Get Your Report for Free!

Machine Learning in AdWords: How & When to Use Smart Bidding

  • 829
  • 12K
Machine Learning in AdWords: How & When to Use Smart Bidding

What a momentous year 2017 was for the evolution of machine learning in PPC.

Google added in-market audiences for search as well as two new bidding strategies (maximize clicks and maximize conversions), introduced predicted click-through rate and optimized ad rotation to AdWords, and launched Google Attribution.

Machine learning is the future of search, without a doubt.

As Google’s supply of data expands, and their AI research continues to make progress, we can only expect the quality of their machine-learning driven features to improve.

The present is a more complicated matter.

Advertisers need to decide between Google’s automation, adopting a manual approach, or using alternative automation to achieve the best results.

There is a whole load of machine learning features to be discussed in relation to this: Dynamic Search Ads, Universal App Campaigns, Google Attribution, Smart Display, Expected CTR – the list goes on.

This article will focus on smart bidding. Here’s how and when to use this generally awesome addition to AdWords.

What Is Smart Bidding?

Smart bidding combines machine learning and contextual signals to optimize bids at auction level. This contextual data ranges from geolocation, time of day, ad creative, and user device to find the conversion opportunity and optimal bid at each auction.

It incorporates billions of these data signals to calculate the likelihood of a conversion, based on the performance targets that have been set.

There are multiple strategies available:

  • Target ROAS
  • Target CPA
  • Enhanced CPC
  • Maximize conversions – the new addition to the family

They each support different marketing goals – there’s no one-size-fits-all bidding strategy, but they all have their own pros and cons, depending on what you need from them.

Why Is Smart Bidding Awesome?

There are loads of reasons. Here are eight (in no particular order):

  1. Looks at search query level rather than keyword level.
  2. Analyzes search contexts to predict likelihood of conversion.
  3. Works for low volume keywords that don’t have enough historical performance.
  4. Adjusts bids for every auction, rather than every hour (maximum in manual).
  5. Incorporates user’s historical behavior to determine bids.
  6. Aggregates performance segments (e.g., geo performance overlaid with device).
  7. Leverages auction-time signals that may be only available to Google (browser, language settings, operation system, app, actual query, ad creative).
  8. Performs cross-signal optimization rather than considering each signal individually.

When to Use Smart Bidding

3 Situations Where Smart Bidding Is Worth Trying

1. When You Can Effectively Align It to Marketing Goals

Not all marketing goals are encompassed by smart bidding – at least for now.

From my experience, about 90 percent of the time, advertisers’ aims tend to align with the Google’s bidding tools. But not always, so make sure.

  • Brand awareness: Increase visibility by using ‘target search page location’, and ‘target outranking share’. This will let you outrank competitors more frequently.
  • Drive more traffic: It’s simply a case of using ‘maximize clicks’.
  • Conversion optimization: Use target CPA and eCPC. Both take into account contextual signals, target CPA and likely conversion rate.
  • Revenue optimization: ROAS, which takes into account the same factors as target CPA and eCPC but also conversion value per click.

2. When Auction Level Signals Are Sufficiently Important

As I said earlier, one of the major strengths of smart bidding is that it incorporates auction level signals that are not included within manual bidding: browser, language settings, operation system, app, actual query, ad creative.

Having access to additional data is never a bad thing, but sometimes there is a trade-off between the benefits of manual and automated bidding.

So, if for example, you have a small account or your campaigns are already performing very well (maybe you are exclusively running brand campaigns), then smart bidding may not be worth the loss of control.

For larger accounts, especially if you are planning to create a portfolio bid strategy (same strategy across multiple linked campaigns), then the benefits of these extra signals can be extremely powerful.

Google reported last year ‘an average 16 percent lift in conversions for advertisers using Target CPA and Target ROAS’, based on their own internal data. From our own accounts, we have seen examples where the uplift has been twice as high.

