I have never had a problem with automation. A large part of what I build now exists specifically to automate work I do not want to repeat manually.
What I do have a problem with is automation that removes the explanation.
That is the part of Google Ads that changed most dramatically for me over the years. I went from working with a system where I could get extremely close to the individual search, click and sale, to a platform where more and more of the decision-making happened inside Google's machinery.
The result might still be good. But I increasingly could not see why.
I liked the detail
I spent years working with AdWords and paid search. I was the sort of person who enjoyed the detail rather than seeing it as a nuisance.
On one of my old affiliate T-shirt sites, I matched referrer and keyword information with Commission Junction SID tracking so I could connect traffic back to actual sales. I used the Google Ads API to create relevant ads and could put particular attention on cheaper long-tail terms.
That level of visibility made the system interesting.
I could form an idea, test it with real traffic, see what happened and alter the next decision based on the result. The value was not simply that Google could send me clicks. The value was that I could learn from the clicks.
Automation itself was not the problem
I was already automating parts of paid search myself.
Using an API to generate relevant advertising from data is automation. Tracking sales back to the traffic that generated them is automation. Building rules around what was and was not working is automation.
But it was automation I could understand.
I knew what information went in, what logic I was applying and what came out the other side. If something behaved strangely, I had somewhere to look.
That distinction has become increasingly important to me.
Then more of the reasoning disappeared inside the platform
Over time, Google Ads moved more decisions into its own automated systems.
From Google's point of view, that makes sense. It has vastly more data than any individual advertiser and can make decisions at a scale no person could realistically reproduce by hand.
My frustration was not that a machine was making the decision. It was that the evidence available to explain the decision became thinner.
If performance changed, I wanted to understand what had changed with it. Which searches were different? Which type of traffic had moved? What behaviour had the system detected? Which assumption of mine had turned out to be wrong?
"The algorithm optimised it" is an answer to what happened. It is not much of an answer to why.
The black box changed what I valued
This did not single-handedly cause me to stop doing client work or start building my own systems. My career did not move in one neat line like that.
But it reinforced something I already cared about: visibility.
I like systems where the data can be inspected. I like knowing where a number came from. I like being able to trace an outcome back through the decisions that produced it.
That is one reason the systems I build for myself tend to collect more context than the bare minimum.
With Control Room, I do not just want a final number if I can preserve the information that explains it. With InPlay, I want decisions, candidate data and timestamps to remain understandable rather than collapsing everything into a single unexplained result.
The same instinct appears in my affiliate systems. A sale is useful. Knowing which video, product, click or piece of content contributed to it is more useful.
AI makes the lesson even more relevant
This matters even more now because I use AI throughout the things I build.
I am happy for an AI system to do work that would otherwise take me hours. But I do not want that convenience to become an excuse for losing track of the source material, the data or the reasoning that made the output possible.
In fact, the more automated a system becomes, the more important I think provenance becomes.
If an article is generated, what facts was it allowed to use? If a trading decision is suggested, what market information was available at that moment? If a recommendation appears in Control Room, where did the underlying data come from?
Those questions are the modern version of the same thing I wanted from paid search years ago.
I want automation with a window in the side
I do not want to go back to doing everything manually.
The whole point of the systems I am building is to automate more, connect more and remove repetitive work.
I just want to be able to look inside.
Google Ads taught me how powerful automation can be. It also taught me how different automation feels when the controls and explanations gradually disappear.
That is probably why, whenever I build something for myself now, I keep coming back to the same requirement: automate as much as possible, but keep enough evidence to understand what actually happened.
