Module 07 Learning & optimisation
Signals the algorithm uses
Give the system a little information, and watch it go looking for the rest.
No list, so the system looks for itself
Broad targeting gives the system little or nothing to match. It uses what it knows about the auction and the people in it to find a group it expects to act. Each dot is one person; the highlighted band is the group it picked. Broad does not mean random. It means the platform chooses.
Your own data is a signal too
A list of past buyers, or the conversions your tag recorded, tells the system what a good customer looks like. In the diagram the small central group is that seed list. Conversions and customer lists are the strongest signals you own.
The system widens the circle
Given a seed, the system can look for more people like it. TikTok’s Smart Targeting will go outside your interest and audience settings when it expects a better result. Snapchat’s Smart Audience adds people beyond your inputs. Expansion is the system spending your signals on new people.
Context counts as well
Google Ads calls these contextual signals: attributes about a person or their context at the moment of the auction, such as device and location. Display & Video 360’s optimized targeting uses campaign data the same way. Who they are and where they are both feed the bid.
Keep a group back to check
If the system chooses the audience, you still need to know whether it worked. A holdout keeps a group out of the delivery so you can compare. In the diagram the dots with a dashed ring are held back. Automation still needs a control group.
| Group | People |
|---|---|
| Entire audience | 300 |
| Seed audience | 20 |
| Lookalike pool | 280 |
| Group | People |
|---|---|
| Entire audience | 300 |
| Seed audience | 20 |
| Lookalike pool | 280 |