In short: "Smart Bidding doesn't work" is almost never true: the problem is in measurement, the target, volume, changes or competition. The bid strategy report explains half the cases in its first line (status, latency, average target). And a rule that saves arguments: a high CPC is not a symptom; a sustained high CPA is.
"Smart Bidding does not work" is the most common diagnosis and the least useful. It almost always means something else: measurement, target, volume, changes, competition. This lesson closes the module with the diagnostic method: from the symptom to the cause, with the specific check for each one, and the tool Google gives you for it.
What is the bid strategy report?
Tools → Bidding strategies → click the strategy (or go there from the campaign's "Bid strategy type" column). It shows:
- Status: learning (and why), limited (by budget, by bid limits, by lack of data), active.
- Average target: the weighted target it has actually used over the period (if you changed the target, you will see the blend).
- Performance against the target: real CPA/ROAS vs target, with a chart.
- Conversion lag: how many days your conversions take to come in — the number that defines your evaluation window.
- Top signals: which dimensions it is differentiating bids on.
Read it before drawing any conclusion: half of the "it does not work" cases are explained on the first line (status).
What causes each Smart Bidding symptom?
| Symptom | Likely cause | Check |
|---|---|---|
| Spend collapses well below the budget | An impossible target; or genuinely low demand; or bid limits | Target vs the real 30-day CPA; lost IS (rank) (if it rises, it is the target); limits in the portfolio |
| CPA climbs with no changes from you | New competition; seasonality; measurement (conversions dropping) | Auction insights; the same period last year; the status of the conversion actions |
| CPA climbs after a change | The change reset the learning | Change history; "learning" status; wait 2-3 weeks |
| Permanently "learning" | Frequent changes; or few conversions | Change history; conversions/month (< 30 → portfolio or bridge) |
| Conversions at zero with normal clicks | Broken measurement | Tag Assistant; conversion diagnostics; a real test |
| Lots of conversions, few sales | Badly chosen reference conversions (an L1 action as primary, spam) | Segment by action; compare with the CRM |
| Volume vanishes on one keyword | Smart Bidding decided it converts badly | That keyword's terms; its historic CPA; it may well have been right |
| The average CPC shoots up | Bidding hard in high-probability auctions; it may be correct | Look at the CPA, not the CPC; if the CPA is fine, it is not a problem |
| Spend irregular day to day | Average daily budget (up to 2×); normal | Judge by the week |
| Different performance by device | The algorithm already handles it; your manual adjustment does not apply | Segment; if mobile converts badly, it is the landing page |
What should you check before blaming the algorithm?
- Is the measurement intact? (statuses, diagnostics, a real test)
- Is the target within 20% of the recent real CPA/ROAS?
- Is there volume (30+ conversions/month in the strategy)?
- Have there been changes in the last 2-3 weeks? (change history)
- Has the conversion lag passed since the last change?
- Has the competition changed (Auction insights) or the seasonality (the same period last year)?
- Are bid limits or the budget strangling it ("limited" status)?
- Do the reference conversions mean money? (lesson 5)
- Has the website changed (landing page, speed, form)?
- Are there new exclusions (negatives, locations) that cut the reach?
If all ten are fine and performance is still bad after two complete cycles, then yes: try another strategy (with a bridge) or rethink the campaign.
In what order should you diagnose Smart Bidding?
- Status: learning or limited? → that is your explanation.
- Lag: am I judging incomplete days?
- Average target vs real: is it far off? → the target is badly set.
- Top signals: is it bidding on what you expected? If it is bidding hard on odd queries, look at the terms and the negatives.
💡 Ninja trick: the symptom table is what the Ninja Scripts Agent checks every day in every account: conversions at zero with clicks, CPA jumps, strategies stuck in learning, spend well below budget. And when something trips, the alert comes with the context ("CPA +80% on campaign X since the target change on the 12th") — the diagnosis half done, so all you have to do is decide.
What you should remember
- The bid strategy report explains half of the problems on its first line: status, lag, average target.
- Every symptom has causes outside the algorithm: measurement, target, volume, changes, competition, the website.
- The ten checks before changing strategy.
- A high CPC is not a symptom; a sustained high CPA is.
That is the end of Module 2 of the Intermediate level. In Module 3, Quality Score in earnest: the three versions of Quality Score, the small print and the 30-day plan.