4 things we learned about automated bidding during the pandemic

Bid management is understandably a favorite topic for PPC marketers because the right bids make the difference between campaigns that lose money and ones that make money. In the old days, search marketers spent a lot of time setting the right CPC bids. But nowadays, in a world where more of the day-to-day of PPC management can be automated, that task has shifted to picking the right automation and feeding it the right settings like targets for CPA and ROAS.

Since the right bids are so critical to success, you’d think it’d be one of the best-understood parts of PPC management but in reality, how automations like Smart Bidding from Google work is still a bit of a mystery to many. And that’s no surprise because the underlying mechanisms for automation are built on machine learning, a technology that can’t easily explain the predictions it makes.

So there are a lot of questions about bid management and automated bidding in particular. For example, advertisers lack confidence in how well these automated systems work during times of unusual volatility and whether pausing campaigns will hurt future automated bidding performance.

The coronavirus pandemic has caused quite a bit of volatility leading to many paused PPC campaigns and an unexpected side effect is that this has given us new insights into how automated bidding works. 

Will volatility break automated bidding algorithms

There was concern that automated bidding might fail at the start of the pandemic. After all, the automations work to detect patterns and use those to set bids and with the whole world changing their lives virtually overnight, those patterns were disrupted. How would machine learning respond to this?

First, a super quick primer on how automated bidding generally works. Smart Bidding sets bids based on factors related to a search that influence the likelihood of the search leading to a click, a conversion, and a transaction of a certain value. The higher the predicted chance of a positive outcome, the more the system will bid for that click.

Here’s an example of how the algorithm might work with something it knows about the user doing the search. It uses historical data to predict if a user who is searching for a printer from an office building is more likely to convert and more inclined to buy a more expensive printer than a user who is doing a similar search from their house.

In this example, the algorithm is using signals from the query to guess if the searcher is B2B or B2C, something that can greatly impact the end result. This prediction intuitively makes sense to us. 

But what happens when a global pandemic closes offices and virtually overnight the whole world starts to work from home? Does the predictive model break? Does it quickly recuperate? And will it continue to set bids that help advertisers reach their CPA and ROAS targets? 

We asked some of the world’s most influential PPC experts and here are the four things we learned…

Lesson 1: Smart Bidding quickly adapts to changes

Peter Oliveira from Google shared during our PPC Town Hall that “Smart bidding uses both aggregated and recent trends, [but] favors what’s been happening recently.” So it quickly adapts its bidding strategy to account for the new normal. It knows it can always go back to its previous strategy when things return to the old way. But it does not wait long to adjust bids because it considers new patterns to be more important than old patterns.

We already knew automated bidding could handle temporary volatility because it handles seasonal events like Black Friday. But what we didn’t know was to what degree the machine used its knowledge of previous years versus data from this year during periods of flux like Black Friday.

Not knowing this is what made advertisers uncomfortable to trust automated bidding during a volatile time that had no seasonal precedent.

But now we know that by-and-large we can trust automated bidding to favor recent data enough that we don’t need to worry about it failing to account for unexpected volatility.

Tip: Use alerts to find when automated bids may be too low

While automated bidding deals appropriately with new changes in conversion patterns, marketers should still set up alerts to be informed about changes in performance. 

By having alerts, you can stay better informed and that allows you to better respond to changes in the industry. While automated bids will appropriately reduce CPCs if something causes conversion rates to plummet, it might in some cases be better to shift messaging or promotional strategies rather than reduce bids.

By reducing bids, you’re basically giving up on prospects when something is causing them to convert less than before. It may then be hard to recover and rebuild your pipeline later on when things return to normal. Instead, help your prospects convert by sweetening the offer or allowing them to drive a micro-conversion, like downloading a whitepaper rather than making a purchase.

Scripts are great for building alerts so check out some of my other posts with free scripts like this recent one that detects local anomalies. 

Lesson 2: Reduce budgets instead of pausing campaigns

We know that the machine learning models for automated bidding are constantly learning, even before campaigns are set on automated bidding. All the system needs to learn is for conversion tracking to be enabled. From that point forward, every query’s data becomes part of the model that calculates the probability of a future query to convert.

