Back to blog

What Does a CMO Decide in the Age of AI? What Automates and What Doesn't

Shusaku Yosa

AI時代のCMOは何を判断するのか|自動化できる意思決定とできない意思決定

Bidding is automated, reports generate themselves, and creative can be produced in minutes. So what is left for people to decide? This article separates marketing decisions into those that can be automated and those that cannot.

Optimisation Is Not the Same as Setting the Objective

This distinction defines the boundary of what automation can do.

Automated bidding makes it clear. Give it a target of lowering CPA and it will adjust bids far better than a person. But it will not judge whether lowering CPA is the right target for your business right now.

Set a CPA-reduction target while trying to reach new customers and delivery drifts toward the existing audience that converts easily. The numbers improve and the business does not move.

What Automation Handles

What you can hand over is fairly clear.

Optimising toward a given target

Bid adjustment, delivery scheduling, narrowing targeting. There is no longer a reason for people to do this work.

Compiling and visualising results

Gathering numbers from each tool and putting them in a table. If people are spending time on this, it is the first thing to automate.

Producing and testing creative at volume

Making several variations, watching the response, narrowing down. This cycle has become far faster.

What to say is a separate matter. Variations of the expression can be automated, but generating candidate messages requires knowing what the customer is struggling with.

Anomaly detection

Catching sudden changes in the numbers. More reliable than a person checking dashboards every day.

What Automation Does Not Handle

This is what becomes the centre of the CMO's job.

What to set as the target

Chase volume this quarter, raise the average value, or try a new market?

This does not come out of past data. It is a judgment informed by the state of the business, competitor movements, and the board's intent.

What to stop

Finding the campaigns with bad numbers can be automated. Whether bad numbers mean you should stop is a different question.

Has it only just launched, is the direction wrong, or has the market shifted? The same number means different things.

Who to aim at

Finding the responsive segments within your existing audience is exactly what automation is good at.

But the decision to go after people you have never reached will not surface, because the data is built only from past contact. Opening a new market is something a person has to initiate.

Being accountable

When an ad causes a problem, "it was generated automatically" is not an explanation anyone accepts.

Even when you automate, decide who reviews the output before it goes out.

Whether to Accept the Recommendation

Easily overlooked, but this is a decision too.

Whether to apply the recommended settings a tool suggests. In most cases, following them improves the short-term numbers.

The problem is that the recommendation only sees short-term metrics. Suggestions to lean on the existing audience appear constantly, but the fact that continuing down that path dries up the pool of new prospects is not part of the suggestion.

If you decide against a recommendation, write down why. When you review later, you can check whether the call was right.

The Work Automation Creates

Automation does not only remove work. It generates new work as well.

  • Setting and revisiting targets: deciding what to optimise for becomes a recurring task.
  • Reviewing output: someone has to check whether what came out can go public.
  • Handling exceptions: the more you automate, the more the exceptions stand out.

Treat the second lightly and problematic wording or factual errors go out as they are. The greater the volume, the more you need a review mechanism.

Improving the Quality of Decisions

The more automation advances, the larger the share of the outcome that rests on human judgment.

What raises the quality of a decision is not more data. It is whether you can read the numbers you already have in the context of the business.

Concretely, put these two in place.

  • Spend and results by campaign, visible in the same place
  • Past decisions, and the reasons behind them, still on record

Many organisations lack the second. Without a record of why a decision was made, the same mistake repeats.

Frequently Asked Questions

Will marketers have less work?

Less execution, but more decisions. When execution filled the day, deferring a decision did not show. Once the execution is gone, an unmade decision becomes a visible delay.

Where should we automate first?

Gathering the numbers. As long as time goes there, no time is freed for decisions. Worth doing before creative generation.

Can we just leave it to automated bidding?

Give it the wrong target and it will proceed accurately in the wrong direction. Revisit periodically whether the target itself still fits the business.

Does a small team need automation?

The smaller the team, the greater the effect. When the person deciding also carries the execution, the execution crowds out the decisions. That said, adding so many tools that managing them becomes the work defeats the purpose.

Freeing Time for Decisions

The point of automation is to cut execution and free time for decisions. But when the numbers are scattered across different places, the gathering work remains.

Xtrategy manages campaign schedules alongside budget and KPIs on a single screen. With the inputs for a decision on one page, gathering time becomes thinking time.

Summary

  • Automation handles optimisation. It cannot set the objective.
  • Hand over bid adjustment, compilation, creative volume, and anomaly detection.
  • Target-setting, the decision to stop, going after new audiences, and accountability stay.
  • Accepting a recommendation is itself a decision. If you decline, record why.
  • Automation adds work: target-setting, output review, and exception handling.
  • Decision quality comes from reading numbers in context, not from more data.

When considering automation, separate whether the work is optimising toward a target or setting the target itself. The former you can hand over; the latter stays with you.

Back to blog