Anyone who has ever worked with Display & Video 360 knows that ready-to-use algorithms are available, but these actually only offer a universal approach to buying impressions. If you as a business want to optimize towards a goal of your own, this has been impossible until now.
For a few months now Google has been working on creating a tool that allows each business to build its own algorithm and thus have an alternative to ready-to-use algorithms.
In this article, we will give you the keys to creating custom algorithms so that you too can create your own algorithm and achieve your goals.
What is a custom bidding algorithm?
The main definition of a custom bidding algorithm refers to a set of bidding rules that are designed to deliver results, aligned to your specific business’s goals. But concretely how does that work ? As a first-party algorithm, the advertiser will have to define custom impression scores in order to build a predictive model that will learn to optimize bids.
Basically, you first need to define and identify a KPI that you want to optimize toward. You can for example optimize towards Transaction value, customer value, brand awareness or even weighted conversions.
Once that step is done, the advertiser will have to build a “simple” script that will tell DV360 how much an impression is worth based on the KPI you just defined. “Simple” can give you a biased sensation. Yes, templates exist but some scripts as for example the one needed for weighted conversions can be rather complex to create.
Finally, DV360 will use this script to assign a value to an ad impression in order to automatically maximize impressions depending on your objectives.
To summarize how it works, your script will firstly tell DV360 how to score each impression. This score will represent how valuable the outcome of the impression is to you. Then the algorithm will learn which impressions are valuable and feed that back into the bidding model using the 40+ signals from the impression to understand which impressions will be valuable to you in the future. When DV360 sees a new impression it uses the data from the model to predict the value of the impression and determine what to bid.
What do you need to start with custom bidding?
The watchword here is DATA. You will have to keep in mind that the more data you have the better your algorithm will be able to optimize. Google states that you need at least :
- At least 1,000 positively-scored impressions for an advertiser
- At least 50 positively-scored impressions for an individual line item
- You can find all requirements here.
Obviously to start with custom bidding you will also need to write your script. For that Google anticipated that we aren’t all coding experts. Thus they wrote guides in order to help you create your script even without technical know-how. You can find samples of scripts and implementation instructions by following those links.
But what can you really do with it?
At this stage, you learned all the basics of what a custom algorithm is. But you’re probably wondering what value it can bring to your company?
As already said, besides all universal algorithms, custom ones will help you optimize towards your own KPI. You will have the possibility to mainly focus on transaction value, weighted conversions, customer value, and brand. In this section, we will review each of them to better understand the ins and outs of the expected outcome.
Transaction value
Optimizing towards transaction value will enable you to score impressions based on the actual value of a purchase. This means that an impression that leads to a purchase of 100€ will be scored 100 whereas an impression that leads to a purchase worth 10€ will be scored 10. As simple as this concept is, results have been striking for companies that already test this custom algorithm compared to classical ones. On average those have recorded a 73% increase in online sales and a 1.7x uplift in return on ad spend.
Weighted conversions
The goal here is to optimize towards your most valuable conversions. Taking the concept that all conversions are not equal, using custom bidding will enable you to optimize towards your most valuable conversions without missing out on other useful ones.
That means that instead of only counting the final conversions as Auto bidding does. You’re weighing different steps of your funnel to give more signals and feed the machine learning algorithm. Your custom algorithm will thus have more data in order to better perform.
Customer Value: u-variables
In contrast to auto-bidding which is only optimizing on purchase and thus considers all purchases as the same, custom bidding allows you to optimize on several sub conversions i.e purchase value, number of items, reward tier, product category,... This will help you optimize towards a higher LTV and increase your revenue. To enrich your custom algorithm you can even use Google Analytics goals to feed the system with more crucial info such as destination (landing page), duration of the session, Pages/screens per session or even based on events.
Brand
This one based itself on signals correlated with your brand in order to get the most of your campaigns. That means that you can either use viewability or video complete or even time on screen in order to make sure that people integrate the message you want to communicate to them.
What’s next?
Google is currently improving its tools in order to facilitate the creation of such custom bidding algorithms. The goal will be to create a fully integrated solution where you will only plug requested information into.
This is already kind of implemented into DV360. If you connect to your DV360 account and then navigate to “Resources” => “Custom bidding” you will find there the tool that Google already created in order for you to create your custom bidding and then share it with a specific advertiser.
A more technical article on how to create a custom script will follow as well as one of our business cases so stay tuned for further information!
Loïck Weber
| LinkedinThis email address is being protected from spambots. You need JavaScript enabled to view it.
