How AI Bidding Strategies in Google Ads Perform When Campaign Structure Is Wrong

Target CPA, Target ROAS, Maximise Conversions, and Maximise Conversion Value all use real-time auction signals that no manual bidder can replicate at scale: device, location, time, audience behaviour, search context, and dozens of other variables processed simultaneously at the moment of each auction.

When Smart Bidding works, it works because the algorithm has enough of the right data to make accurate predictions. When it fails, the failure is almost always traced back to the same place: campaign structure that prevents the algorithm from learning correctly.

Understanding why that happens requires understanding what Smart Bidding actually needs to function.

What Smart Bidding Requires to Learn

Smart Bidding is a machine learning system. Like any learning system, it requires sufficient and relevant data to produce reliable outputs. The primary data source it learns from is conversion data generated within the campaign itself.

Google’s own guidance suggests a minimum of 30 to 50 conversions per month per campaign for Smart Bidding to move out of the learning period and into stable optimisation. Below that threshold, the algorithm is operating on too little signal to make confident predictions. It is not broken in that state. It is uncertain, and uncertainty produces inconsistent bidding behaviour.

Campaign structure determines how conversion data is distributed across the account. A structure that fragments that data across too many campaigns, too many ad groups, or too many conversion actions prevents any single campaign from accumulating the volume the algorithm needs.

How Fragmented Campaign Structure Starves the Algorithm

The most common structural mistake in accounts running Smart Bidding is excessive campaign segmentation. Teams that built their account structure for manual bidding, where granular segmentation gave a human bidder precise control, often carry that same structure into Smart Bidding without recognising that the logic has inverted.

With manual bidding, more campaigns mean more control. With Smart Bidding, more campaigns mean less data per campaign. An account generating 200 conversions a month spread across ten campaigns gives each campaign 20 conversions on average. That is below the threshold for stable learning in every single campaign simultaneously.

The algorithm responds to data scarcity by bidding more conservatively, missing volume, or oscillating between overspending and underspending as it attempts to gather more signal. The campaign appears to be underperforming. The instinct is often to adjust targets or switch strategies. Both responses make the data problem worse.

Mixed Intent Within Ad Groups

A separate but related structural issue is ad group composition. Smart Bidding uses the query, ad, and landing page together as part of the signal set it evaluates at auction time. When an ad group contains keywords with meaningfully different intent, the algorithm receives conflicting signals about what a click from that ad group is worth.

A single ad group containing both broad awareness queries and high-intent purchase queries will generate conversion data at very different rates depending on which queries are serving on a given day. The algorithm interprets this as volatility in conversion probability and adjusts bids accordingly, often in ways that do not align with the actual value of individual queries.

Tighter ad group composition gives the algorithm a cleaner signal. It knows more reliably what kind of intent it is bidding into, which makes its conversion probability predictions more accurate.

Broad Match in a Fragmented Structure

Broad match and Smart Bidding are designed to work together. Google’s position is that broad match gives Smart Bidding a wider pool of auction opportunities to learn from, which allows the algorithm to find conversion-generating queries that exact or phrase match would exclude.

That relationship holds when campaign structure is sound. When structure is fragmented, broad match compounds the problem. A campaign with insufficient conversion volume running broad match expands into a wider query space before the algorithm has established reliable conversion predictions for the queries it already knows. The result is wasted spend on low-intent traffic while the campaign simultaneously struggles to hit its targets.

Broad match earns its place in an account after the algorithm has established a reliable conversion pattern, at adequate budget, with sufficient historical data. It is a scaling tool, not a learning tool.

Budget Constraints During the Learning Period

Smart Bidding’s learning period typically runs for one to two weeks following any significant change to a campaign. During that period, the algorithm tests a wider range of bids to gather the signal it needs to calibrate toward the target.

A campaign with a budget that constrains daily impression volume during this period extends the learning period because the algorithm cannot generate enough auction data quickly enough. A campaign with a very tight budget running Target CPA will frequently exit the learning period before it has gathered sufficient conversion data, then re-enter it the next time a structural or target change is made.

The practical implication is that Smart Bidding requires adequate budget to learn efficiently. A campaign that is consistently budget-limited is a campaign that will struggle to exit the learning period and reach stable performance.

Conversion Action Configuration

The conversion actions a campaign optimises toward are as important as the campaign structure itself. Smart Bidding optimises for whatever conversion action it is given. When the conversion action is too high in the funnel, page views or time on site for example, the algorithm optimises for actions that do not correlate with business value. When multiple conversion actions with very different values are pooled together, the algorithm cannot distinguish between them and optimises toward whichever generates the most volume regardless of quality.

Configuring Smart Bidding to optimise toward the conversion action that most accurately represents business value, a qualified lead, a completed purchase, a booked call, is a structural decision. It sits outside the bidding strategy itself but determines entirely what the bidding strategy is working toward.

What a Structure That Supports Smart Bidding Looks Like

The account structures that allow Smart Bidding to perform consistently share the same characteristics. Campaigns are consolidated enough that each one accumulates conversion data above the learning threshold. Ad groups are grouped around coherent intent rather than keyword volume. Conversion actions reflect actual business value. Budgets allow the algorithm to operate without daily constraint, particularly during learning periods.

The shift from manual bidding to Smart Bidding is not simply a settings change. It is a structural change that requires rebuilding how campaigns are organised around the data needs of the algorithm rather than the control preferences of the human manager.

Smart Bidding performs well when the inputs are right. The inputs are determined entirely by campaign structure.