AI Didn’t Solve This Problem. We Did.

How an hour spent wrestling with Google Ads conversion tracking became a lesson in the future of digital marketing.

There’s a narrative floating around right now that AI is going to replace marketers, developers, writers, analysts, consultants, and pretty much anyone else who spends their day sitting in front of a computer solving problems. Personally, I think that’s a gross misunderstanding of what AI actually does well.

This week gave me one of the clearest examples I’ve experienced so far.

I was implementing conversion tracking for one of our law firm clients, which, on paper, sounded almost laughably simple. I wanted to measure two things that actually matter to the business: when someone submits an intake form and when someone clicks the firm’s phone number. Those are the moments where an anonymous website visitor starts becoming a potential client, and if we’re spending money on Google Ads, those are exactly the signals I want Google’s machine learning optimizing toward.

It’s a simple objective but not so simple implementation. If you’ve worked with Google’s marketing ecosystem recently, you’ll know exactly what I mean. Somewhere along the way Google managed to create multiple legitimate ways to measure the same conversion. You can create conversion actions directly inside Google Ads, define events in Google Analytics 4, send those events through Google Tag Manager, import them back into Google Ads, or maintain a mixture of all of the above if you enjoy future versions of yourself wondering why nothing makes sense anymore.

I had no interest in building another tracking system that only I could understand. My goal was to build something we’d be comfortable deploying for every ClearBox client going forward, because good agency systems shouldn’t depend on tribal knowledge or crossed fingers. Six months from now, someone should be able to open the account and understand exactly how conversions are being measured.

That’s where AI entered the picture. Not as the expert. Not as the decision-maker.

More like an incredibly fast research analyst who never gets tired, never forgets what we’ve already tested, and can synthesize information from five different places faster than I ever could on my own.

That’s an unbelievably useful capability. It’s also completely capable of leading you straight into the weeds if nobody is driving.

Throughout this implementation, AI and I were constantly working together. It would analyze the GTM container, compare triggers, review tag configurations, reason through redirect behavior, interpret Tag Assistant, compare what we were seeing inside Google Analytics, and suggest possible explanations for what wasn’t working. Sometimes those suggestions were exactly what we needed. Other times they were technically plausible but completely disconnected from reality, and that’s where the human part of the partnership became incredibly important.

At one point we had convinced ourselves that something deep inside Google’s ecosystem had to be broken. We were discussing consent mode, redirect timing, Google Ads processing delays, GA4 propagation, browser caching, and several other explanations that all sounded wonderfully sophisticated. The AI was doing exactly what it was designed to do by generating possibilities and helping narrow them down through testing.

Then I noticed something sitting quietly in the corner of Google Tag Manager. “Workspace Changes.” The container hadn’t been published. That’s it.

We had built everything correctly. We had tested everything thoroughly. We’d analyzed configurations, discussed edge cases, debated Google’s engineering decisions, and spent the better part of an hour wandering through increasingly complicated theories before I finally realized we’d forgotten the one step that actually makes any of it exist. I laughed. Mostly because I knew exactly what had happened. This wasn’t an AI failure. It wasn’t a Google failure.

It was a perfect demonstration of why AI still needs someone at the wheel.

Artificial intelligence is incredibly good at exploring possibility space. Give it enough information and it’ll happily generate ten perfectly reasonable explanations for why something isn’t working. What it doesn’t naturally do is stop every few minutes, look around the room, and ask the wonderfully boring question every experienced technician eventually learns to ask:

“Did we publish the container?”

That’s human judgment. That’s human experience. That’s human curiosity directed by context instead of probability.

The funny part is that after publishing the container, almost everything immediately started behaving exactly the way we’d expected. The conversion tags fired correctly. The form submission event appeared. The phone call event worked. Google Tag Manager looked healthy. Google Analytics started receiving data.

Well…eventually.

Because Google wasn’t quite finished messing with us.

Tag Assistant insisted everything was working perfectly, while GA4 DebugView stared back at us with the digital equivalent of a shrug. For a few minutes it genuinely looked like one Google product was contradicting another. Rather than assuming we’d broken something again, we slowed down and started collecting evidence. We disabled the VPN, opened a clean browser session, repeated the tests, watched the Realtime reports, and compared what each platform was actually telling us instead of reacting to whichever screen happened to be open.

Sure enough, the events began appearing exactly as expected.

That turned out to be another useful reminder. Google’s products are connected, but they aren’t synchronized. GTM can tell you a tag fired immediately. GA4 Realtime might acknowledge it moments later. Google Ads may not make that event available for import until considerably later. If you don’t understand that relationship, it’s very easy to mistake normal processing delays for technical failures.

AI was genuinely valuable during this stage because it never lost track of what we’d already tested. It remembered every configuration, every trigger, every event name, every screenshot, and every conclusion we’d already ruled out. It could synthesize information faster than I could reasonably hold it all in my head, which meant I could spend my energy making decisions instead of trying to remember whether we’d already checked a particular setting thirty minutes earlier.

That’s a very different relationship than the one people usually describe when they talk about AI. I wasn’t asking it for answers. I was interrogating its reasoning. Sometimes I’d accept its recommendations. Sometimes I’d reject them. Sometimes I’d tell it that it was overthinking the problem (in not so kind words, lol).

And yes, at least once during this adventure I basically had to tell it to shut the fuck up because we’d wandered miles away from the obvious answer.

Oddly enough, I think that’s exactly what healthy human-AI collaboration looks like.

By the end of the implementation we had accomplished something much more valuable than simply getting conversion tracking working. We had developed a measurement architecture that we’ll now use across every future ClearBox client. Google Tag Manager collects meaningful business events. Google Analytics 4 becomes the single source of truth for those events. Google Ads imports those conversions instead of maintaining its own competing definitions, and redundant conversion actions are removed so every lead has exactly one definition throughout the entire system.

More importantly, we walked away with something that can’t be downloaded from Google’s documentation. We walked away with a better process. That’s something I’ve been thinking about a lot lately.

Everyone seems obsessed with asking whether AI will replace experts, but I think that’s the wrong question entirely. The more interesting question is what happens when you combine an experienced professional who’s deeply curious with a machine that’s exceptionally good at organizing information, recognizing patterns, and remembering everything that’s happened during an investigation.

This week gave me my answer; the AI didn’t solve the problem. But I clearly didn’t solve the problem alone either.

The solution emerged because one of us could process an extraordinary amount of information while the other could recognize when we were drifting away from reality, challenge assumptions, decide which evidence actually mattered, and occasionally remember to click the fucking Publish button.

That’s not replacement by AI. That’s augmentation and it’s kinda like a super power.

And if this week is any indication, I think that’s where the real future of digital marketing is going to be found.