Automation Fails to Remove Vague Goals

It has never been simpler to delegate marketing tasks to automation. Platforms handle bidding, targeting, and creative assets. CRMs score leads, trigger workflows, and recommend next steps. You can feed performance data into an AI assistant and receive optimization suggestions before your coffee finishes brewing.

None of this is exaggeration. These systems genuinely and continuously work toward the goals you set, and their recommendations keep improving.

But notice what remains unchanged in that equation: the goals you set.

Automation cannot fix a vague objective. Handing a fuzzy goal to a person usually leads to messy results and unclear direction. The same holds true for AI.

When automation does exactly what you asked

Give a fuzzy goal to AI, and you will likely get overly confident results heading in an overly confident direction. The system will find the most efficient path, even if it leads the wrong way.

Ask for higher ROAS, and automated bidding will happily lean into branded search, warmer prospects, and retargeting. People who were already going to buy. The ROAS climbs.

Ask for more signups, and your campaigns may fill the funnel with low-intent volume that never activates. Signups go up.

Ask for lower CAC, and the system will quietly shrink your reach down to the easiest audience you have. CAC drops.

In every case, the metric improves, but the business might not. The automation did exactly what you said. It just did not do what the business actually needed.

Stop giving automation a direction. Give it a field.

“We need higher ROAS” is not an objective. It is a direction. An optimizer will follow a direction forever, right past the point where it stops helping the business.

What automation actually needs is a playing field. Clear sidelines on both sides. What counts as a win, and just as explicitly, what counts as a loss, even if the metric looks good.

Take a brand running paid campaigns at an 8x ROAS. Leadership wants growth, so the real objective is not protecting the 8x. It is acquiring more new customers. Said properly, the target sounds like this: we will accept ROAS coming down from 8x to 5x if new customer volume grows with it. Below 5x, we stop and reassess. That is the floor.

Now the AI has room to move. It can expand audiences, chase incremental customers, and spend into less efficient territory, because someone decided in advance how much efficiency the business is willing to trade and where the line is. Without that floor, you get one of two failure modes. Either the team strangles the campaigns protecting a ROAS number nobody actually needs, or the system spends its way down with no agreed stopping point.

The skill is not just picking the metric. It is defining both sidelines before the game starts.

Decide what to turn off before you turn it on

The same thinking applies to newer AI features inside the platforms themselves, and this is where I see teams skip the homework.

Consider an advertiser in a regulated industry, insurance for example, turning on Google’s AI Max. The default posture is to enable everything and let the system optimize. But for that advertiser, the loss conditions include things no dashboard will ever flag. The AI rewriting carefully reviewed ad copy into something compliance never approved. Brand terms getting pulled into broadened matching where they do not belong.

So the right move is deciding what to disable before enabling anything. Text customization off. Brand excluded. Then let the AI work hard inside what is left.

That is not distrust of the technology. It is the opposite. It is giving the system a field it can sprint on. Guardrails are what make autonomy safe enough to actually use.

Automating a guess is still a guess

One more version of this, on the CRM side, because it is easy to assume the problem only lives in paid media.

It is very easy to build elaborate automated workflows. Trigger this email, assign this task, nudge the customer toward this action. The question that rarely gets asked is whether there is any data showing that customers who take that action actually retain better. Plenty of workflows are engineering effort layered on top of an assumption. The automation runs. The assumption was never tested.

If you would not have a person do the task manually because you cannot say what it improves, automating it does not make it smarter. It just makes the guess run on a schedule.

Where the humans go

None of this is an argument for less automation. The tools are good and getting better, and refusing to use them is its own kind of risk at this point.

It is an argument about where human judgment now earns its keep. Not approving every bid change or reviewing every suggested action. The human job is owning the definition of the field: the floors, the exclusions, the tradeoffs the business will accept, and the wins that do not count. Then watching for the moments when the system is winning the metric but losing the game.

Automation will hit whatever target you give it. That is exactly why the target deserves more thought than it is getting.

So here is a question worth sitting with: if every automated system you ran hit its number this quarter, how many of those wins would actually move the business forward?