Sure, you know how to turn the key and sit behind the steering wheel, but do you actually know how the Google Ads Target CPA engine works?

You may have heard the term station car before… It’s the junky car that you drive to and from the train station. You’re not proud of it but you only put a mile or two on it each day.

My dad always had a station car. He still does, in fact, and he calls it the La Bamba. It gets the job done, even if there’s a slight chance that it could go up in flames at any moment.

I’ve seen a lot of La Bamba ad campaigns over the years.

With a better understanding of the bidding algorithms that drive your PPC campaigns, you’ll be able to avoid breakdowns and even reconstruct the engines so that they don’t shake when you push the gas to 80 mph and beyond.

My team and I have spent a lot of time working in this garage. This under the hood look will train you to think differently about how these algorithms work, allowing you to come up with creative solutions to solving your own unique business problems.

This is our manual on proper implementation of Target CPA campaigns. Feel free to take it for a test drive.

In this post, we’ll cover:

• The basics of Google’s Target CPA bidding algorithm
  ‍
• How to determine minimum campaign budgets
  ‍
• How to use conversion rate to determine your break-even point
  ‍
• How to determine your initial Target CPA
  ‍
• How Conversion Window, attribution models, and other conversion settings impact the bidding algorithms
  ‍
• How you think about your campaign goals compared to how Google interprets your goals
  ‍
• The impact of outliers on bidding algorithms
  ‍
• The impact of drastic performance swings on bidding algorithms
  ‍
• How to optimize your Target CPA campaigns
  ‍
• The basics of the Learning Phase in machine learning
  ‍
• The impact of changes to your tCPA targets and/or campaign budgets
  ‍
• How to use all this information to unlock new, profitable growth opportunities
  ‍

Buckle up.

‍

Disclaimer: These concepts have been over-simplified for the sake of a crash-course and likely do not do justice at explaining the brevity and complexity of this bidding algorithm.

My goal here is not that you walk away thinking that 10% of your budget will always be reserved to learning (a number that I am completely speculating on). Rather, my intention is that you will have a better understanding of how and why a portion of your budget is always reserved for learning (even if I cannot report on that specific number) and therefore you will be able to make more informed decisions when attempting to guide the machine.

Ultimately, the handful of levers at your disposal have serious ramifications regarding the success or failure of your smart-bidding campaigns. Misuse or a misunderstanding of any of these features can have disastrous effects.

I hope that you can come away from this article more confident and more capable in utilizing the Target CPA algorithm to help drive success… These tools have had an incredible impact on our client’s campaigns and we’re excited to share our insights with you.

‍

Introduction: Not All Conversions Are Created Equal…

We start with a reminder that advertising is best described as the act of buying customers at a profit.

Customers, or conversions, vary in cost. We can theorize that all probable conversions that could be attained by a given business would create a scatterplot that looks something like this, where the first conversion costs very little to acquire, but the 50th conversion costs exponentially more to acquire:

‍

Let’s pretend that our client is a custom ping pong table manufacturer. They’re a lead generation business that earns $5,000 of profit per sale and has a 10% close rate.

We can therefore afford to pay $500 per conversion if we want to break even.

However, the previous statement is somewhat misleading. It is not that we can afford to pay $500 per conversion, rather, we must average $500 per conversion.

Theoretically, we could afford to pay $4,000 for a single conversion, if we also bought 10 subsequent conversions for $9.99 each.

‍

How we got that number:

($4,900 x 1) + ($9.99 x 10) = $4,999.90

‍

I’ll revisit this example later in the post to prove that it’s not as ridiculous as it might currently seem.

The total addressable market for our ping pong business might actually look like the following, whereby the cheapest customer only costs $75 to acquire, and the most expensive customer costs $925:

‍

That is, there are attainable conversions that could be bought given our unique conditions for profitability. If a competitor of ours has a more efficient Google Ads account or a better website conversion rate, then their addressable market, as it relates to the PPC bidding landscape, will look different.

If there was a major change in the market or to our own conditions for profitability, then these numbers could change completely.

