Smart Bidding Is Really Speed x Signal Bidding
What is your take on Google’s Smart Bidding? It’s a make-or-break question we ask every Google Ads applicant and one that more and more clients are asking Agencies. The answer should be very much the same; it isn’t perfect, but it’s worth it (testing at least).
To some, it’s a controversial and even personal question—a conversation starter that can have radical responses. Some take it as undermining their experience to imply artificial intelligence could be more efficient. It can elicit survival instincts… my job is to bid, if no more bidding, no more job, no more food, no more life.
What we’ve come to find with new products is there is no one right answer, but there is one right stance and that is one of objectivity. What I mean by that is, there is not a universal truth that Smart Bidding is better than Manual or that Broad Match is better than Phrase Match. But to discover progress requires an open mind that it can exist, that one doesn’t simply have it all figured out for eternity. One may have it figured out to the best of their ability now, but the humility that it requires to say one’s abilities do not defy space and time is oddly polarizing. An interest in achieving the best results as opposed to being the best result, and compounded, that decision over time, over and over again, typically results in superior performance, happier clients, and sought-after talent.
The red flag 🚩 is always the implication that someone is “smarter than Google”
The fact is, even if you were a modern-day Albert Einstein, you couldn’t beat Smart Bidding. Why? It’s not a measure of intelligence or even seasoned intuition; it’s much more black-and-white. It’s about signals + speed, which is why we need to embrace it.
Google’s Machine Learning can analyze 70 million signals (and growing) in 100 milliseconds. 🤯
Scientists at Google say that they have achieved quantum supremacy, a long-awaited milestone in quantum computing. Google in Mountain View, California, says that its quantum computer carried out a specific calculation that is beyond the practical capabilities of regular, ‘classical’ machines1. The same calculation would take even the best classical supercomputer 10,000 years to complete, Google estimates. – Nature.com
Although you have hundreds of data points available in the interface, you can only bid or add bid adjustments on a handful and definitely don’t have it on an individual user level. Sure, you can see how devices and demographics have performed as an aggregate in the context of campaigns, ad groups, keywords, and general ads. But you can’t bid per person based on their unique probability of converting and certainly not per auction for a given person.
And even what you do have, you don’t have it all that fast.
Even if you were a superhuman octopus with 8 arms and could memorize the bid ranges and historical performance for the hundreds or thousands of keywords in your account—let’s pretend that is possible for a minute—with four mice, four keyboards, and a quantum computer… the medium for which you must bid through, the browser interface is not real-time.
If you read further about Google’s data freshness, you’ll find it’s not all that fresh at all. On the surface, it reads like you’ll be lucky if impressions + clicks are within 3 hours of real-time, but if you’re using any of the preferred non-last-click attribution models, it could be more like 15 hours for conversion data.
Google’s Machine Learning will almost always win—but guide it in the right direction
That’s why Google’s Machine Learning will typically win whatever specific metric you are testing it on. These data-rich signals are informed by Google’s access to (at least) 7 different properties, each with 1+ billion users. These include Google, Gmail, YouTube, analytics, android, Google Maps, and Chrome. All this to say, the machine is fed by incomprehensible amounts of data being processed in microseconds. It is the only way to effectively handle infinite permutations.
However, it also has tunnel vision. If you ask it to optimize toward a target CPA, it will do just that, with little regard for your ROAS. And vice versa, don’t expect a reasonable CPA if you ask it to focus on ROAS.
The concept of algorithms used to be really abstract to me, until I watched a documentary that described what they are and engrained this visual concept in my head of these sorting algorithms.
In the context of Google Ads, it’s less about “sorting” by bid, which is why bidding by a keyword + match type is less important these days. Instead, it’s about sorting by probability. And to calculate that probability, you need a lot more data than you have available, and you need to analyze it in real-time… like between the time it takes to hit “search” and the results rendered.
In every algorithm shown, the sort doesn’t just happen perfectly or instantaneously. No matter what, it takes time. There is a “learning phase” to achieve the best result, and it doesn’t happen on the first go but through some trial + error. Eventually, math prevails, and the goal is achieved. The same concepts hold true for Smart Bidding algorithms.
Where human experience comes in here is not too dissimilar to where humans thought their value was previously. With Smart Bidding, you still identify target “bids” (on CPA or ROAS). But, instead of at a keyword level w/ adjustments you do so at an ad group level. You are still only as successful as the performance you achieve, with the campaigns you’ve built, producing the results you’ve optimized for. Your experience is only as diverse as your mind is open. In fact, nothing really changed at all except for maybe the perception shifted from one that sees a Paid Search manager as a bidding octopus buying clicks all day to a… gasp… performance partner?
So in the spirit of performance partnership, we’ve identified some key pillars we’ve experienced to be true to ensure a productive test + launch and hopefully save you a ton of trial and error.
Those key Smart Bidding pillars of success are:
- Start with max clicks or impression share, aggressively with top-of-page bids (high range) that you see in the keyword planner, to develop momentum.
- Don’t bother on budgets < $1,000, unless it’s a brand or remarketing campaign. For prospecting at that budget, you might get a click or two a day so you might as well take a flyer on DSA or Pmax.
- Experiment 50% or less before switching, test + prove.
- Start on the higher side of bids and refine as you go otherwise inertia may be paralyzing, try not to choke it from the beginning and wait for it to build up slowly.
- Don’t be ridiculous. Set realistic CPA targets based on your market. You can check this in the keyword planner or ask your agency to show you. If your top-of-page bids are $50 and the historical conversion rate is 33%, you should allow ~$150 CPA and refine from there. Don’t be unreasonable and set a $50 CPA and expect any volume or quality, Google will just enter the cheaper auctions and generate low-intent leads.
- Don’t make drastic changes if you can avoid it. But if you have to, try to keep it under 20% at a time, otherwise, it goes back into a learning period.
- Upload Customer Lists to give Google more signals to work with.
- Check that your conversion actions count one instead of every, otherwise you are optimizing for duplicates instead of uniques, and your CPA is skewed.
- Ensure some kind of qualifying criteria on call length, or better yet use a CRM integration that only reports back qualified leads.
- If using chat or SMS integrations, use submission action, not start or “click to” whatever.
- Don’t just set bid at campaign level, adjust at ad group levels too.
As we always harp on, there is no universal answer. However, we’ve found success more often with Smart Bidding when following the above guidelines. But, the most important point to progress is to approach with a critical, but open mind.
You can hear more about these concepts on a webinar I joined recently hosted by ServiceTitan. If you have any questions or comments, as always, hit me up at brianna@adleverage.com.
