Smart Bidding (aka Automated Bidding Algorithms) is touted as an intelligent way to help manage bids on ad accounts. It is purported to work better than human bidding, save tremendous time, and achieve greater financial results. Machine learning will save us from a life of drudgery and thinking for ourselves!
But there are tremendous downsides to using Smart Bidding algorithms when you use them in the wrong situation or don’t understand what they are really optimizing for. They all have strong biases that you need to understand. In this article, we're going to dive into the main smart bidding methods, and look at where they fall short, to help you avoid making mistakes that will drive up your costs and reduce profit.
First, I’m going to take a certain perspective.
I'm going to assume that we're looking at smart bidding algorithms from the perspective of an ecommerce merchant that wants to maximize their total profit. If you're operating a different kind of business, some of the assumptions I'm going to make here won't apply to you, so you'll have to think about things just a bit more in order to consider the best course for you.
Why am I taking this perspective? Simply because Psyberware is primarily in the business of driving profitable growth for ecommerce merchants through digital advertising. While some of our clients also have other goals, our default is driving profit to the bottom line and doing it with clients who run ecommerce sites.
And, for the sake of argument, I'm not going to give evidence about whether a particular smart bidding method actually works or not. Let's just assume that they all work very well. Let's judge them based on the available documentation and assume they actually do what they say they will. My reasoning here is that if they do what they say they will, but still aren't appropriate, then there really isn’t any need to evaluate what their performance is relative to non-artificial intelligence methods.
That's it. Let's dive in!
Target Impression Share Bidding
This bidding algorithm allows you to set a goal such as…
“I want my ad to show at the absolute top of the Search Engine Result Page (SERP) 65% of the time.”
It will then make dynamic changes to your bids in order to achieve this result.
For an ecommerce merchant, where the real goal is to make sales at a cost that is profitable, this is problematic. Given all the targeting methods available in Google Ads, Search Impression Share is not a great metric for optimizing. If you have very precise exact match keywords, then the theoretical goal of 100% search impression share may be useful, since that represents all available searches for a desirable keyword. However, if you are using broader keyword targeting or dynamic search ads, the more you bid, the more irrelevant "related" terms will match.
The other factor at work is that valuable keywords will tend to be bid up higher by competitors - those keywords aren't just valuable to you. If you try to maximize search impression share, Google will focus on cheaper keywords that are less relevant and less likely to lead to sales. You will get more search impression share, as claimed, but that is not what you need to maximize profit.
The only time you would want to use it is when you don't care about profit, and you just want to keep certain ads in top slots to maintain market share or push competitors out of top spots. Since it will waste money though, this is not a sustainable business practice, and should only be used with very precise keyword targeting.
For ecommerce merchants focusing on profit, Target Impression Share bidding is almost never appropriate.
Maximize Clicks Bidding
So ok, we don't want to maximize impression share, but what about maximizing clicks? Nope!
This bidding method has almost exactly the same issue as the previous one. In order to get more clicks, Google is going to shift focus away from high-cost, high-value clicks toward those that are cheaper. There's a reason they're cheaper. They don't lead to conversions, so your competitors don't bid on them. With less competition, the prices are lower for those terms.
Another perspective on clicks is that you actually want the fewest clicks possible. Why? Because this is PPC advertising: PAY-PER-CLICK. You only pay when you get a click. Most clicks don't lead to sales. Wouldn't it be great to not have all those clicks that don't lead to sales? Why would you want more clicks? Ideally, you want fewer clicks, but the right clicks.
Maximizing clicks is going to tend to get you less valuable clicks, and clicks for irrelevant search terms that match your broad match keywords and dynamic search ads.
The Maximize Clicks Bidding method also has another unwanted effect. If you read the documentation, this method will "get as many clicks as possible within your budget". Put another way, this method will spend all available budget, always. Once it has gotten all the clicks that are valuable, it'll just keep spending what's left of the budget on less valuable clicks until that budget us used up. On normal campaigns, a budget is set to limit the amount that's spent. On a campaign using Maximize Clicks Bidding, the budget is set in order to specific precisely how much to spend. This nuance matters a great deal. There is no mechanism for this bidding method to take any feedback on what clicks are more valuable and stop spending at the right level or to make any intelligent decisions about the tradeoffs of spend versus value. There is simply no way to steer this towards improvements in profitability.
Ecommerce merchants should almost never use Maximize Clicks Bidding.
Maximize Conversions Bidding
If we target maximizing conversions, at least we're actually focusing on the thing that we want to achieve, unlike targeting clicks or impression share. But… not really.