It’s clear that a Target CPA bid strategy is likely going to help us achieve the results we need. But under what conditions?

‍

The Basics of Target CPA Bidding

To start, Google recommends that Target CPA should only be used when you can afford to set your daily budget to at least 10x your Target CPA. They claim that this is the only way to ensure that the system will have enough freedom to run tests and disprove the null hypotheses.

This makes perfect sense when you consider how silly it would be for us to set a budget of, say, $600 per day and expect to drive results.

In this case, Google would recommend that our daily budget be set to $5,000.

However, this isn’t possible for many clients… Not every ping pong table manufacturer can afford a $150K monthly budget out of the gate. (And this is just for one measly search campaign!)

We’ve tested this theory in many accounts and have concluded that it’s not always necessary to set a 10x budget. Volume of clicks (and conversions) is more important than actual spend.

Keep in mind, you can always set your daily budget to 10x your Target CPA, and monitor results on a daily basis. Setting a higher daily budget than what you’re comfortable spending over the course of a month isn’t a commitment to Google to actually spend that amount of money.

More data in less time is a better way to set yourself up for success.

Look at it this way – If you’re a personal injury lawyer seeking a $500 CPA and competing for clicks that cost at least $100 each, Target CPA won't work for you under a limited budget.

However, if you’re a discount furniture manufacturer seeking a $500 CPA and competing for clicks that cost $2 each, you can likely drive results with a less than 10x budget.

Excluding these outlier scenarios, I’d recommend setting that budget to at least 4x your Target CPA.

As a rule of thumb, you should estimate a realistic conversion rate that you can achieve, then factor in the average CPC for the traffic you are targeting, and then determine if you can even afford to receive multiple conversions in a given day.

‍

Conversion Count Minimums

If you cannot afford a budget at least 4x your tCPA, I would recommend starting with Maximize Conversions. Under a limited budget, the Maximize Conversions bidding algorithm will be forced to optimize for the cheapest possible conversions.

Regardless of your smart bidding strategy, you should still be realistic about what you expect a conversion to cost and whether you can afford to buy multiple conversions per day.

That is, if you’re seeking a $500 CPA, you are setting yourself up for failure if your daily budget is $500 or less for a given campaign.

In this event, I would suggest to the client that they outline what they planned to spend over the next six months and spend it all in three months (or sooner), so that you can run an appropriate test.

One conversion per day, or less than one conversion per day on average, will not give the algorithms the data necessary to drive results and scale.

The algorithm would essentially be throwing darts, blindfolded, at a wall and hoping something occasionally lands near bullseye.

There is no specific number of conversions that are required on a daily or weekly basis to effectively run Target CPA campaigns. My own personal rule of thumb is that I’d like to see at least 30 conversions per week before completely turning over to Target CPA.

(This varies by account… Don’t hold me to that number, but at least this gives you somewhere to start.)

Thankfully, Google’s smart-bidding algorithms consider signals from your entire account when setting bid prices. That is, conversion data obtained from Campaign A will help guide the algorithm to confidently bidding on behalf of Campaign B.

Therefore, you don’t necessarily need 30 conversions per week per campaign, to effectively run Target CPA campaigns, so long as you’re earning 30 conversions per week from smart campaigns across your account.

However, each campaign should still be able to afford at least one conversion per day (7 per week).

Regarding your actual CPA goals: As an intelligent PPC manager, you may be able to show your client that their actual CPA goals too aggressive. 

Maybe you can help them see that the LTV of their customers is more than what they thought. 

Maybe you can help your client see the benefits of breaking even and forgoing profit for the time being, effectively increasing your Target CPA.

Maybe you can help your client optimize their overall business, increasing revenue per customer and subsequently increasing Target CPA. 

In other words, don’t be in the habit of assuming your client’s (or your own) Target CPA is intransigent. If we had more data on hand, we could likely find that our ping pong table client could afford to spend beyond $500 per conversion, but that is a can of worms that I'll ignore for now.

‍

Factoring in Conversion Rate to Determine Break-Even Point

Let’s assume that the average market price for ping pong table keywords is somewhere near $5 per click. We would need a 1% conversion rate to break even.

