It has never been easier to outsource marketing work to automation. Ad platforms will handle bidding, targeting, and creative. CRMs will lead leads, trigger workflows, and suggest next action. You can enter performance data into the AI assistant and get recommendations for improvements before your coffee is even finished.
None of this is hype. These systems truly work towards the goals you give them, automatically and continuously, and the suggestions keep getting more and more useful.
But notice what hasn’t changed in that sentence: the target you’re giving them.
Default doesn’t fix a vague goal. A vague goal given to someone often produces messy results and unclear direction.
Your customers are searching everywhere. Make sure it’s your product he appears.
The SEO toolkit you know, and the AI visibility data you need.
Start a Free Trial
Start with
If the automation does what you asked
Give the AI a vague goal, and you’ll likely get overconfident results and overconfident guidance. The system will find the most efficient way in the wrong place.
Ask for a high ROAS, and automated bidding will happily rely on branded searches, warm leads, and retweets. People would buy anyway. ROAS increases.
Ask for more signups, and your campaigns may fill the funnel with low intent volume that never works. Enrollment is increasing.
Ask for a low CAC, and the system will silently narrow your reach down to the easiest audience you have. CAC is decreasing.
In all cases, the metric improves, but the business may not improve. The automation did what you said. It didn’t do what the business needed.
Stop giving way to automation. Give it a field.
“We need high ROAS,” is not the goal. It is a guide. The optimizer will follow the index forever, past the point where it stops helping the business.
What automation needs is a level playing field. Wipe the sides on both sides. What matters as a win, and obviously, what matters as a loss, even if the metric looks good.
Take a brand that runs paid campaigns at 8x ROAS. Leadership wants to grow, so the real goal is not to protect 8x. It gets more new customers. Well said, the target sounds like this: we will accept the ROAS drop from 8x to 5x if the new customer volume increases as well. Less than 5x, we stop and recheck. That’s it down.
Now AI has room to move. It can increase the audience, chase growing customers, and spend in an inefficient area, because someone has 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 ways to fail. Either the team is throttling campaigns that secure a ROAS number that no one needs, or the system is wasting its way down outside of the agreed upon area.
Skill is not just about choosing a metric. Explains both sides before the game starts.
Decide what to close before you open it
The same thinking applies to new AI features within the platforms themselves, and this is where I see teams skipping homework.
Consider an advertiser in a regulated industry, insurance for example, to unlock Google’s AI Max. Auto stop enables everything and allows the system to run at full capacity. But for that marketer, loss situations include things that no dashboard will ever flag. AI rewrites have carefully revised the ad copy to something that has never been approved. Brand terms are drawn to extended analogies where they don’t belong.
So the right move is to decide what to block before allowing anything. Text customization is disabled. The brand is not included. Then let the AI work hard inside the rest.
That’s not distrusting technology. It’s the opposite. It gives the system a field in which to run. Guardrails are what make independence safe enough to use.
A default guess is still a guess
Another version of this, on the CRM side, because it is easy to think that the problem remains only in paid sources.
It is very easy to create detailed automated workflows. Start this email, assign this task, move the customer to this action. A question that is rarely asked is whether there is any data that shows that customers who take that action retain better. Most workflows are engineering efforts superimposed on imagination. Automation works. The assumption was not tested.
If you wouldn’t have someone do this work for you manually because you can’t tell what you’re improving, automating it doesn’t make you smarter. It just takes the guess work out of the schedule.
Where people go
None of this is an argument for less automation. The tools are great and getting better, and refusing to use them is your own kind of risk at this point.
It is an argument about where human judgment now ends. Not allowing all bid changes or reviewing all suggested actions. A person’s work owns the meaning of the field: low, excluded, trades that the business will accept, and winnings that cannot be counted. Then look for times when the system wins the metric but loses the game.
Automation will beat any direction you give. This is exactly why the target needs more consideration than it gets.
So here’s the question you have to sit with: if every automated system you use hits its number this quarter, how many of those benefits would move the business forward?