The thing is, we don't actually want to achieve the most conversions possible. We want profit. Maximizing profit means that we need to optimize the tradeoffs between cost and the gross profit margin we're bringing in on each sale. This method does allow us to get more conversions for a given budget, assuming it works as stated. But it doesn't take into account anything on the other side of that equation. It doesn't understand revenue, margins, and ultimate profits.
So just think about what this bidding method is going to focus on. Google wants to be able to report to you that it got a lot of conversions. Success! This is going to lead to decisions that favor cheaper conversions, and more of them. This bidding method will skew away from more expense, and potentially more profitable conversions. And since it has no idea of the margins, this bidding method will happily spend more per conversion than the actual revenue generated if you have a product mix that includes inexpensive items.
Like the other "Maximize X" bidding methods, Maximize Conversions bidding will spend all available budget, even if it isn't getting additional conversions. We have seen a number of cases where merchants increased budgets, attempting to get more conversions, and this strategy happily spent all the additional budget, but with no additional conversions at all. Basically, all the conversions available given the current configuration and targeting were already being achieved, and as the budget was increased, the algorithm happily spent it all on increasingly wild experiments with irrelevant "related" searches that stood no chance of working.
This is not a reasonable business objective, except in a very limited use case. Perhaps you have a selection of products all priced the same with exactly the same margins? Then the problems with this bidding method won't matter. It won't be able to optimize for lower-profit items that kill your profits. But unless everything you sell has the same margins, or you have some other really compelling reason that you've really thought through, avoid this bidding method.
Target CPA Bidding
What about Target CPA? This method allows goals such as…
“I want to achieve as many conversions as possible at $10 per conversion.”
If it can achieve a consistent advertising cost of sale, then perhaps this can be useful for ecommerce merchants. But there are several key problems here that kill ecommerce profit.
It's very unlikely that all of your products have the same average order values and margins. If you’re selling some $20 products with 50% gross profit margins, a $10 cost of sale will destroy all of your profit on the sale. Perhaps you have some other products at $200 with 30% margins, though. In that case, however, you might be willing to spend more in order to get more sales. And perhaps you've got some products that are often bought bundled with some up-sells, where the gross margins are only $5 on each product, but people tend to buy a dozen at a time. A flat target cost per acquisition just doesn't make sense for any ecommerce merchant with a large portfolio of products at different price points, margins, and average order values.
In these cases, Google is going to learn that searches leading to conversions on the more profitable products cost more (because competitors are bidding them up), and those will be avoided. Meanwhile, your cheaper products with low margins will get a disproportionate amount of attention, because those conversions are cheaper to get, and they help achieve the CPA target.
Another issue here is that the Target CPA goal is based on an average cost. With a $10 Target CPA, you will get some conversions for $1, but some will cost $100. They will all be averaged together and reported in aggregate. Any bidding strategy that's built around average goals will have this characteristic. In reality, perhaps you're willing to spend up to $10 on a sale, but you really don't want to spend more than that. Given that a $10 limit would get conversions cheaper than that, up to the limit, if this is your goal, then you might think the answer is just to set a lower Target CPA of perhaps $7. But this still doesn't stop you from buying the stray $50 or $100 conversion. This method doesn't allow you to cap what you're willing to spend at the margin.
You might be tempted to think that this method could be appropriate for businesses with limited product lines that all have similar prices and margins. But in that situation, you're most likely better off using Maximize Conversions Bidding to get the most conversions possible.
For more information, read Should You Set A Target CPA in Google Ads?
Enhanced Cost Per Click Bidding
If you set a campaign to Manual CPC bidding in Google Ads (which now requires hunting through a couple layers of submenus that hide the option), a friendly looking checkbox will default to being selected…
“Help increase conversions with Enhanced CPC”
That sounds very wonderful. But what does it mean?
Enhanced CPC once meant that Google had permission to increase your bid by up to 30% if they deemed this search as being one more likely to convert. Of course, there wasn't a really good way to test this. But many people tried, and some found it seemed to work. Others found it didn't seem to work. In tests I tried early on, I just couldn't find enough evidence it worked or that it didn't. So I didn't have really strong feelings about it, but I'd turn it off if I felt I should be more conservative in a given situation.
The documentation claimed that it would also drop bids in situations less likely to convert. But I never saw any evidence of this. Just think about it like this: Most clicks do not result in conversions. Shouldn't it have dropped costs enough to save a ton of money given this reality? I never saw a scenario where selecting this checkbox saved money on a campaign.