‍

How we got that number:

$500 CPA / $5 per click = 100 clicks per conversion

1 conversion / 100 clicks = 1% conversion rate

‍

If we cross-reference this data with historical conversion rates that we’ve seen in Google Analytics and similar accounts, we believe our conversion rate will be better than that. We estimate our conversion rate to be at least 1.5%, and we are actively seeking to improve our site conversion rate up to 2% over the next three months.

We can estimate that a budget of 4.5x our break-even CPA, $2,500, is an appropriate starting daily budget. We can estimate that we’ll achieve around 500 clicks per day.

At 500 clicks per day and 1.5% conversion rate, we should generate an average of 7.5 conversions per day, or 52.5 per week.

Again, this is our baseline that helps us determine a daily budget. This may change as we determine our specific goals and other variables that I’ll outline below.

The key takeaway from this exercise is that we’ve determined that our daily budget should be at least 4x our tCPA. Now we have to find out what those targets will be…

‍

Determining Your Starting CPA Targets

Here’s the next issue: A $500 CPA would allow us to break even. We don’t want to just break even… we want to earn a bit of profit.

So how should I determine my Target CPA? We know that ideally retain a CPA below $500, but how much profit per conversion should we aim for from the onset? 

We used historic conversion rates on the site to predict our baseline budgets and break-even point, but ultimately, smart bidding should optimize for higher conversion rates over time.

While we want to drive as much profit as possible, we also want to be reasonable about what we can achieve in the short term. If we set our targets to be too conservative from the onset, we might throttle the total amount of learning and ultimately results that we can achieve.

Therefore, choosing a realistic CPA target is important.

Here's a few things to consider:

1. If this is an entirely new campaign, I'd recommend using your own estimation of conversion rate on the site, while factoring in knowledge gained from other areas of the account and/or similar accounts.
   
   But here’s the good news: As mentioned above, the algorithm now considers signals (data) collected from other campaigns.
   
   If no historical data is available (if you just created your Google Ads account yesterday…), Google will pull signals from accounts competing in similar auctions.
   
   That is, if you’re an auto parts dealer bidding on the keyword thule bicycle rack, the tCPA algorithm will consider signals gleaned from the accounts of your competitors when determining how to set the appropriate bid.
   ‍
2. If this is a new campaign in an account that has historical data, Google will consider signals gathered from other campaigns inside of your account. This is a relatively new addition that is extremely helpful.
   ‍
3. If this is an existing campaign that has been running on manual bidding, I would first check to see Googles recommendation for your Target CPA. Even in this scenario, the system is still aware of what it could potentially achieve given existing data.
   
   I’m often pleasantly surprised when I uncover an extremely low suggested tCPA.
   ‍
4. If this is an existing campaign, and the results have been nowhere near profitability, then it’s even more important to take your budget into consideration when setting your tCPA.

‍

You might first think that if the campaign has been returning at an $800 average CPA, and your goal is to be below $500, that it would be best to gradually bring down your tCPA settings until you achieve profitability.

However, if budget is limited, you would be better off making a drastic change with a much lower CPA.

If you can only set aside $2,000/day for this campaign and apply this one-step-at-a-time strategy, you’ll never allow the system enough freedom to run the appropriate tests and conclude hypotheses with statistical significance.

A tCPA of $750, for example, will give you less than 3x your tCPA in daily budget. At most, this will yield 18 conversions in a given week.

If I need to drastically move the needle to achieve profitability, I am not comfortable with the fact that the algorithm only has 18 conversions per week to optimize based off of.

Instead, a drastic tCPA reduction of $200 would allow the freedom for 10x tCPA to daily budget ratio. Sure, this will likely shock the system and it will take a few days, if not weeks to recover, but the algorithm will be much more confident with the results it drives.

In the event that this doesn’t ultimately drive the results you need, you’ll need to be more creative with your bid strategies, budgets, and possibly which actions you’re counting in your conversions column.

Or you can just hire AdVenture Media. I mean… figured I’d throw it out there.

