Smart Bidding was, in hindsight, the start of a new era in performance marketing. The algorithms were often worse than manual bids at first. But that was not the point. What was taking shape was a shift of power: away from the marketer operating every lever individually, towards a system that takes over execution and pushes the human into the role of strategist, whether they want it or not.
Performance Max then accelerated that process massively. Suddenly an algorithm decided where your budget landed. Search, Display, YouTube, Shopping, all in one black box. The marketer could watch, set guardrails, but detailed control was gone. For people like me, who had spent years being proud of optimising bids manually, that felt at first like a loss of control. And at the same time like a liberation, if you were willing to see it that way.
The wrong question
Many agencies responded to that shift by selling AI as an accelerant for existing processes. Do the same thing faster, more output for less input. The classic efficiency narrative.
We asked: “How do we do this faster?” The right question would have been: “Why are we still doing this ourselves at all?”
I told myself exactly that story for a while. At SlopeLift, together with the AI team, we built workflows, tested them and tore them down again. We launched pilots, some of which caught fire immediately while others missed completely. And at some point, after enough torn-down workflows, it struck me that the problem was not the technology but our frame of thinking. We asked: “How do we do this faster?” The right question would have been: “Why are we still doing this ourselves at all?”
That sounds like a small difference. In practice it is enormous. Accelerating a bad process only produces more of the wrong thing, faster. Anyone who has lived through that (and I have) does not forget it. The first question always has to be: does this process belong in the shredder before we put a new tool on top of it?
The shredder before the tool
Out of that grew something at SlopeLift that we call the anti-to-do list. A systematic search, not for tasks that should go faster, but for tasks that can be handed over to an agent entirely. Three categories crystallised out of it, and they show up in a similar form at almost every company.
First, the work nobody wants to do. Data cleaning, formatting, repetitive reporting steps. The things qualified people consider beneath them, but which still have to get done, and which therefore often get left lying around or done half-heartedly. Second, the knowledge gaps: quick research into new industries, competitive analyses, briefing preparation. Everywhere that missing information blocks the process. And third, the blank-page blocks: the jump from zero to something. First drafts, concept scaffolds, overcoming the emptiness that stands at the beginning of every creative process.
The anti-to-do list sounds simple because it is simple. But it changes the whole perspective on AI in a company. The question is no longer “which tool do we need?” but “what do we no longer want to do ourselves?”
The logical consequence
What came next was not career planning. It was the only step that still made sense.
As CRO at SlopeLift I had led the shift from manual reporting to automated dashboards. Done. But from 2024 onwards it became clear that the overlaps between data analysis and AI application were too large to manage in separate responsibilities. AI projects need decision-making authority and their own budget. They need someone who knows the operational reality of the agency and at the same time understands where the technology is heading. In October 2025 that became the CAIO title. Chief AI Officer. Sounds like a corporate buzzword, and yes, it is a bit of that too. But the core is real: AI cannot be a side project. It needs central ownership.
My guess as to why 56% of CEOs report no measurable AI ROI, according to the PwC Global CEO Survey (Davos, January 2026, 4,454 respondents, 95 countries): in most of those companies the wrong question was asked. Or none at all.
Back to Part 1 | Continue with Part 3: Human-on-the-Loop