The AI Bidding Transparency Gap: What Smart Bidding Doesn't Tell You
Target CPA and Target ROAS are bidding strategies, not explanations. What the algorithm does inside the black box — and why the settings around it still matter.
Smart Bidding works. The evidence that machine learning outperforms manual CPC management in most accounts, at most scales, is now fairly settled. What Smart Bidding doesn't do is explain itself. Target CPA and Target ROAS are instructions you give the algorithm, not windows into what it does with them. The gap between "the algorithm is optimizing" and "I understand what the algorithm is doing" is where most account problems live.
The algorithm is optimizing for what you told it to optimize for
The most common Smart Bidding failure mode isn't a bug — it's a misconfiguration. The algorithm faithfully pursues whatever conversion action it was told to pursue, at whatever target it was given, using whatever signals it has access to. If the conversion action is low-quality (a page view instead of a purchase, a lead form fill instead of a qualified lead), the algorithm will optimize for low-quality conversions. If the target is too aggressive, it will sacrifice volume. If the signals are degraded — missing conversion data, incorrect attribution windows, lost enhanced conversion matching — the algorithm works with bad inputs and produces bad outputs.
What you can't see
Smart Bidding does not expose which signals it weighted most heavily for a given auction, which users it decided were worth bidding up, or why it chose a particular bid for a particular impression. Google's Auction Insights and Search Term reports give partial visibility, but the core logic — the per-auction bid calculation — is opaque by design. That's the trade you make: better average performance, less per-decision transparency.
What you can see — and need to track
The settings that govern Smart Bidding are visible and controllable: the target, the conversion actions in scope, the attribution model, the bidding strategy type, the budget. These are the inputs to the black box, and they are the things that change. When Smart Bidding performance shifts unexpectedly, the answer is almost never inside the algorithm — it's in the settings that changed around it. A target that drifted, a conversion action that was added or removed, an attribution window that was updated. The record of those changes is what makes the algorithm accountable.