‍

The Importance of the Conversion Window

Conversion Window is an often-overlooked component of the conversion settings, but equally as important as the attribution model you select.

(Attribution Modeling Hint: Never choose Last-Click attribution when creating your Google Ads conversion actions. Go with Data-Driven if available, Position Based if you average 2.5+ clicks to conversion, else Time Decay).

Conversion Window is the length of time after an ad click that any conversions will be attributed back to said ad click.

The default conversion window in Google Ads is 30 days. Under this scenario, if a user clicks on your ad on the first of the month but returns on the 29th (even without clicking on a second Google Ad), a conversion will be attributed back to the original ad click.

It’s a common practice to extend the conversion window to 90 days so that you capture as much conversion data as possible (it also helps the PPC Manager receive more credit…)

This isn’t exactly a nefarious practice. Clients with a longer lead cycle benefit from an elongated conversion window. Analyze your Time Lag report to help make this decision on your own.

If a significant portion of your revenue comes at least 12 days after first click, then you would be doing yourself and your client a disservice with a 30-day conversion window. Smart bidding algorithms will be ignoring conversion data that should be factored into testing.

However, an inflated conversion window is detrimental to accounts that do not need it. Here’s why:

The Target CPA bidding algorithm is attempting to meet your average tCPA goals over the course of your conversion window.

If your conversion window is set to 30 days and your tCPA is $500, Google's algorithm will optimize with the goal of averaging a $500 CPA (or lower) over a 30-day period.

If you have a bad week where your average CPAs are trending toward $800, the algorithm will become much more conservative. You will likely see lower spend and fewer overall conversions as the algorithm conserves your budget to bid in auctions where there is extremely high confidence that an ad click will result in a conversion.

Under a 90-day conversion window, the algorithm will pace out its tests to achieve that goal over three months. Every ad click technically starts a new conversion window.

That is, the algorithm does not optimize to meet your goals by the end the quarter, for example. An ad click that takes place is factored into a 90 day window from now until 90 days from now, whereas an ad click tomorrow is in a slightly different conversion window.

But either way, the algorithm will pick up on overall trends in performance. If you are trending under your performance goals, Google will throttle your spend and you will likely see a decline in the quantity of conversions you acquire. Similarly, if you are trending ahead of your goals, the reverse is likely to take place.

Second Attribution Modeling Hint: Ecommerce typically has longer lead time than Lead Gen. Lead Gen requires much less commitment (a conversion is often a form fill, as opposed to ecommerce where a conversion includes handing over credit card information), so plan accordingly.

This might not seem like a huge deal… If you want to pay an average of $500/conversion, and that number is likely not changing for the next year, then why would it matter if you’re optimizing for a 30-day or 90-day window?

In fact, wouldn’t the system benefit from a longer attribution window, given the incremental conversion data that would come over the next two months, that the system can now optimize with?

No. For two reasons: Outliers and dramatic swings in performance.

‍

The Impact of Outliers on Google’s Smart Bidding Algorithms

Outliers can affect the algorithms in a negative way (especially if you are not using the Data-Driven attribution setting).

For ecommerce companies, it is extremely common (and also logical) that larger orders have a longer lead time. Smaller orders (i.e. conversions with less conversion value) require less financial commitment, and therefore convert at a faster pace. This is illustrated by the disparity in conversions-to-revenue in the ecommerce account Time Lag report below:

‍

As you’ll see at the top of the chart, 17.5% of conversions (sales) are attributed to ad clicks that were within a one-day window of the first ad click. This represents conversions where a user clicked on an ad and converted within 24 hours.

However, only 16% of conversion value (revenue) is attributed to these instances.

Conversion Percentage > Conversion Value Percentage

The rest of the data represents the amount of time that passed between a user first clicking on an ad and ultimately converting.

The bottom of the graph represents that 34.5% of sales take place at least 12 days after the first ad click, but these conversions represent 36% of total revenue.

Conversion Percentage < Conversion Value Percentage

The difference in these two numbers proves that larger orders take longer to convert. But it is very possible (and can be proven by digging into your actual sales reports) that outliers are driving the numbers on the polar ends of this report.

