How Smart Bidding works
Smart Bidding adjusts your bid for every individual auction using signals not accessible to manual bidders: the specific search query, the user's device, location, time, whether they are on a remarketing list, their browsing behavior, and patterns from similar auctions across Google's network.
For Smart Bidding to work well, it needs enough conversion events to learn from. Without sufficient data, the algorithm makes poor predictions and either overpays to force volume or underbids and misses opportunities. The quality of your conversion data matters as much as the quantity.
The four Smart Bidding strategies and when to use each
Google offers four Smart Bidding strategies. Each suits a different account maturity and business goal.
- Maximize Conversions: get as many conversions as possible within the budget. Best for new campaigns with no historical data — no target required.
- Target CPA (tCPA): optimize for a specific cost per acquisition. Requires 30–50 conversions/month. Best once you know your sustainable CPA.
- Target ROAS (tROAS): optimize for a specific return on ad spend. Requires 50+ conversions/month with revenue values. Best for e-commerce and lead gen with known lead values.
- Maximize Conversion Value: maximize total conversion value within the budget. Best for accounts with high variance in conversion values — e.g., different products at different price points.
The data requirements before switching to Smart Bidding
The most common Smart Bidding failure mode is switching before conversion data exists. An account with five conversions per month cannot build reliable predictions. The algorithm enters a semi-permanent learning phase, spending aggressively while learning, then settling on a behavior that is based on insufficient data.
Minimum recommended thresholds before activating Smart Bidding: 30 conversions/month for tCPA, 50 conversions/month for tROAS. If your campaign is below these thresholds, use Maximize Clicks (which improves impression share and drives traffic) or manual CPC while building conversion volume.
Setting targets without over-constraining the algorithm
The most common configuration mistake with tCPA and tROAS is setting an initial target that is too aggressive. If your account average CPA is $45 and you set a tCPA of $25 from day one, the algorithm dramatically reduces impression share looking for below-$25 opportunities that may not exist in sufficient volume.
Start with a tCPA target 15–20% above your recent average CPA. Let the algorithm stabilize for 2–3 weeks. Then reduce the target gradually — 5–10% per adjustment — while monitoring conversion volume for drops.
- Set initial tCPA at 115–120% of your recent actual CPA
- Allow 3 weeks before evaluating — the algorithm needs time after any target change
- Reduce targets no faster than 10% per adjustment with a 2-week gap between changes
- Watch impression share and absolute conversion volume, not just CPA — a lower CPA with 50% fewer conversions is usually the wrong trade
When to revert from Smart Bidding to manual
Smart Bidding underperforms and should be reverted in three scenarios: conversion tracking breaks and the algorithm starts optimizing for the wrong signal; the campaign goes into a prolonged learning phase with no exit; or the target is set so aggressively that impression share collapses.
Reverting is not failure — it is the right call when Smart Bidding conditions are not met. Switch to manual CPC, fix the root cause (conversion tracking, data volume, target calibration), and re-enter Smart Bidding when the conditions support it.
Using MCP Ads to monitor Smart Bidding health
Ask Claude to check your Smart Bidding status: "Review my Google Ads campaigns and check each one's bid strategy status. Flag any in learning phase, any with limited conversion data for the strategy selected, and any where CPA or conversion volume has dropped more than 20% in the last 14 days."
That gives you a Smart Bidding health report across all campaigns without manually opening each one. Claude can also recommend whether to adjust targets, switch strategies, or revert to manual based on the current data.
Doing this with Claude connected to the account
The hard part of Smart Bidding is not choosing a strategy, it is knowing whether the one you chose is actually working — and the UI spreads that answer across the bid strategy report, the conversions page, and the change history. With Google Ads connected through MCP Ads, Claude can pull all three at once.
Prompt 1: "For every campaign on a Smart Bidding strategy, show the strategy type, its target, actual CPA or ROAS over the last 30 days, conversion count, and whether it is still in learning." Claude calls google_ads_smart_bidding_report, which returns the strategy status and target-versus-actual for each campaign, and google_ads_portfolio_bid_strategies for anything on a shared portfolio. What comes back is a table you can scan in ten seconds: which campaigns are hitting target, which are 20 percent over, and which are stuck in learning because a change reset the clock. As an illustrative example, three campaigns at target and one at $85 actual CPA against a $60 target with only 11 conversions immediately tells you which one to look at.
Prompt 2: "Show me the daily conversion trend for that campaign for the last 60 days and overlay every bid strategy or budget change from the change history." This combines google_ads_daily_trend with google_ads_change_history. The result usually explains the CPA problem on its own: a target lowered on the 14th, conversions dropping the next week, and the learning phase restarting. You do not have to reconstruct that story by hand.
Prompt 3: "Raise the tCPA on that campaign from $60 to $72 and leave everything else alone." Claude uses google_ads_update_target_cpa (or google_ads_update_target_roas for value-based strategies). Because this is a write, Claude shows the campaign name, the current target, and the new target and waits for your approval before applying it. Nothing is changed silently, and the same confirmation applies to budget changes through google_ads_update_campaign_budget.
If you would rather not type these each week, the weekly bid optimizer skill runs this exact sequence and hands you the recommendations with the approvals still in your hands.
Where the manual way still wins
The API reports the learning status label and the target, but not the internal auction-time signals the algorithm is weighting or the bid it placed on any given query. Claude can tell you a strategy is underperforming; it cannot tell you why Google bid what it bid. Seasonality adjustments and data exclusions are also easier to set in the UI, where the date picker and the affected-campaign list are visible together.
The decision that should stay human is the target itself. A tCPA is a statement about what a conversion is worth to the business, which depends on margin, sales close rate, and cash flow — none of which live in the ad account. Let Claude find the gap between target and actual; you decide whether the target was right in the first place.
