prioritize work, but human leaders still decide whether the output is suitable for use. Preliminary checks can then happen automatically while people retain responsibility for the resulting communication. This also gives teams a way to keep review effort manageable as the volume of generated work grows.

Because output and models can change, O’Connell recommends monitoring performance and assessing when a different model could improve accuracy or reduce costs. Review findings and departures from brand requirements can guide adjustments after the initial setup. Lightweight controls and repeated changes can help manage quality without requiring people to perform every inspection manually. Model selection, review results, and continuing adherence to brand requirements remain management responsibilities throughout operation.

Approved access can provide more oversight than blanket bans

These controls depend on employees using AI through arrangements the organization can oversee. During the audience Q&A, O’Connell described employees who face strict AI restrictions while still being expected to meet deadlines. Some respond by using personal AI accounts, moving the activity outside approved oversight. This observation describes a response to delivery pressure; it is separate from the audience poll’s measure of adoption.

To bring that activity under organizational supervision, O’Connell recommends working with employees and providing approved, secure access. A prohibition sets a rule, but it leaves deadline pressure in place, which can encourage the use of personal accounts. Approved access gives employees a permitted way to address that pressure and gives the organization visibility into how they use AI. With that visibility, teams can identify where review, access restrictions, and other controls need to apply.

Key executive takeaways

  • Preserve decision context: Marketing managers need to record brand subtleties, customer history, and the reasons behind past decisions in agent instructions. Translate that context into briefs with clear deliverables, boundaries, and review requirements.
  • Test the whole workflow: Marketing operations owners need to identify bottlenecks and downstream capacity before assigning work to agents. Use a bounded pilot to measure whether faster execution frees usable time after supervision and review.
  • Assign human ownership: Marketing leaders retain final approval and accountability for agent-assisted work. Set review requirements according to customer consequences, with close attention to sensitive renewals, executive communications, and messages requiring empathy.
  • Keep evaluation active: Brand and editorial reviewers need explicit criteria for checking relevance, accuracy, and brand alignment. A second AI model can flag output for human attention, while ongoing monitoring informs changes to instructions and model selection.
  • Provide approved AI access: IT and marketing leaders need secure, approved tools that fit employees’ delivery demands. Bringing AI use under organizational oversight makes access controls and review requirements easier to apply.

Alexander Procter

November 17, 2026

9 Min

Okoone experts
LET'S TALK!

A project in mind?
Schedule a 30-minute meeting with us.

Senior experts helping you move faster across product, engineering, cloud & AI.

Please enter a valid business email address.