It is common practice in statistics to remove outliers from your analyses. However, by dragging out your conversion window and not selecting the correct attribution model settings, you are indirectly telling Google that you care about those outliers just as much as you care about every other customer.

Put simply: I would rather Google focus its machine learning efforts on going after my average customer than being distracted by potential outliers.

And if you adapt the same approach you will drive better results.

Third Attribution Modeling Hint: Position Based attribution modeling will place much more emphasis on outliers. In the event where one user clicked three ads over the course of three months and produced $5,000 in profit, a Position Based attribution model will reward $2,000 in value (40%) to the first click, even if it took place 89 days prior. This might be correct and effective for some businesses, but definitely not all.

‍

The Impact of Performance Swings on Google’s Smart Bidding Algorithms

Under a 90-day conversion window, a breakout performance during a Memorial Day sale can still be impacting algorithm’s decision-making process throughout the end of July.

Remember: The algorithm’s goal is to meet your average CPA goals over the course of the conversion window. Your conversion window does not start or end on any particular day; rather, Google attempts to average this over a rolling 90-day period.

Recall my original example from the top of this post, where we had a goal of a $500 CPA and achieved that goal with one conversion that costed $4,900, and 10 conversions that costed $9.99. This is an example of a dramatic swing that would average out to meet our $500 CPA goal…

Let’s assume that you set a tCPA of $500 and a daily budget of $2,500. You are basically telling Google that, for the next 90 days, I want you to buy me 450 conversions with a total budget of $225,000.

How we got that number:

$2,500 daily budget / $500 CPA = 5 conversions per day

5 conversions per day * 90 days = 450 conversions

What you see is a daily budget with a $500 CPA target. What Google sees is a goal of spending $225,000 to earn 450 conversions. This is an important distinction, because Google is incentivized to spend that entire budget, given the parameters that you’ve set.

Plotting this out over a 90 span would look like the following:

‍

… and so on, until you reach day 90 where your remaining budget is $0 and your total conversions is at 450.

Graphically, it would look like the following, with the constant blue line representing daily average CPA and the rising green line representing total conversions earned over this period:

‍

Reality Check: How Performance Actually Plays Out

Let’s assume that demand for ping pong tables was so high over Memorial Day weekend that you earned 50 conversions over a three-day period. Actual CPA over this period averages to just $150!

‍

How we got that number:

$2,500 daily budget * 3 = $7,500

$7,500 / 50 conversions = $150 CPA

‍

This is great!

However, this might have negative implications for how the algorithm behaves for the next 87 days. Google did a great job of buying you conversions far below your target, which means Google has earned the right to dedicate more of its budget toward learning (more on this later) and be much more experimental overall.

In short, Google has earned the right to drive worse results, unless you say otherwise. When you log into your account, you see that the Target CPA is still $500, but the algorithm sees an adjusted CPA Target, given the conversion window parameters.

‍

After the first day of strong performance (15 conversions at $166.67 CPA), the backend of Google now has an Adjusted Target CPA of $511.49.

This does not mean that automatically, Google will only deliver conversions at an average cost of $511.49. Rather, this means that Google can now spend up to $511.49 for each day, for the remainder of the month, and still net out at an average of a $500 CPA over a 90-day period.

The following two days of Memorial Day yield similarly strong results, driving that Adjusted Target CPA even higher. Keep in mind that you, as the advertiser, still think that the daily Target CPA is $500.

All things equal, after Memorial Day Weekend, Google now only has to average out 4.6 conversions for the remainder of the 90-day period to meet your goal. This won’t necessarily be the case, as there are tons of variables at play, but here’s how this scenario could in fact play out:

‍

Sure, over this 90 day stretch you can still average out to $500 per conversion, but the last 30 days look like this:

‍

Total spent: $75,000

Total conversions: 137.93

Average CPA: $543.75

‍

It’s not fun attempting to explain to your client that performance probably suffered because everyone was on vacation for the 4th of July and not shopping for ping pong tables…

‍

Final Point Regarding Performance Swings

You may have realized that even under a 30-day conversion window, the strong performance on Memorial Day would possibly deter performance for the next 27 days. This is also possible.

