Programmatic Advertising 106— Optimisation Options

Armed with a background in targeting and reporting metrics we can consider how to optimize a campaign toward its goal (or goals). A campaign will have three conflicting objectives:

Conflicting Campaign Objectives

We have covered each in prior posts, a brief reminder:

  • Pacing: The rate at which a campaign is delivering impressions against the target impression count for the campaign. [Delivered impressions within timeframe / Expected impressions within timeframe]
  • Performance: Achievement against the advertisers campaign goal — i.e CPA.
  • Margin: % of revenue retained as profit for the company (Profit / Net Revenue)

To understand how these are determined, lets have a look at a typical IO request from a customer:

Sample IO Request
Note: The IO states a start and end date, the format of the creative to be used and budget. The budget (Price) is simply the (number of impressions * Unit Price/1000 — since unit price is CPM) in the first example above: 489,631*£4.34/1000 = £2125.00

You’ll see that the goal is not defined in the IO, but is discussed and agreed based on discussion between the customer and sales team. The delivery and cost data is included — and usually it is a requirement of the campaign to deliver a set number of impressions or revenue, split between different formats.

This leaves two variables Performance and Margin — the better the performance the happier the customer and more likely to spend more in the future. Performance drives long term revenue growth whereas Margin drives profitability of the company. The decision between the two is usually taken by the Trader — more margin will be taken if the campaign is performing ahead of expectations.

There is a direct link between performance and margin — so to simplify the conflict stated above — you can consider the following priorities:

  1. Achieve Pacing goal
  2. Maximise performance against goal of the campaign
  3. Determine how much performance should be kept as margin

The first two can be determined through analysis — the 3rd should be determined by the trader.

Goal

While the goal is not defined on the IO request, it will be agreed verbally as a target. Most common goals are defined in the prior metrics post. A brief reminder of our most common goals:

  • DR Goals: CPA, ROI, CPC — Each goal measure the response to the ad — When a fixed campaign budget is given this means finding as many convertors or clickers as possible within that fixed budget.
  • Brand Goals: CTR, Viewability, CPCV- Rather than focusing on a user reaching a specific goal-Brands are interested in engaging eyeballs on their brand message so measure ratio of users that are engaged by viewing a video or impression or clicking an ad.

Bid price considerations

There is a vast pool of online inventory available; winnable at a wide range of prices. Highly engaged users on premium websites may command a very high bid to win the auction, whereas unidentifiable users on obscure websites may be available at close to no cost.

The objective of targeting is to predict the likely-hood of a user taking the required action and bidding an optimal amount, for CPA campaigns:

Bid price = (probability of user taking action * CPA target) — Margin

It may be tempting to target just those users who are highly likely to convert — however, you may find that those users are too expensive to drive performance.

Lets look at a very simplified example to understand why reach is an optimisation parameter not just a target. Imagine the following two scenarios:

Target size effect on performance

Because the auction model prices each user differently it is important to ensure you have the luxury to be selective in bidding. High priced users, even those who are highly likely to convert — are often not worth bidding on, given we do not know the price necessary in advance for given users — it is important to ensure the target of users is significantly larger than the delivery need of the campaign to prevent being forced to bid on high priced users. It is often true that an un-targeted campaign will perform better than a campaign which is over targeted due to inflated bid pricing.

Worse — since you will never win bids on all users in a segment if the campaign is over targeted — you will never reach the pacing needs of the campaign and miss the primary goal.

Always keep in mind that each targeting option combines to create the union of users that match. While the initial inventory availability is vast, with just a few targeting constraints, the available inventory could be too small to achieve the reach goals. While you may feel you have plenty of users in your segment, with additional geo-location or demographic filters applied, it may be far smaller than expected.

Targeting Options

Since we cannot define the exact bid at a user level through our DSP partners, the standard methodology for increasing performance and choosing to distribute this performance between the Advertiser and Captify is to reduce the number of users at a certain bid price through targeting.

NOTE: AppNexus have recently introduced a feature to allow user level bid adjustments on segments, so for user targeting we may wish to use this rather than sending segments of users with equal bid value

An ideal line item would have a single campaign strategy that is known to work best, with optimal settings to provide performance at reach. This is typically not plausible — a cluster of users may exhibit very high performance but not meet reach needs, so multiple strategies are used with different bid amounts to meet the performance and pacing needs of a campaign. You should consider however, any budget spent on a second campaign is diluting your overall campaign performance — so an ideal state is the lowest number of strategies to achieve goals.

