Maximising the return on AI
Lessons from Google's advertising platform.
The most instructive public example in applied AI is still Google's advertising platform. Not for the technology, which is well documented, but for the pattern of how the value was captured.
The returns did not come from a single algorithmic breakthrough. They came from a decade of compounding decisions: aligning the auction mechanics to advertiser outcomes, instrumenting feedback loops that got tighter every quarter, and reorganising the commercial team around what the system actually optimised for. The AI was the engine. The business model, the data access and the operating structure were what let the engine do work.
Most companies trying to replicate "AI-driven value" skip the second half. They buy the engine and bolt it onto an operating model designed around human judgement and quarterly reviews. The engine runs. The returns do not compound.
The lesson is not that every company should become an ad platform. It is that AI returns are structural, not episodic. They come from reshaping the feedback loop between what the system learns and what the business does, and that reshaping takes years of patient, unglamorous work.
There is no shortcut. There has not been one for twenty years.
The decision: budget AI returns on a multi-year compounding curve, not a quarterly uplift — or do not budget them at all.