Smart bidding works, and the migrations that go badly nearly all go badly for the same two reasons: the account was optimising toward a conversion definition nobody had checked, or the change was made at the same time as three other changes so nothing could be attributed.
The algorithm is only as good as the signal it is given. Handing over bidding while the signal is wrong simply makes the account very efficient at buying the wrong outcome.
Fix the conversion definition first
Before changing anything about bidding, look at what the account counts as a conversion. If a fifteen-second phone call, a newsletter signup and a genuine quote request all count equally, smart bidding will chase whichever is cheapest, and that is almost never the one that pays.
This is the highest-value hour in the whole migration and it is routinely skipped because it feels like housekeeping rather than strategy. It is the strategy.
Check you have the volume
Smart bidding needs enough conversions to learn from. The commonly cited floor is around thirty in thirty days per strategy, and in practice comfort starts somewhat above that. Below it, the model is making decisions on noise and performance can get worse rather than better.
If volume is thin, the answer is usually to consolidate rather than to abandon the idea: fewer campaigns, each with more data, will outperform a fragmented structure where every campaign is starved.
Most failed smart bidding migrations were doomed before the switch. The account had six campaigns each with four conversions a month, and no algorithm can learn from that.
Olivia Grant, Head of Paid Media, Media @ Marsons
Change one thing at a time
Switch the bid strategy and nothing else. Not the budget, not the targeting, not the ad copy. If three things change together and performance moves, you have learned nothing about which caused it, and the temptation to revert everything is strong precisely when patience is most needed.
Where the account is large enough, use an experiment so the old and new strategies run against split traffic. That converts an argument about whether it worked into a measurement.
Expect a genuine learning period
Roughly one to two weeks of unstable performance is normal, and it frequently looks worse before it looks better. The single most common way a migration fails is being reverted on day four, which wastes the learning and leaves the account no wiser.
Decide in advance how long you will hold and what result would make you revert. Writing that down before the change removes the emotion from the decision at the point you are most inclined to panic.
Start with a target you can actually hit
Setting a target cost per acquisition well below anything the account has historically achieved is a common and expensive mistake: the strategy throttles delivery trying to hit an impossible number, volume collapses, and the conclusion drawn is that smart bidding does not work. Start at or slightly above your current actual figure and tighten gradually once it is stable.
What to watch afterwards
- Cost per genuine enquiry, not cost per conversion action.
- Impression share, to see whether the strategy is throttling delivery.
- The spread of spend across campaigns, which often shifts noticeably.
- Search terms, since the mix usually changes and the negative list needs revisiting.
When to stay manual
Very low volume accounts, brand-new campaigns with no history, and situations where you have strong outside knowledge the platform cannot see, such as a supply constraint or a service you have temporarily stopped selling. In those cases manual control is the better instrument rather than nostalgia. The PPC hub covers how this plays out by sector.
Seasonality confuses the model, and you can warn it
Bid strategies learn from recent history, which means a predictable annual spike looks to the model like an anomaly. A business whose demand triples for three weeks will find the strategy underspending exactly when it should be leaning in, because nothing in the recent data suggested it was coming.
Seasonality adjustments exist for this and are widely unused. Set one for a known event with a defined start and end and an expected conversion rate change, and the strategy enters the period with the right expectation rather than spending the first week discovering it.
Portfolio or campaign level
A portfolio strategy pools conversion data across campaigns, which helps thin accounts learn faster and removes the ability to hold different targets per campaign. Campaign-level strategies give control and starve on low volume. The usual answer is portfolio for the long tail and campaign level for the two or three campaigns whose economics genuinely differ.
Nervous about handing over bidding?
We will tell you whether your account has the conversion volume to support smart bidding, and what to fix first if it does not.