A campaign should start by casting its net as wide as possible to explore performance under all conditions, unless prior knowledge on performance exists. For a rebooked campaign by the same advertiser — it makes sense to skew the initial targeting based on what worked during the prior campaign. Even in this instance — it makes sense to reserve some budget to explore alternate targeting strategies in case of new discoveries. As the campaign progresses, targeting can be tightened and performance is expected to increase.

Allocation of budget for learning

User targeting

The DSP offers many parameters on the user profile to reduce your target group from all users (Run of Network) to a set of users expected to perform well. These largely align with those discussed in the Audience metrics section of the reporting page. This makes complete sense — since you need to report on something to measure its impact. Examples of how user targeting may impact performance:

Frequency — Showing an Ad to a user once is cheapest, but often a user needs to see an ad several times before converting. Clearly showing an ad endlessly to a user is wasteful — the optimal number of exposures for a given campaign varies.

Geo — Some geographic regions will perform better than others — usually there is a fixed geo (e.g: UK) to start the campaign and then further geo targeting can be applied if reporting shows performance skews by region.

Demographics — Products can be more interesting to users based on their Age or Gender, an advertiser often has an opinion on this and can be a required target audience, though it is useful to consider analysis during the campaign.

Day/Time — Showing products at weekends, or in the evening may prove better than early in the morning, however — consider than inventory costs change by time also — getting the user engaging in the middle of the night may just be performant due to low inventory costs.

Site Retargeting- Probably the highest driver of performance is to target those users who have visited the checkout page already — they have shown very high interest in the product. While performance is very high, you are limited to a small audience so reach is constrained.

Audience Segments — any other first or 3rd party data can be used to identify performing sections of an audience. Here Captify can leverage its high intent search data and split the audience by interest levels. Providing different bids based on their intent can be a very powerful targeting criteria.

Inventory Targeting

Finding the right audience can dramatically increase performance — further optimization can be delivered through finding the right placement and creative type.

Site: Show an Ad on the financial times and a wealthy investor may show interest, show the same ad on a long tail gossip site and the same user may consider it less trustworthy. The host site for the inventory can play a large party in the quality of the user experience, but you generally pay for this quality.

Format: Different creative formats were discussed earlier but they can have an effect on engagement — just as the quality of the creative can. A large intrusive ad may cause user disengagement — but is also more impactful. Depending on the quality of the advertisers creative different formats may perform differently, though often we are stuck with the formats the advertiser decides upon.

Viewability: Just as the brand advertiser wants to ensure the Ad is viewable, it makes sense on a DR campaign to check the impact of viewability on performance. If the Ad is partially obscured it will be cheaper but less likely to engage the user.

Fraud & Brand Safety: Often an advertiser pre-requisite — Disney don’t want their brand being shown on alcohol or adult sites. The trading team will place these required constraints on the campaign and there are 3rd party data providers who offer site list filtering to block different types of site (Adult, Alcohol, Violence, Religious etc). There are also 3rd party providers who offer user segments against likely fraudulent behaviour — it is wasteful to bid on click fraud bots.

Context: Companies such as Grapeshot and Peer39 offer a catalogue of page category. In the same way you can prevent ads showing on adult sites, you may target sites with specific page content such as sports or travel. This is often a strong targeting mechanism since you can align your ad context to the page it is shown on.

Conclusion

There are infinite possibilities to target an efficient audience, the key to strong targeting is to gather as much data about the target audience and calculate the lowest bid possible per probable goal event.

Begin by applying any prior knowledge you know about the expected campaign outcome — which may be nothing — to the campaign target. As you identify strong intent indicators through analytics, apply them to the campaign until you have found the optimal level of targeting.

For Captify we may automate the identification of strong audience signals within our search data to seed the campaign with an efficient audience. Remembering that the trading team may themselves identify strong indicators within the DSP. The combination of targeting can lead to over targeted audiences and therefore the need to over bid to retain pacing.

There are no golden rules — constant analysis of all campaign metrics is the only way to succeed in delivering a superior performance outcome for the campaign. Superior performance will lead to increased revenue and profit for the company over the long term.

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.