Smart Bidding is only as good as the signal you feed it.
An account we took on was running on tCPA, the target was set sensibly, and the bidder was sticking to it. The problem wasn’t the strategy. It was what counted as a conversion: thank-you page views that also fired on a page reload, a newsletter sign-up, a PDF download. Google was optimising for exactly what it had been asked to. Those things just weren’t leads.
That’s the part most Smart Bidding conversations skip. It’s almost always about the knobs: should I run tCPA or target ROAS, how aggressive should the target be, do I raise the budget. But the bidder optimises toward the conversions it’s handed. If that signal is wrong, no setting saves it. You end up with a strong algorithm running very efficiently at the wrong target.
Garbage in. Smart Bidding doesn’t see leads. It sees conversion events. If things that aren’t leads get counted, if the action that actually matters isn’t measured, or if a sale falls outside the attribution window, that’s the reality the bidder trains on. A phone-driven account where the calls aren’t tracked is the clearest case: the bidder steers toward the form fill or the directions click, because that’s what it can see. We wrote about that separately in why “call tracking doesn’t work in the Netherlands” costs you on bids.
When Smart Bidding wins, and when it doesn’t. Whether the bidder adds value depends a lot on how much data an account produces month over month. On very large accounts the algorithm has enough to learn from and does work a person can’t keep up with by hand. On very small accounts there’s too little data to bid well manually, and Smart Bidding still helps. The middle is where it gets murky: there an experienced hand with spreadsheet discipline can often steer bids just as well or better. So the right strategy is account-dependent. But none of those regimes rescues you from a bad conversion signal. Smart Bidding doesn’t fix the small account with a polluted conversion. It just makes it efficiently wrong.
The order we work in. Signal first, then strategy. Before we touch tCPA or a target ROAS, we check what’s being counted. Are the conversions actual leads, or are there thank-you page refreshes and double-counts in there. Is the most important action measured at all. Does the attribution window match how long it really takes for a click to become a customer for this client. Only once that holds up does it make sense to tune the bid strategy. Do it the other way around and you’re adjusting the knobs to compensate for a measurement error, which never lasts.
What you stay honest about. A clean signal doesn’t make Smart Bidding infallible. It’s still a black box you can’t steer keyword by keyword, and on a thin account it can wander for weeks before it learns anything. Smart Bidding isn’t always the answer. But the debate about the knobs only matters once what you’re feeding it is right.
On nearly every audit this is the first thing we look at, before we look at the campaigns themselves: is this account measuring leads, or is it measuring noise that happens to look like a conversion. Checking it takes half an hour and it’s free.