Google Ads · Smart Bidding
Smart Bidding is not a black box you switch on and hope. It is only as good as the data you feed it and the discipline you keep while it learns. Most accounts pick the wrong strategy for their data, or change things mid-learning, then blame the algorithm. We match the strategy to the account and protect the learning phase.
If you run Target CPA or Target ROAS on any campaign that is limited by budget, the behaviour changed. Before, those campaigns could systematically overachieve the target you set, and performance swung whenever you adjusted the budget. From August 17, 2026, Google's bidding systems deliver more consistently toward the target you actually entered, including through budget changes. A campaign that has been quietly beating its target will trend toward that stated number, and Google will not adjust targets or budgets for you.
The change covers Search, Shopping, Performance Max, Demand Gen, Display, Hotel, and Travel campaigns across Google Ads, Search Ads 360, Display & Video 360, Ads Editor, and the API. App campaigns and Video reach and view campaigns keep their previous behaviour. There is a Bid Target Adjustment Tool live in the account that surfaces affected campaigns and lets you match the target to recent performance. The practical upshot for anyone scaling: budget increases no longer destabilise efficiency the way they used to, which makes the target you set the single most consequential number in the account.
Free Google Ads Audit
On a discovery call we check your bid strategy against your real margins and conversion data, and tell you what we would change first.
For advertisers spending $100K+/month. No cost, no obligation. A person reads every submission.
Request a Discovery Call →The algorithm rarely underperforms because it is bad. It underperforms because it was handed too little data, the wrong strategy, or a fresh change every week.
Flip a low-volume campaign to Target ROAS before it has the conversions to support it and it thrashes. Add and remove keywords, launch new ad groups, or switch strategy mid-flight and the learning phase resets every time, so the algorithm never stabilizes. Then the account concludes Smart Bidding does not work, when the real issue is how it was run. Getting it right means feeding it real revenue through offline conversion tracking, choosing the right strategy, and leaving the learning phase alone.
Four things decide whether Smart Bidding compounds or stalls.
Every campaign starts on Max Conversions. When conversions carry different values, such as revenue, we move to Max Conversion Value and apply Target ROAS. If every conversion is worth the same, we stay on conversion count with a Target CPA.
After a significant change the algorithm re-calibrates before performance settles. Google notes this can take up to around 50 conversion events or 3 conversion cycles, depending on how much conversion data is coming in.
Adding or removing keywords, adding ad groups, or changing the bid strategy are composition changes that restart learning. Per Google, changing a target does not reset what the system has learned, and budget swings are safe on a mature campaign.
Smart Bidding optimizes toward whatever you feed it. We feed offline, revenue-weighted conversions rather than form fills, so it learns to find customers, not clicks.
Moving from tCPA on form fills to tROAS on tier-weighted deliveries: subprime auto leads →
Choosing the strategy is half the job. The other half is knowing whether the algorithm is actually stabilizing, and the tell is not the daily cost per acquisition.
We read the campaign Insights graph, which shows performance inside a projected variance range: the band of where results can land. As the algorithm learns, that band narrows. A tightening range means the campaign is stabilizing and can take more spend. A range that stays wide means it needs more data, or that something was changed too soon and learning restarted. Watching the variance tighten, rather than reacting to a single day's CPA, is what keeps us from touching a campaign right when it is about to settle. The specific target math that sits on top of this lives in our guide to target ROAS.
The reason we can run six-figure daily spend without white-knuckling it is that we know precisely when a campaign has learned enough to trust.
The readiness bar is concrete: at least 50 conversions in a rolling 30-day window, on a conversion cycle that captures about 90% of events within seven days. Clear that, and the algorithm is learning fast enough to bid reliably. Miss it, and every change is a guess dressed up as optimization. Understanding the speed at which conversions are actually learned is what prevents premature changes, and premature changes are the single most common way operators reset their own progress. When the data flows fast enough and the strategy is matched to it, scaling spend is a decision, not a gamble. This is the discipline behind our our Google Ads agency practice.
The same discipline applied to the biggest black box: Performance Max for lead generation →
A deliberate sequence, held steady long enough for the algorithm to learn at each stage.
Google is productizing multi-stage feedback: bidding that learns from every stage of the lead-to-sale journey rather than a single conversion event. That is the same principle this page describes running today through offline conversion values; the beta, confirmed in May 2026, brings a native version of it. Background coverage.
Smart Bidding Exploration is now available in Performance Max and AI Max campaigns, giving target-constrained campaigns controlled flexibility to reach adjacent search categories without abandoning efficiency targets. Announced among Google Marketing Live 2026 lead generation launches.
Start on Max Conversions to gather data. If your conversions carry different values, move to Max Conversion Value and apply Target ROAS so bidding optimizes to revenue. If every conversion is worth the same, run Target CPA on conversion count. The choice follows your data, not a preference.
Composition changes reset it: adding or removing campaigns, ad groups, or keywords, and changing the bid strategy. Google's documentation on the learning period and what affects it also notes that changing a target does not reset what the system has learned, and that learning can last up to around 50 conversion events or 3 conversion cycles.
Watch the projected variance range in the campaign Insights graph, not a single day's cost per acquisition. As the algorithm learns, the band of where performance can land narrows. A tightening range means the campaign is stabilizing; a persistently wide range means it needs more data or was changed too soon.
Our readiness bar is at least 50 conversions in a rolling 30-day window, on a conversion cycle that captures roughly 90% of events within seven days. Below that, the algorithm cannot learn fast enough to bid reliably, and changes are guesses.
Yes, with a portfolio bid strategy, which applies a shared goal across grouped campaigns. Per Google, portfolio bid strategies work across Search, Display, and Shopping campaigns, though not Performance Max. We use them mostly for multi-location accounts that need budgets and targets coordinated across campaigns.
Smart Bidding needs conversion volume before it behaves. When no single campaign has it, a portfolio bid strategy pools conversions across Search campaigns, and the bidder learns from all of them at once.
Free Google Ads Audit
If you are spending $100,000 or more per month and your campaigns keep resetting their own learning, or bidding on the wrong signal, we should talk.
For advertisers spending $100K+/month. No cost, no obligation. A person reads every submission.
Request a Discovery Call →