Determining the Optimal Frequency Cap in an Ad Campaign

Introduction

When a digital ad campaign is being run by an advertiser, they’d want their dollars to be spent in reaching exactly the right group of users. Usually this means targeting the most relevant demography or audience. However, sometimes the campaign may keep showing ads for same/similar items to users who’ve already made the purchase for that item. Or keep showing ads to people who have no intention of buying the item at all.

In the context of an ad campaign, the issue here is with the frequency settings. We use a frequency cap to limit the number of times a user may see the same ad. This can be useful if you want to prevent users from seeing your ads too many or too few times.

It should be apparent by now that the frequency settings when showing an ad are extremely important. If we show too few ads to a user, we may be giving up too soon. That is, the user may have converted if we’d shown the ad a few more times. On the other hand, if we keep on showing ads for too long, we are effectively wasting our budget on a user who isn’t likely to convert or has already converted.

As a result, there is an optimal frequency cap which needs to be determined so that the campaign performs well.

In this article, we’ll look at how to increase or decrease the frequency cap so that we can find the most optimal value using a data driven approach. Also, we’ll discuss how frequently we need to evaluate whether our frequency cap is optimal or not. We shall restrict our scope to the Google’s DV360 as a DSP, and use only the settings and data that it provides

Definition of Frequency Cap

DV360 introduces frequency cap with the following lines

“Use a frequency cap to limit the number of times users may see the same ad. This can be useful if you want to prevent users from seeing your ads too many or too few times.”[1]

Approaches to determining frequency cap

We can set the optimal frequency cap the following ways:

Intuition/ domain knowledge: traders generally have an idea of what an optimal frequency cap is for the campaign based on several factors[2]. However, this generally involves broad generalisation and may turn out to be sub-optimal.

LI lifetime frequency:

• This data provided by DV360 gives details about how many times the ad was shown to a user in total (as opposed to how many they were shown the ad daily, for instance).
• Using this data we can look at the point where the CPA starts to get really bad. We can then decide that showing ads to users beyond this point is wastage. Let’s say we find that showing an ad 15 times to a user in their lifetime is a good number.
• We’ll also now need to decide the number of days for which we want to show the ad to the users. Let’s say we decide that a threshold of 3 is good, That is, showing the ad for 3 days is good enough.
• Accordingly, we can get the frequency cap:

Frequency Cap = LI lifetime frequency threshold / number of days ad shown

• In practice, this is slightly tricky as we need to make sure that we find two good thresholds separately such that their combination is good

LI daily frequency: this is data provided by DV360 about how many times the ad was shown per day before converting. This has worked well for us in practice, so let’s see how we can use this data. We’ll look at how to decrease the cap and increase the cap as per the situation.

Decreasing frequency cap

We will use this logic when we are spending budget on showing ads at an unnecessarily frequent level

The following steps will be followed:

• Get the performance stats for each frequency level — that is for users shown ads once per day, what were the impressions served, cost spent, conversions obtained etc; similarly for ads shown 2 times/day and so on
• If the frequency cap is too high, we’ll find that at the higher frequency level, we’re spending too much for very few conversions. We can safely remove these higher frequency levels
• Also make sure that we’re not losing out on too many impressions by removing the higher frequency levels. So ensure that by capping the frequency, we’re still retaining 80–90% of the overall impressions

Increasing frequency cap

As mentioned above, a cap may cause issues in delivery. Hence when we have a budget increase, we would like to ease the frequency cap limit. This would ensure that the frequency cap is not a bottleneck in delivery

What challenge would we face when trying to do this ? If our frequency cap is set to 5 daily, we won’t get data of how much cost and conversion we get by showing ads >5 times/day. That is, do increase the frequency cap, we’ll have to estimate how much delivery we can gain by doing so, we don’t have concrete performance numbers to use here

• What can we do?
• For each frequency level, we have the impressions served at that frequency. So we can find the number of unique users shown in the ad once, twice and so on. Accordingly, we can find the Percentage of Users Retained (PURe) across frequency levels
• Now, we’ll find that the PURe is approximately the same for an LI across different budgets and different date ranges. This means that even if we increase the budget, we can expect the user fall percentage to remain about the same
• With these set of numbers, we have some ground for extrapolation. We can have two possible cases -
• Case 1: no point in increasing frequency cap to aid increase in delivery. For example we may see here that the increase in frequency cap may only give an increase of 100 impressions, so not much point in increasing the cap
• Case 2: increasing cap would help in greater delivery (though at the cost of performance)

We can now extrapolate find out for which LIs the frequency cap is worth increasing, and upto what limit

When should you change the frequency cap?

Optimal frequency of showing ads is a question that is largely determined by the psychology of users (i.e. how many ads should a user see that they are most likely to convert?) , and the nature and objective of the campaign (For e.g. questions like is it a product like automobile or jewellery or insurance which generally users like to think a lot before converting? Or is it a branding campaign? etc.) Hence it doesn’t vary drastically/abruptly with time. So frequency cap isn’t something that you’d like to change every week or two.

However, seasonality does matter for many campaigns. For e.g., for jewellery campaigns around wedding seasons and festival seasons users are more likely to convert, so it may make sense to raise the frequency cap around that time.

In short, while the optimal frequency may not change very drastically with time, it does change. So we should look to monitor how each frequency level is performing and increase/decrease it as per our needs every one or two months.

Conclusion

A general intuition/experience based frequency cap is a good starting point when setting the frequency cap. However, it leaves some questions unanswered:

• We should have a framework to be able to determine of the frequency cap is working well or not
• LIs may be set up differently and may have different frequency cap needs. We need a data driven way to be able to customise the campaign in such a way that we can identify the frequency needs of different LIs, and be able to monitor and tweak it to find the best frequency cap

Because of these limitations, its best to use the data that our DSP provides to find an optimal frequency cap

PS: I need to thank Naga Preethish Pandurangi and Stephanie Pinto for coming up with ideas and helping me with the analysis!

Sources

[1] https://support.google.com/displayvideo/answer/2696786?hl=en

[2] https://www.facebook.com/business/news/insights/effective-frequency-reaching-full-campaign-potential