Evolving your strategy – Using the tCPA Smart bidding approach
Have you heard about machine learning recently? Only a dozen times a day?
Here is a practical take on using machine learning in managing your toughest and most important task as an SEM manager.
At Four15 Digital we have many medium, small, and large accounts taking advantage of the smart bidding approach powered by Google, which is an advanced form of machine learning algorithms. It uses data at a scale to help determine different bid amounts for each user and query to impact conversions or conversion value, factoring in a wider range of signals that impact performance than a single person or team could compute.
In other words, the algorithm takes into account auction-level signals. I know that experienced SEM managers can surmise intent from a keyword. In most cases we can. It’s a skill one develops with years of looking at the Keyword Tool projections. Keyword intent is a mighty signal, no question about it.
But what if a keyword that always converts also always fails to convert when certain conditions are met? I don’t presume to demystify the workings of Google algorithm, but the way to understand it is to consider vast amounts of data processed to learn which signals are present more often than not when a conversion occurs. Correlation and “hopefully” some causation analytics is also involved to determine the signals that matter, how much they matter, and whether or not they cause the conversion, or just correlate with it. So if someone who searched for that [insert your always converting keyword] is also exhibiting geographic or behavioral signals suggesting that they represent the case of a non-converter, using a smart bidding technique will result in a lower bid for that auction. Smart bidding is in essence a way to take more into account when bidding on an auction than the keyword itself.
But honestly, smart bidding algorithm works well coupled with standard knowledge of SEM, as it alleviates the need to fine tune the bids. You still have to select the right keywords, write ads and extensions, point to conversion-optimized pages, ensure correct tracking, and do the work of an SEM manager, but you won’t need to fine tune your bids.
There are several smart bidding techniques available today (aug 2019):
Target CPA approach is the one that I’d like to discuss given how much success we’re having using that approach. It is an approach that allowed us to improve account performance for a variety of performance-based accounts.
The way to migrate to tCPA is quite easy actually.
After you’ve accomplished the heavy lifting of building a campaign and accumulated enough data for it, your account will push automatic recommendations suggesting that you transition to a tCPA bidding like this:
The above is one way to learn what the system is suggesting in terms of a CPA target. Another way would be to navigate to each of your campaigns’ settings to retrieve information on the suggested tCPA. For instance, as I see my campaign’s settings, they look like this:
Clicking on the blue link in the image above, I see the following shortcut to applying the recommendation that pops up:
Clicking on the down arrow on the right hand side will take you to switching to tCPA the long way. First you’d click to select the bid strategy directly:
Then this:
At this point you can make a few strategic decisions that will make or break your approach. For starters I would not edit the recommended target bid too much. I would either not switch at all or switch with the recommended target.
If you have many campaigns already running on tCPA in the account, then I would only edit the target slightly for the first transition. If you recommended target CPA is $131, I wouldn’t slash it in half to lower it, make it $10 lower at most. However, there is a very important point to consider at this point. Your algorithm is suggesting a target based on average. I’ve seen many campaigns that have suggested a very high target, but delivered at a lower CPA, and quickly self-perfected to improve the CPA even further. This is the time to trust the algorithm and take it’s suggestion. This is also the time to watch closely.
Franky, I haven’t seen too many problems with the tCPA approach. It is indeed well designed to decipher the signals that matter and to self-perfect with more data it accumulates.
Have you had issues with your tCPA approach? How about success? Share with us below!
Comments are closed here.