To avoid this, it’s often helpful to raise your Target CPA for the three days after Memorial, to $575 for example, to force the system to become more conservative. You can change this back to your original $500 after 3 or 4 days.

This same methodology should be applied to Black Friday / Cyber Monday, and other times of the year where dramatic swings, positive or negative, are expected.

‍

Optimizing Target CPA Campaigns

Back to the ping pong table example…

Considering all these variables including the inner workings of the tCPA algorithm and the specifics of the custom ping pong table market, we can summarize our main takeaways as:

• Based on our own profitability goals, we’ve determined that our actual Target CPA should be $400
  ‍
• Based on the average we expect to pay-per-click, our daily budget should be at least 4.5x our tCPA (So, at least $1,800)
  ‍
• Based on our understanding of the ping pong table market, we expect the average customer to convert within 7 days of first ad click. That is, if a user does not convert into a lead with 7 days, it is very unlikely that they will ever convert (Keep in mind, lead-gen conversion cycles are almost always shorter than ecommerce).

‍

Therefore, our campaign settings become:

Conversion Window: 7 Days

Attribution Model: Data-Driven

Target CPA: $400

Daily Budget: $1,800

‍

We launch our campaign and hope for the best. Seven days later we evaluate the results. It is important to wait until after your conversion window has completed before making conclusive decisions about performance.

The daily conversions and actual CPA are measured below:

‍

The campaign got off to a rough start, with just 2 conversions for a $900 CPA after Day 1, but steadily improved as the algorithm exited the learning phase and began making more confident decisions regarding expected conversion rate for Google Ad auctions.

On the final day of the week we earned 7.5 conversions at an astonishing $240 CPA! The totals for the entire conversion window results as such:

‍

Total spend: $12,600

Total Conversions: 31.5

Average CPA: $400

‍

Given a break-even CPA of $500, we’ve earned $100 in profit per conversion, so:

‍

Total 7 Day Profit: $3,150

‍

The campaign appeared to be spending nearly all its budget each day, so an increased daily budget would likely result in additional spend and incremental conversions.

You take all this information and send it over to the client.

Her response includes the gif of Michael Scott and Dwight raising the roof, and a note that they’re willing to spend $150K/month on this campaign, which would more than double your current spend levels.

Adjustments to the budget and tCPA would help you scale spend, but you must first consider the learning phase as part of this process.

(Side note: It is extremely rare that you will want to set a 7-day conversion window, even for lead gen accounts. However, this example simplifies the concepts in a much easier to understand way…)

‍

The Learning Phase

The learning phase is the period of time where a machine learning algorithm is gathering data from initial tests. When a significant change has been presented to an algorithm, the machine must test these new variables until it learns about predicted outcomes with statistical significance.

Google’s DeepMind documented the process of a learning phase by recording an instance where a machine learned how to play and master Brickbreaker on Atari. The machine was not given any previous rules of the game; rather, it was just instructed to find a way to break all the bricks in the game.

The paddle moved somewhat randomly at first but eventually learned that the pixilated ball would remain in play if it made contact with the paddle, increasing the likelihood that more bricks would be broken. After just 120 minutes of learning, the machine mastered the game.

Most digital advertisers are familiar with this concept, but many do not realize that a portion of a Target CPA budget will always be dedicated to learning, even after you’ve exited the original learning phase.

The Always-Be-Learning construct is a fundamental principle of machine learning and is essential to continued improvement and optimization.

Once you’ve exited the learning phase, we can technically breakdown your Target CPA budget into two components: The percent of the budget dedicated to confidence, and the percent of the budget dedicated toward learning.

Under confidence, the tCPA algorithm can manage your budget in a way that it reasonably believes will help deliver the results that you’ve outlined. The algorithm is confident that it understands the primary signals to consider or ignore as part of the bidding process.

However, the small percentage of budget dedicated toward learning is always looking beyond confidence to test new variables and hypotheses.