Because many advertisers temporarily halted online ads, we now have some new insights into what happens to automated bidding algorithms when you starve them of recent data, the type of recent data Peter Oliveira said is so important for the machine to set good bids.

We first heard from Navah Hopkins that campaigns whose budgets were reduced rather than paused returned more quickly to normal levels when the recovery started. She said “Clients who stayed the course throughout the flux are in an amazing position; the ones who pulled back their spend are experiencing a far more intense recovery. The former have been able to capitalize on cheaper CPCs and really own the ‘compassion conversation’ to stay top of mind.” Several other panelists agreed with this finding that campaigns that were paused were facing a long road to recovery and in some cases had to revert to manual bidding.

Campaigns that were paused seemed to stop learning and when re-enabled were confused about the new state of the world and needed significant time (~2 weeks) to again understand how to help set the right bids to help advertisers reach their goals.

Lesson 3: Take advantage of all bid automations

I’ve never been a fan of some bid automations like ‘maximize clicks. It’s not part of Smart Bidding (the class of automations that set bids at auction time) and it also seems to be more aligned with a visibility goal rather than a direct response goal.

But strangely, during volatile times with many advertisers temporarily pausing campaigns, these bid automations worked better than usual. 

And it sort of makes sense… During normal times, a ‘maximize clicks’ strategy buys the lowest quality clicks. That’s because the way to get the most clicks for a set budget is to buy the cheapest clicks. And the cheapest clicks are those that are less likely to convert and hence get lower bids from performance-oriented advertisers. 

But in unusual times, many good clicks also became cheap, not because they had little chance of converting, but simply because ad budgets got slashed. This meant that even an automated bid strategy like ‘maximize clicks’ would lead to better than typical lead quality.

Advertisers like Kirk Williams figured this out right away. He’s embracing the role of the ‘PPC doctor,’ a role that will keep humans relevant in a more automated PPC world, something I describe in my book. When we understand how the ad auction works and how the different automations work, we can make better decisions that help us and our clients get through unusual circumstances like the ones we now face. 

Lesson 4: A high Impression Share can be bad for automated bids

A final finding from the pandemic’s impact on PPC is that Smart Bidding needs some leeway to explore new opportunities. 

While we usually think of Smart Bidding as purely a system to automate bids, we have to consider that it also works to automate targeting. Let me explain… When ads show for close variants or other new queries, we’re allowing the machine to explore new sources of conversions. With manual bids, this exploration needs to be closely controlled because the close variant inherits the bid of the keyword that triggered the ad.

But sometimes the close variant is not so close to the original meaning. It may still have some relevance, but not nearly as much as the keyword the advertiser selected. Smart Bidding can automatically set an appropriately lower bid so that the advertiser gets the benefit of incremental conversions, but while still respecting the target CPA or ROAS.

In normal times, it’s fine to limit the exploratory capabilities of Smart Bidding by tightly controlling the positive and negative keywords in an account. After all, we know more or less how users behaved in the past and we can use that to guide the machine to waste less money on tests we think have little chance of amounting to anything.

But in strange times, user behavior changes and so may the searches they do. It is then helpful to have the benefit of the machine to help make the pivot to the new normal. The machine can identify the new batch of relevant searches that deliver conversions. 

So how do you know if you’re limiting Smart Bidding techniques like tCPA and tROAS too much? You can look at the Impression Share. If you have less than 5% Search Lost IS (Budget), it’s likely that the automation is mostly controlling bids and doing very little exploration of new opportunities. 

By adding more budget, loosening up match type restrictions and being less restrictive with negative keywords, you allow the machine to shift more of its benefits to finding new conversions.

There is obviously a risk of giving the machines too much leeway. So as I said before, alert and notification tools and scripts are a critical part of the process you should have in place to avoid unnecessary wasted spending.

Conclusion

Automated bidding continues to get better and has earned recent kudos from search marketers. When the ‘new normal’ becomes the status quo, we’ll be able to look back on the lessons learned in 2020. And as we learn how automated bidding fits into the bigger picture of what we are trying to achieve as search marketers, it’s clear there is a lot of value from having a human expert who remembers the lessons and uses those to guide advertisers to PPC success.