Advertising algorithm leverages understanding of users and their interactions with your products/services and similar products/services to predict likelihood of a required outcome. It uses this logic to bid. It record the results and course corrects till an advertisers goals have been met (generally speaking).
Unlike humans who can be irrational, an advertising algorithm operates in a rule based world with a very clear and single outcome that it is looking to maximize. Advertising platforms will generally allow you to set typical constraints (threshold values) for these rules, so that the algorithm can work within those constraints. If these constraints are too tight, the algorithm will be unable to deliver enough outcomes to learn, self-optimize and deliver. And if these constraints are too loose, the algorithm might not give you the right level of efficiency.
Advertising algorithms generally tend to optimize bids to achieve specific outcomes that are deemed as most important by the advertiser. Some of the important outcomes are:
- Conversion action like lead submission, application submission, newsletter subscription
- Reach i.e unique impressions or Video view
- Traffic to site i.e clicks to site. An important point to note is that not every click leads to an actual visit on the site. So always looks at Cost of Online Visit
- App Install
- Ad position – mainly for paid search channels
- Impression share – main for paid search channels
Advertising algorithm generally goes through a few clear/distinct phases to reach its optimum performance within the constraints you have set:
- Learning Phase: In this phase, the algorithm does aggressive testing to learn what kinds of users are responding and find the best combination (bid, creative ,customer traits etc) that can deliver the results you want. An advertising algorithm needs a minimum number of results within a given period for it to learn enough and emerge out in the next phase
- Delivery Phase: This is the phase right after learning phase where the advertising algorithm thinks that it knows enough to meet your objectives. In this phase, the algorithm starts doing more of what is working to scale up performance.
- Evaluation Phase: Evaluation phase runs in tandem with delivery phase because the context around which the advertiser’s ads are being shown is constantly changing. Competition could be showing aggressive offer ads or the traits associated with performing customers might be changing. In this phase, the advertising algorithm is constantly making minor modifications to what works to ensure that the results continue to scale. Sometime, this evaluation phase can also get kicked in when creative changes happen. While it might not require a complete re-learning via the learning phase, it would still require the advertising algorithm to test, record results to adjust and adapt the winning combination.
Think of advertising algorithm as your sales team which works 24×7 to drive results for you. They will always be by your side and are always willing to work for you. All you need to do is give them the right customer guidance (targeting criteria), right marketing collateral (creatives), the right objective and enough budget so they can do their job. As with managing any sales team, you need to monitor their performance closely in the beginning to ensure they are working well. If you have done the best job with the above things, then the most important virtue for you to have is patience because unlike a human sales force, machines still can operate at scale but they cannot be as flexible in making pivots as humans.
Now, that you understand how an advertising algorithm works, you might also be interested in reading about what makes Facebook’s advertising algorithm one of the strongest.