#6 Ways to Improve CPA with Smart Bidding

1) Targets

The simplest way to influence smart bidding behaviour, is by actively adjusting CPA (or ROAS) targets.

And the thing to remember with CPA targets is that they aren’t set-and-forget instructions, where you simply choose the CPA you want to achieve and leave the rest to the machine…

Rather they are a flexible tool to nudge the algorithm’s priorities in a certain direction along the volume<->efficiency continuum.

The reason why you might set a target at anything higher than your genuinely ideal level (I mean, a £0.01 CPA would be nice, right?) is that the algorithm has a protective element. If the algorithm can’t reach your ultra-low CPA target, it will protect your spend by throttling itself.

But just how much will it throttle itself? At what point? How much time and how much spend will it tolerate when it can’t bring CPAs down to your target level?

We don’t know exactly how much licence Google gives itself to fall short of your stated target and still continue spending your budget – but we know that it will do so, and the bottom line is: the target doesn’t represent any kind of hard limit. Not even close to it.

So instead, targets should be seen as flexible levers for directional guidance of the algorithm.

If your campaigns are failing to get down to your target CPA – rather than accepting a less ambitious goal, lower your target further, and increase the gap between the actual CPA and the targeted level. 

This puts more pressure on the algorithm to take seriously the fact that it is not being cost effective enough, and to start bidding more cautiously, being more selective about which auctions it enters, and competes for aggressively.

Conversely, when CPA is below your target and yet the campaign is failing to make use of all available impression share, raise the target (again increasing the gap between target and actual CPA) to nudge the algorithm into the expansive approach you’re after, dragging the actual CPA up and – more importantly – pursuing more conversions, more freely.

So target management is the closest optimisation tool we have to old fashioned max CPC bids, and it’s the main lever we have while working ‘with the algorithm’.

Then there are the other methods… when you have to take matters more into your own hands…

With these we need to identify areas of underperformance, intervene in the smart bidding’s work, to cut those segments out of the picture.

In an ideal world, the smart bidding system would identify those segments more reliably, more quickly, and more precisely than we would… but it’s just not that good in practice yet. And while that’s the case, a bit of old fashioned manual intervention is often exactly what’s called for.

2) Search Terms

The search query report is one of the dimensions where you can often see that uneven performance, and deal with it through negatives, but don’t neglect these other  dimensions…

3) Locations

The Locations report is one of the most reliable places to find and act on disparate performance. With any reasonable amount of data, differences in CPA by location will start to show up fairly quickly.

Campaign >
Locations >
Click into your targeted location >
Select your level of granularity and assess performance by ‘matched location’ >
Exclude the locations that are dragging up your CPA

As with all of these variables – choose the unit at a level of granularity appropriate to the volume of data

e.g. when looking under matched locations, if you’re targeting a country, then take ‘counties’ (UK) / ‘states’ (US) as a variable rather than ‘cities’ (unless you’re in super-high volume territory)…

4) Audiences / Demographics…

First add several audiences on Observation for analysis. (Smart bidding has a handle on the interest-based audiences whether you add them or not, but the data will be valuable for us…)

Include low-performance audience in this – because the main lever we have here when using smart bidding, is excluding under-performing audiences.

When it comes to demographics, don’t forget to treat unknowns as a distinct category.
Google generally assigns users to a demographic group based on their Google account settings. This relies on the user being signed in to their account.

Because of this methodology, the ‘unknown’ segment in demographic reports does not represent a random sample of users. Certain groups, and types, of user will be less likely than others to be logged into Google – for whatever reason – and therefore more likely to be categorised as ‘Unknown’.

Being a non-random sample, we would expect to see ‘Unknown’ users deviating from the average when it comes to campaign performance… and typically, if you pay attention to your demographic reports, you will see exactly that.

5) Devices

It’s no surprise that mobile traffic tends to perform quite differently from desktop, with both the user and the site being in obviously different modes between the two.

But when we do see big discrepancies (and we’re using smart bidding) our options are limited to:

• excluding one of the segments – which – with only two serious segments (tablet being invariably small) – is a BOLD move (but sometimes a good one)…

• separating the segments into separate campaigns for separate treatment. I don’t like this structurally, but when the cap fits…

6) Ads

Back to the campaign contents…

But now with RSAs, which are a much more complicated beast than ETAs (we could have done without the enforced switch, but there we are…)

And RSA best practices are still very much up in the air.

Google’s recommendations – and the ‘Ad Strength nudge’ – are for maximising the number of assets and minimising the use of pinning… (and I’ve seen best practice advice from some very reputable sources that mirrors that guidance).

Although RSAs have notoriously underperformed ETAs on conversion rate, the common counter argument is that RSAs gather incremental impressions, negating a like-for-like comparison.

But that assumption is starting to wilt under scrutiny. As the ever-sharp Brad Geddes (founder of Adalysis) says:

And with study after study (Optmyzr, Adalysis, PPC Hero) showing that RSAs fail to beat ETAs – and high-Ad Strength RSAs fail to beat ETA-esque RSAs – on performance… That leaves us with very little reason to favour high-Ad Strength practices at all.

So when it comes to optimising RSAs, don’t be afraid to take back the reins and pin your messaging down, and test it through multiple pinned RSAs or the Ad Variations tool.

p.s.

By failing to make room for landing pages in this list. I’ve made the common PPCer mistake of leaving this crucial piece of the picture as an afterthought…

Because whatever we do on the campaign side – and however well we do it – it can only multiply the effectiveness of what’s already in place in the site / offer / landing page. If that isn’t positive, then multiplying it won’t get us anywhere…

The landing page is the foundation of any success (or other) we have from PPC…

So don’t settle for a poor landing page, and don’t neglect to test!

Share this post

Lastest Posts

Google Ads Updates - Q1 2024 Roundup

No slowdown in the pace of change with Google Ads updates in Q1 2024… Here’s…

Aggregation vs Articulation: Competing Virtues of PPC Management

Three ‘A’s Two concepts have been central to the recent evolution of Google Ads: aggregation…