But in 2018, Google made a change to ECPC bidding. They removed the 30% threshold on bid increases, and the documentation now states that it can raise bids on searches more likely to convert by any amount. That sounds concerning. But… the documentation goes on to say, it will ensure that the bids aren't raised by too much, and the average actual cost per click won't go above the Enhanced CPC bid that's set. Or rather, it will try to not do that. Hmmm…
Think about it. They redefined what ECPC means. But they kept that name, since it has such a good feel. If we were all a bit more honest about this bidding method, however, we'd call it "Target Average CPC Bidding", because that's what it is now.
Enhanced CPC will seek to get more conversions, while achieving a particular average CPC.
So how does this impact profitability? It's like a slightly worse Target CPA method, because it constrains the cost target to a metric based on clicks (which are further away from our goal of profit than conversions are). But like Target CPA, it has all the problems of overbidding for marginal conversions. It will spend far too much on terms that are too expensive as it averages those out with some that are achieved for mere pennies to try to achieve a given average CPC. It doesn't take into account anything about variations in average order values or profit margins.
Enhanced CPC bidding just does not make the right choices to drive increased profitability for typical ecommerce merchants.
Target ROAS Bidding
Target ROAS Bidding is one of the most widely used smart bidding methods for ecommerce merchants, and it has almost all of the same problems as Target CPA. Let's think about it.
Target ROAS is a goal based on averages. Let's say you set a 5:1 goal (aka 5X or 500%). That's a ratio between revenue and ad spend, and the same as saying you're willing to spend 20% of revenue. Notice that if you were selling $50 products and set a 5:1 ROAS goal, that's the same as saying you have a Target CPA of $10.
Target ROAS is actually just Target CPA with one minor tweak to it. Instead of having a fixed target cost per acquisition, the target cost per acquisition scales up or down with product revenue. So this method is an improvement. It isn't as strongly biased against bidding on more expensive products. It'll happily bid on more expensive products if they convert, even at higher costs, since the higher revenue is now taken into account.
But this bidding method still has all the other biases. It has no idea how product margins vary across your product line, and it's based on achieving a certain average. So it will dramatically overspend to get some sales, since that's averaged with other sales that are achieved at significantly below the target. This last point is key: Target ROAS is less of a performance goal, and more of a budgeting method. Once this method gets all the lower-cost conversions it can, it will overspend past your target and likely lose money on a lot of more expensive conversions, so it can fully spend the money available to it. There is no way to tune this method to avoid this, and this problem occurs with every bidding method engineered around averages.
Target ROAS is an improvement on most of the methods we've looked at so far, but it is biased against profitability, and makes very inappropriate decisions that reduce profit across many of the sales that it achieves. For most ecommerce businesses it has limited utility. A more nuanced strategy based on setting different bids and bid adjustments manually across the account can achieve greater total profit.
For more information, read How ROAS Bidding Kills Ecommerce Profit.
Maximize Conversion Value Bidding
Google released Smart Shopping campaigns not too long ago, and they included a new bidding algorithm called Maximize Conversion Value Bidding. This is exciting, because it brings us just a bit closer to a smart bidding algorithm that can be focused on achieving better performance than the previous attempts at automation.
But, yes, there are problems.
As of late 2019, this bidding method is still only available on Smart Shopping campaigns. Smart Shopping campaigns are an interesting hybrid of shopping and display ads, centered around products rather than search ads. They are really fully automated, with almost no configuration options and almost no ability to tune them. Google touts them as being amazing. But they've purposefully removed a lot of reports from these campaigns, so we can't peer inside to see where they're doing well and where they're screwing up. And there's no ability to fix the mistakes made by the machine learning algorithms. Early performance has looked good for some merchants, and horrible for others. Most people in the industry are taking a wait-and-see approach to them, perhaps with limited experiments. At this point, we can't recommend them. But we also can't say they are useless. Let's wait and see.
The Maximize Conversion Value bidding method takes some of what's best about Target ROAS and Maximize Conversions and blends it into something better, at least in theory. For a given budget level, Google will try to get the most revenue possible. Assuming it works, we're getting closer to being able to use smart bidding to try to maximize profit… except for this one pesky detail. This bidding method is really going to maximize revenue.
The bias should be clear: Maximize Conversion Value is going to tend to find cheaper conversions on lower-margin items that competitors aren't bidding up as high. This bidding method will tend to focus on low margin products. High margin products will tend to require higher bids, and Google is trying to maximize revenue by default. But if you have a product selection that tends towards similar margins, then that issue really doesn't matter. In that case, you won't have lower margin items, so this method's biases won't actually hurt your profitability… at least not for this reason. But this is not the only issue.