For example, a confident tCPA algorithm might not currently be evaluating weather patterns as part of its process; however, it may begin testing these signals to help determine whether colder temperatures play a role in the demand for ping pong tables.

The algorithm might similarly be testing to see if CEOs of tech firms that are shopping for commercial real estate are more likely to purchase a custom ping pong table, and whether they are more likely to be influenced by an early-morning banner ad or an evening YouTube ad.

It’s all worth testing, so long as the advertiser still meets their defined CPA goals.

We don’t know for sure what percentage of our budget is dedicated toward learning, and I am certain that it changes all the time and is unique for every advertiser. However, let’s assume for a minute that at least 10% of our budget is reserved for learning.

In this scenario, the tCPA algorithm will need to adjust its actual CPA goals to be 10% more conservative than the nominal goal of the advertiser.

That is, if our tCPA is $500 and we set daily budget of $2,500, 10% of which will be reserved for tests that might yield 0 results, the algorithm needs to adjust and overcompensate for this potential loss so that it averages out to $500 per conversion at the end of the conversion window.

‍

Looking at it mathematically, we get:

Daily Budget: $2,500

Daily Budget Dedicated to Confidence (90% of daily budget): $2,250

Daily Budget Dedicated to Learning (10% of daily budget): $250

30 Day Total Budget (assuming 30-day conversion window): $75,000

30 Day Conversion Goal: 150

30 Day Confidence Budget: $67,500

Adjusted Target CPA: $450

‍

If Google spends 100% of your budget over this 30-day window, and if every penny invested toward learning is unsuccessful in driving incremental conversions, then Google would need to make up 150 conversions from its confidence budget in order to meet your end goals.

What you see is a $500 tCPA, but Google would be viewing that as a $450 tCPA.

Ultimately, you need not worry about Google’s interpretation of your Target CPA so long as you are reaching your goals.

‍

The Impact of Increased Budgets & tCPA Adjustments

Any major adjustments to your budgets or Target CPA parameters will have an impact on your learning-to-confidence ratio.

Let’s say, in the previous example, we took our $2,500 daily budget and doubled it to $5,000. Up until this point, Google has been spending $2,250 with confidence, and additional budget won’t necessarily increase that number right away.

In this case, you go from a 90/10 confidence-learning-ratio to a 45/55 confidence-to-learning ratio, where more than 50% of your budget is now in the learning phase.

A massive increase in your Target CPA will cause a similar impact. Either way, your short-term results will likely take a huge hit…

My colleagues at both Google and Facebook have stated that budget changes up to 15% will not drastically impact the learning phase. While this isn’t documented anywhere, Facebook does allude to this in one of their support docs.

(Facebook is much more transparent about the inner workings of their algorithms, which is one of the reasons why this entire article focuses on Google as opposed to Facebook. However, the core principles that make up the tCPA algorithm can help you better understand all machine learning algorithms in the digital advertising space, regardless of the platform or specific strategy.)

Any changes you make to your Target CPA campaign, especially if they are greater than 15%, should go untouched for at least 14 days. Just don’t do it. The algorithm takes time to work itself out.

I cannot promise that you will see the results that you are seeking within 14 days or within your conversion window, but if you are spending thousands of dollars on a test and making decisions about the outcome before the results are conclusive, then you have effectively wasted every single dollar dedicated toward this test.

If your goal is to scale your campaign but you are afraid to shock the system, here’s my recommendation: Increase both your Target CPA and your daily budget, but increase your daily budget by a greater margin than you increase your Target CPA.

Here’s why: Think back to my original example of how Google interprets your goals. You say, “I want to spend $2,500 per day and pay $500 per conversion,” but what Google hears is “OK Google, I’ll let you spend $75,000 over the next 30 days if you earn 150 conversions.”

If you double your budget without adjusting your Target CPA, you are now telling Google, “You can spend $150,000 over the next 30 days if you earn 300 conversions.”

300 conversions might be hard to come by in the ping pong table market. It will likely take more time and more money to reach a point where Google is even on the right path of confidence.