3. When Manual Isn’t Getting Good Results

This one doesn’t take much explaining, but is worth emphasizing.

A general guiding principle of search is to be experimental, so testing out manual vs automated is a natural part of the job.

If you aren’t getting remarkable results, smart bidding is definitely worth a shot.

By the way, you should always at least start with manual.

Smart bidding won’t work unless there’s some historical account data to build upon, and you may find that manual works very well for your particular needs anyway.

3 Situations Where Smart Bidding Might Be Worth Avoiding

1. When Alternative Automation Is Better

Google has some of the most advanced machine learning technology out there and is easily the industry leader in search advertising.

Still, they can’t think of everything.

Every search account is different, and if you’re a true PPC superhero, you’ll continually be developing your own automation to optimize performance.

For bidding, my company has created some awesome stuff over the years, like real-time bidding and 24/7 bidding (plus a whole load of bespoke tech we sadly can’t share publicly).

There are some decent third-party bid management providers, who are always worth comparing against the incumbent. And there’s a whole sheet-load of open-source scripts here to sift through.

What’s so great about AdWords is that its API facilitates innovation from its users, so get creative!

2. When Marketing Goals Don’t Align with Smart Bidding Strategies

Smart bidding strategies work great when attribution is straightforward. When attribution is more complicated, there is a risk of damaging wider marketing aims by optimizing for the wrong AdWords goals.

Sometimes you have generic, high volume terms which function as a gate to your website. Your returns on these terms are probably not good (CRs are low) but they increase traffic a lot. The problem is that you can’t determine the overall effect of these terms on your business.

User journeys are complicated. It’s hard to give the correct “value” of a touchpoint early in their journey.

Switching to a data-driven attribution model is an imperfect compromise, as there’s often not enough data to know whether a click early in the journey contributed to the conversion at its end – and Google like to be fairly certain about something before claiming it.

This means you have generic terms which from a PPC-metrics perspective are too expensive, but in reality are driving a lot of important traffic to your website – so you don’t want to bid down on them. As you can’t exclude individual terms from a smart bidding strategy (unless you pause them), these terms will likely get pushed down by smart bidding as it deems them unworthy, and therefore it’s best to try and exclude them from the strategy.

3. When Control Is Key

Until AI surpasses general human intelligence, search advertising will always require some degree of human involvement. For some campaigns, human control is needed at a far more granular level.

This might be when an account lacks data, has a fluctuating or limited budget, or requires personally chosen bid weightings on individual keywords.

For example, if you’re running a new campaign with little historical data, you might want to see how it performs with manual bidding first.

Or, if your conversion data is unreliable or incomplete (like in the case of technical or compliance issues), a manual approach is the better option.

Smart bidding strategies can’t account for some of the more unpredictable factors involved in PPC, and this can be especially problematic for campaigns that are underperforming.

Some advertisers operate in a highly unpredictable environment. We had a client in the charity sector, for example, who needed maximum flexibility with their bidding. Being able to respond as quickly as possible to world events was essential to their success.

Smart bidding is fast, but it’s not currently as fast as a semi-automated approach (i.e., a human account manager being supported by custom automation).


I’m sure there are lots more examples of when it’s good and when it’s bad to use smart bidding – just wanted to get the ball rolling on this very 2018 discussion!

Get in touch with me if you have some thoughts, and stay tuned for my next post on this theme where I’ll share my thoughts on Dynamic Search Ads.

More PPC Resources:

Image Credits

Featured Image by Dan Gilbert, March 2018.

CategoryPaid Search

Subscribe to SEJ

Get our daily newsletter from SEJ's Founder Loren Baker about the latest news in the industry!


Daniel Gilbert

CEO at Brainlabs

Daniel Gilbert is founder and CEO of Brainlabs, the smartest digital marketing agency on the planet. As the self-proclaimed superhero ... [Read full bio]

Read the Next Article
Read the Next