Once again, like all of the "Maximize X" series of bidding methods, Maximize Conversion Value will happily spend all of your additional budget any time you increase it, even if no additional sales are possible. It optimizes for the maximum conversion value it can achieve with a given budget. It doesn't try to spend less budget if it's unable to get more sales. It will try increasingly ridiculous bids on experiments that stand no chance of working, happily doing its job of trying to find more things that work. Figuring out the right amount to spend, in order to prevent waste at the margins, is very tricky.
Unfortunately, this bidding method still doesn't get us all the way to the holy grail of having a really good smart bidding method that can be focused on achieving maximum profitability. But it's one more step in the right direction.
At the time of writing, Maximize Conversion Value is being rolled out to some accounts and some campaign types in a beta test of expanding it beyond the smart shopping campaigns, and we're looking forward to experimenting with it. We'd like to try this bidding method with normal shopping campaigns where we have more control over tuning and optimization, for example. But even with this new bidding method, its limitations may result in it being unable to outperform an expert human who understands the business model and cost structures of a particular ecommerce site.
Other Advertising Platforms
I've focused on Google Ads, as they have a rich variety of automated bidding algorithms with more options than other platforms typically have. Other advertising platforms have their versions of these bidding strategies. But in just about every case, they have almost exactly the same issues and biases. Facebook Ads can be configured to drive a certain ROAS, for example, or set to maximize clicks within a certain budget. If you understand the general principles at work, you can quickly map those bidding strategies on to the examples above of Google's smart bidding methods and understand how the biases inherent in the algorithms will drive waste and push your profitability down.
Notice that there are several common themes that run through all of the issues we have when trying to use smart bidding for profitable growth for ecommerce merchants. We could boil these issues down like this:
- Bidding algorithms that don't understand profit margins cannot drive profit
- Averages hide poor performance at the upper cost margin
- Focusing on more of a thing, gets you more of them… but not the ones that are valuable
- If you're focused on anything besides profit, you don't get more profit
- Maximizing anything within a budget means spending it all, even with no extra results
The challenge with every system for automated bidding is that none of them are focused on achieving profit for advertisers. It would be trivial to engineer a new generation of tools for merchants to achieve profitability goals. And every one of these ad platforms employs armies of PhDs in Economics and Artificial Intelligence who know exactly how to do it. The reason they don't is that those experts are focused on engineering systems to make the ad platforms the most money possible—not you. I think that's unfortunate, and hope that with increased competition, they will someday be forced into more transparency and better tools for bidding. You could argue that there has been a steady improvement in the tools over time. But we aren't there yet. More improvements are needed.
Machine learning algorithms are generally good at doing what they're programmed to do. But you have to read the descriptions of what they do carefully and read between the lines to consider the unintended consequences of each of them. When you pause to reflect, it should be obvious that seeking more of something like conversions will result in a bias towards selecting a higher volume of cheaper conversions, and you can think through what that will mean for sales of your high-ticket items and your underlying profitability.
What's The Solution?
As more ecommerce merchants are lured into using these automated bidding strategies, it is actually creating some wonderful opportunities for the rest of us. In order to drive increased profits from advertising, it requires a more nuanced understanding of a business, its cost structure, and how value is created by balancing the costs of advertising against the marginal returns. When our competitors prioritize getting more low-value conversions, for example, competition for higher-value conversions goes down. This can be wonderful for those capable of taking a more nuanced approach and putting in the work to figure out how to get them.
Machine learning algorithms can truly achieve spectacular results at what they are programmed to do. But so far, no bidding strategy from any major ad platform has been engineered to drive increased profit. Many of them appear to be explicitly created to drive more profit to the advertising platform, actually, while being as opaque as possible to avoid too much scrutiny about what they're actually doing under the hood.
This is a key reason many advertisers are unwilling to move to any automated bidding method. When these systems are adopted, profit usually goes down. Sure, they're great at getting more of what they're optimized to get. More impressions. More clicks. More conversions. But those are not the right goals. Make sure you're measuring the right goal. Profit is the goal. Not clicks. Not conversions. Not ROAS. Profit.
Until the advertising platforms release some tools to help ecommerce merchants better manage these systems, it's probably best to stick with simpler bidding methods that give you more control and more understanding of what is going on. Two decades after Google unleashed AdWords on the world, you can still drive more profit in most scenarios with Manual CPC bidding and solid financial analysis than you can with any automated bidding strategy that has been released since. With the right techniques, bids can be varied across large accounts to take advantage of what drives conversions and how variations in those bids will impact marginal costs and the ultimate profitability of the products being advertised.
If you'd like to have a conversation about achieving profitable growth from your digital advertising, reach out to us.
Psyberware specializes in managing online advertising for ecommerce businesses, and nothing else. If you want to build a great relationship with a group of dedicated people who really understand ecommerce, get in touch with us.