So instead, let’s raise our budget by 30% (to $3,250/day) and our Target CPA by 15% (to $575 tCPA).

Here’s what Google now sees: “You can spend $97,500 over the next 30 days if you earn 169.5 conversions.

That seems like a much more reasonable goal. As such, the tCPA algorithm will likely be able to exit the learning phase sooner and be on pace to achieve that goal in less time.

Seems like a good deal to me.

‍

But wait… In this hypothetical example, didn’t you say that the client’s breakeven CPA is $500? Why are you suggesting we raise the Target CPA beyond that??

First, the tCPA numbers you set in your account are oftentimes arbitrary. That is, it’s very common to see tCPA campaigns averaging CPAs far below the target. In that event, an increase in tCPA is nothing more than a signal that says “Good job, Google, you can be more aggressive now and go after more expensive conversions. I’m willing to pay relatively more per conversion if you can earn me more conversions.”

Here's an example from one of our lead gen accounts where this has taken place. Below is a 14-day snapshot of a Target CPA campaign, compared tothe previous 14 days. In the previous period (July 1 - July 14), the Target CPA was set to $300.

Over those 14 days, we spent $6,700 on 43.60 conversions with an average CPA of $240.

On July 15 (the first day of the current 14-day period shown), the Target CPA was raised from $300 to $350. Over that time, spend increased 62%, conversions increased 56%, and CPA only increased by 3.9% to $249 - Still far below our original CPA Target of $300.

‍

‍

In addition to allowing the algorithm to be more aggressive, an increase in your Target CPA has an additional positive impact: Increased budget dedicated toward learning.

‍

Why It’s Important to Dedicate Budget Toward Learning

Even once you exit the initial learning phase under these new settings, you are still dedicating nearly 10K/month into learning experiments (assuming a 10% learning budget, which again, is a number that I am speculating on for the sake of this example).

This additional 10K per month in learning might in fact be exactly what you needed to uncover the breakthrough that completely changes your market potential.

Compare this to the impact of a charitable donation. If you care deeply about helping individuals that suffer from a rare disease, where would your donation best be spent? Should you support organizations that help families cover the medical expenses associated with this illness, or should you instead support organizations that perform research, with the goal of developing better treatments, drugs, and ultimately a cure for this disease over the next 10-20 years?

There’s no correct answer to this question, but many would say that it’s important to focus on both… Help those that are currently affected, while also seeking innovative breakthroughs that will have a larger, long-term impact.

Google has taken the ethics out of this question and decides this for us… As an advertiser on Google, a portion of your budget will be spent toward driving short term results, and the rest will be dedicated toward uncovering major breakthroughs.

Perhaps for our ping pong table client, thanks to the increase in learning budget, Google was finally able to test a hypothesis that a growing segment of the population: Married couples that share an interest in wine, and are also parents of teenagers, and also have a basement; have the highest conversion rate compared to any other audience of people that are in the market for ping pong tables.

“Just buy the damn table so that we can enjoy this Cab Sav and watch Stranger Things in peace!”

While this may represent a small segment of the market, a confident algorithm would ensure that you bid aggressively in every auction where these conditions are met.

‍

Realized Growth: Unlocked Market Potential

This sort of breakthrough is the whole purpose of machine learning. With this knowledge (that is now added as part of the confidence arsenal and improves your performance by 20%), you have unlocked an entirely new combination of potential customers and the costs required to acquire them.

The chart below illustrates our original market trend line in blue. The orange trend line represents a 20% shift, also allowing for an additional two customers that can now be acquired under our new conditions.

‍

Our original (blue) market conditions earned 10 conversions at an average CPA of $500… Breaking even on our advertising efforts.

Our new (orange) market conditions can earn 12 conversions at an average CPA of $445… generating $660 in profit.

Once we re-evaluate these results, taking a full conversion-window into consideration, we can continue to optimize this campaign for even better results, or use these profits to invest in completely new initiatives.

For further reading on Target CPA and other Google Smart Bidding algorithms, see our post, “A Guide to Understanding Google’s Automated Bidding Strategies.”

‍