Frequency

The frequency cap has been around for 15 years.

Why does every ad network use it?

And if software is eating marketing, what new opportunities are there for this user-level optimization?

Source: Microsoft Advertising

What is frequency capping?

Typical networks allow marketers to set a hard frequency cap for each new campaign. For example, once a user has seen your ad, say, 2 times in a 24-hour period, stop showing ads to the user. Marketers can typically adjust two variables: number of exposures; and, sometimes, the time period. They usually have no idea what to use for these numbers and learn by trial and error.

How important is it?

A user’s first exposure to an ad is 3–30 times more likely to lead to a conversion than subsequent exposures.

Why does it work?

Well, there are 3 types of users seeing your ads:

1. Ignorers (70%)

• They will never convert, no matter how many times they see the ad. Maybe they’re not interested in your offer; maybe they’re one of those people who keep telling me they never click on ads.
• Immersed in playing a game or reading. Don’t want to be interrupted.
• They come back every day to this publisher; they’re hardcore users.
• They’re familiar with the app and have learned to tune out ads.
• They’ll be on this one app for a really long time.
• The higher a user’s frequency, the likelier the user is in this ignorer group.

2. Maybes (20%)

• As the ad is repeatedly shown, the user has a wider opportunity to engage.
• But as the novelty of the message wears off, the conversions go down.
• As frequency increases, conversion rates either remain the same or go down.

3. Adventurers (10%)

• They’re bored and looking for something new. In a jumpy and receptive mood.
• They’ve probably been on many different sites in the past few minutes.
• They could easily get lost in a new game (maybe yours!).
• These are users we see only on the first few exposures for this publisher. The higher the frequency, the less likely the user is an adventurer.

Frequency works because it limits exposures to the ignorers (#1) and prioritizes the adventurers (#3).

What’s next for frequency?

Just as RTB enabled retargeting, so will it enable more profitable user-level optimizations.

What should the new generation of mobile DSPs do? Usually they copy the frequency cap feature from the old ways of media buying. But so much more is possible!

With RTB, machines process every impression. They can track every different creative, at what time, our users have seen. With machine learning, there are lots of ways to use this data for the marketer’s benefit:

• Make models of sessions: Continuous time intervals where the user is seeing ads. Start to learn user behavior and how quickly they move through different activities.
• Every impression isn’t created equal. Some ads might be shown for a longer time, be on a different location on the device, or are one after another.
• Integrate with other behavioral and demographic models. For example, what are the effects of time of day (tiredness/attention span) and day of week (mood/free time)?

AdWill

Here’s what’s currently implemented at AdWill, as of December 2014.

Depending on how many creatives a user has seen, across all campaigns, our system learns a sliding scale of decreasing probability factors. For example, if a user is 3 times less likely to convert after having seen an ad twice, then the algorithm should consider this user 3 times less valuable.

(Storing data on hundreds of millions of users isn’t your everyday programming challenge. We use a version of bloom filters, inspired by Carter Schonwald and Ilya Grigorik.)

Along with frequency, other user attributes — device, location, time, etc. — affect the probability of conversion. The machine constantly updates its bidding strategy based on what it learns from users’ behavior, with the goal of generating as many conversions as possible for our clients.