Google Ads Target CPA held spend flat 1pm–7pm while CPL ranged $286–$1,003 by hour
The case for smart bidding includes a specific promise: it takes signals a human can't act on (device, location, time of day, audience, query), and adjusts every bid at auction time accordingly. Ad schedules, the argument goes, are a manual-era relic. The algorithm sees the hour and bids for it.
Here is ten days of hourly performance from an account running a mix of automated and manual bidding.
Hour | Spend | Clicks | Conversions | Cost per lead | Conversion rate |
|---|---|---|---|---|---|
07:00 | $3,259 | 379 | 13.0 | $251 | 3.4% |
08:00 | $3,883 | 437 | 16.3 | $238 | 3.7% |
09:00 | $5,914 | 529 | 32.5 | $182 | 6.1% |
10:00 | $5,472 | 527 | 22.0 | $249 | 4.2% |
11:00 | $5,275 | 501 | 18.0 | $293 | 3.6% |
12:00 | $5,346 | 485 | 15.7 | $341 | 3.2% |
13:00 | $6,058 | 556 | 13.4 | $453 | 2.4% |
14:00 | $5,889 | 522 | 12.4 | $475 | 2.4% |
15:00 | $6,371 | 588 | 14.1 | $453 | 2.4% |
16:00 | $6,523 | 589 | 6.5 | $1,003 | 1.1% |
17:00 | $6,099 | 560 | 21.3 | $286 | 3.8% |
18:00 | $4,995 | 473 | 14.2 | $353 | 3.0% |
19:00 | $6,197 | 621 | 14.4 | $431 | 2.3% |
20:00 | $5,032 | 492 | 9.0 | $559 | 1.8% |
21:00 | $4,429 | 456 | 5.0 | $887 | 1.1% |
Look at the spend column from 13:00 to 19:00. It is almost flat: $5,900 to $6,500 an hour, every hour. Now look at cost per lead across the same window: $286 at 17:00, $1,003 at 16:00. A 3.5× difference in what a lead costs, between adjacent hours, with no difference in what was spent.
09:00 and 16:00 received nearly identical budgets. One produced 32.5 conversions at $182. The other produced 6.5 at $1,003.
Day of week says the same thing
Day | Spend | Conversions | Cost per lead | Conversion rate |
|---|---|---|---|---|
Monday | $5,523 | 26.0 | $212 | 4.6% |
Tuesday | $9,749 | 21.0 | $464 | 2.1% |
Wednesday | $18,673 | 49.0 | $381 | 2.9% |
Thursday | $19,534 | 47.4 | $412 | 2.8% |
Friday | $19,461 | 79.4 | $245 | 4.1% |
Saturday | $6,667 | 14.0 | $477 | 2.0% |
Sunday | $6,227 | 15.0 | $415 | 1.7% |
Saturday and Sunday together: $12,894 for 29 leads. Monday alone: $5,523 for 26. The weekend spent 2.3× as much as Monday to produce roughly the same number of conversions.
What the algorithm is actually doing
Smart bidding does incorporate time of day. That is documented and it is true. But it incorporates it as one signal among many, weighted by how much conversion data exists at each hour, and it is optimising to a target across the whole campaign, not to an equal cost per lead at every hour.
If 16:00 traffic converts at 1.1% and the algorithm has a $150 target, the correct response is to bid much lower at 16:00. The observed spend says it didn't. Either the model hasn't learned the pattern (plausible, several of these campaigns had been on Manual CPC until days earlier), or it has learned it and is bidding down but the auction floor at 16:00 is high enough that it's still spending. Either way, the outcome is the same: budget flowing evenly into hours that return very different things.
Explanation | Implication |
|---|---|
Model hasn't learned hour-of-day yet | Wait, but verify, and know that "learning" can take longer than the account can afford |
Model has learned it but competition at 16:00 is fierce | Bidding down isn't enough; exclusion is the tool |
Hourly conversion attribution is noisy | Some of the swing is measurement, but not a 5.5× swing on 6,500 clicks |
The caveat
Hour-level conversion data is noisier than it looks. Conversions are attributed to the click's hour, and a 16:00 click that converts on Tuesday morning still lands in the 16:00 bucket. Ten days is a short window; 6.5 conversions at 16:00 is a small number. Before acting on the 16:00 figure specifically, you'd want a month.
But the shape is not noise. Morning hours converting well, mid-afternoon converting badly, late evening converting worst, weekends underperforming weekdays. That pattern held across every campaign in the account and matches what the business's own sales team would tell you about when enquiries come in.
What to do
Ad schedules still exist. Under smart bidding, bid adjustments on schedules are ignored for most strategies, but exclusions are not. You can still tell the campaign not to serve from 21:00 to 06:00, or to serve on Saturday and Sunday from a separate campaign with a lower target.
Approach | Compatible with smart bidding | Effect |
|---|---|---|
Schedule bid adjustments (+20% at 09:00) | No, ignored by tCPA / Max Conv | None |
Schedule exclusions (don't serve 21:00–06:00) | Yes | Removes the worst hours entirely |
Separate weekend campaign with its own target | Yes | Lets the algorithm learn weekend behaviour separately |
Portfolio strategy with hour-of-day seasonality | Partially | Advanced; rarely worth it |
Check your own hourly table. If spend is flat and CPL isn't, the algorithm is not doing what the sales pitch said. The schedule is still yours to set.
