Online Advertising
Automated bidding is nearly always better if one condition holds. If it does not, it is markedly worse than a steady hand.
The question of the right bidding strategy is usually argued as a matter of belief. It is a matter of data volume.
A learning system needs feedback. As a rule of thumb: 30 measured conversions in 30 days per campaign is the minimum, 50 is comfortable. Below that the system sees too little to tell pattern from chance.
Note that it counts per campaign, not per account. An account with 200 conversions spread across twelve campaigns averages 17 each and sits under the threshold.
| Strategy | Steers on | Needs | Suits |
|---|---|---|---|
| Manual CPC | position | nothing | start, very small accounts |
| Maximise clicks | volume | nothing | building reach, short term |
| Target CPA | cost per conversion | 30 conversions | services, enquiries |
| Target ROAS | revenue per unit spend | 50 valued conversions | retail |
| Maximise conversions | conversion count | 15 to 30 | budget must be fully spent |
| Target impression share | visibility | nothing | brand defence, competitor terms |
A target is an instruction, not a wish. Set a target of £30 per enquiry when actual cost is £80 and you do not get cheaper enquiries, you get almost no delivery. The system simply cannot find auctions it can win at that price.
Four situations keep manual bidding ahead: below 15 conversions a month, where any automation steers in noise; campaigns with no measurable conversion, such as pure brand visibility; the first three weeks of a new account, while no data exists; and strongly fluctuating capacity, such as a trade business with full order books.
That last point gets overlooked. A firm booked until October does not want optimisation for maximum enquiries. It wants visibility at low cost, and a manual strategy with a low cap is the right tool.
If you know of a period when conversion rates multiply, tell the system in advance. A seasonality adjustment exists for exactly that. Without it the automation needs several days to notice the change and just as long to unlearn it afterwards.
Use the tool sparingly, for windows of one to seven days. For longer shifts, adjusting the target is the right move.
After the learning phase three things should be visible: cost per conversion oscillating more tightly around the target, click prices fluctuating more than before, and impressions distributed less evenly across the day. All three are good signs. The system is deliberately paying more when the probability is high and less when it is low. That is what you switched it on for.
Seven to fourteen days, longer at low volume. Results swing widely during it; that is normal and not a reason to intervene.
Technically yes, practically no. Every switch starts a new learning phase. Two switches a month means the campaign never runs normally.
More from Online Advertising
Online Advertising
Cost per click is the most widely misread number in online marketing. Look only at it and you will pay more per customer than the competitor with the pricier click.
Online Advertising
Most accounts grow wild. After two years nobody remembers why a given ad group exists. A structure that is still readable after a hundred changes.
Online Advertising
A campaign type that plays across every channel and shows you almost nothing. For some firms it is the strongest lever available; for others an expensive trap.