What is ROS (Run of Site) or RON (Run of Network)?

Definition

Run of site (ROS) is a phrase used to define inventory (page loads) that publishers are unable to forecast ahead of time. As a result of this they don’t have advertiser spending lined up. This causes them to dump such inventory at dirt cheap prices. Another form of ROS inventory is inventory coming from some sections of publisher site which cannot be sold as premium inventory. For a newspaper site, such sections could include, job postings, obituaries, personals etc. Such sections have very low value for an advertiser and are thus given up as ROS inventory.

A lot of ad networks buy ROS inventory from publishers and try to make a profit by either reselling it or running direct response or branding campaigns on them. Such a collection of ROS inventory on an ad network is called Run of network (RON).

Typically ROS/RON inventory is prices quite cheap compared to premium inventory. As a result it is expected that it will perform poorly compared to premium inventory.

Why this happens?

Pages that do not have good content cannot be sold as premium inventory. There is nothing much one can do about such pages. They will end up as ROS inventory. A bigger problem for the publisher or an ad network is when they have a surplus of impressions for a given month and no advertisers lined up to serve ads. Such a case can happen either when a publisher does not make accurate forecasting about their future traffic or there is a spike in the traffic coming to the publisher. As a result the sales team sells only what was forecasted. If the publisher receives more traffic than forecasted then it will end up wasting those extra impressions. Since media buys are done weeks if not months ahead of time, any changes in inventory cannot be sold immediately.

A balance is met when another ad network or advertiser steps in and offers to buy such inventory. The catch is that the advertiser will not pay premium price for such inventory. They will just buy it at ROS price. This lets publishers dump their surplus inventory at whatever price they can get and advertisers get a steal deal. The only catch for advertisers is that they can potentially end up with a mixture of ROS surplus inventory and real ROS inventory.

There is a big difference in the prices of premium and ROS inventory. For example, the CPM price of premium inventory might be $10.00 while ROS might be as low as $1.00. This is the reason why forecasting is one crucial aspect of publishers’ ad operations.  

Yield Optimization

Definition: Yield optimization is a technique utilized by ad servers to improve the performance of a given advertiser creative.  In this technique the ad server tries to identify publisher impressions which are working well as per campaign parameters from the impressions that are not.  It then tries to place more and more creatives on the impressions that are working and less on the ones which are not.  Eventual goal is to place all creatives on the impressions which are working well. Yield optimization could be as rudimentary as tracking CTR (click through rate) for a given site and optimizing creatives based on it. On the other hand it could be as sophisticated as feeding a host of campaign specific parameters, like time of the day, publisher, ad size, geographical location, channel, price etc, into a machine learning system and letting the machine make the decision based on all those parameters about creative placement.

What you need to know about yield optimization:  There are a couple of things you need to know about yield optimization:

1.       Not everybody supports it.  If you are an advertiser and have bought some inventory from a publisher and are using an ad server, Atlas for example, it will not optimize your creatives  for you. The reason for this is that Atlas was not built to do so. It assumes that you know what inventory you are buying and have already done so, thus it makes no sense for it to decide whether it should place a creative on the impression or not. It assumes that you have made the decision already when you did the buy. It just does the placements for you.  

Typically, ad exchanges are places where you could see really sophisticated yield optimization.  This is so because you are trying to buy the inventory on the fly and it makes sense for you to know whether a given impression has any probability of doing well for your campaign or not.  RightMedia’s yieldmanager is one such ad server which supports very robust yield optimization.  You can use yieldmanager to run your own campaigns even if you did not want to participate in their ad exchange.  Although there is a queue to get on to RightMedia exchange.

Regardless of the fact whether you are buying inventory in bulk from an advertiser or from an exchange you could be smarter about how your place your ads on it if the server you are using supports yield optimization. Here is a scenario – you have a couple of campaign running on a couple of buys from different publishers.  Instead of placing your creatives evenly or per your media plan you can try to let them ‘compete’ with each other and let the ad server do the optimization for you.  Certain creatives might do well for a given publisher and their user audience.  Lots of agencies and advertisers are either unaware of this feature or do not utilize it fully.

2.       It needs to learn.  When you employ an ad server which can do yield optimization based on machine learning, you will end up losing some impressions to learning. When the campaign starts, the ad server will try to place the creatives on all publishers and then it will slowly weed out the ones which are not performing and zone in the ones which are.  The more you let it learn the better it will perform. In yieldmanager, for example, you can specify how many impressions you want it to use for learning.

Yield optimization is a great tool at the disposal of an advertiser that they can utilize to improve the performance of their campaigns and specifically their creatives. Creatives play an important role in the success of your campaign.  Some creatives work well for one scenario while others work for another.  If your ad server can pick and choose these scenario and creative combination smartly, your campaigns will tend to be more successful.  As an advertiser you should try to find out how you can use this feature of your ad server.  

AdNetwork

Definition

An ad network is a company which has relationships with a large number of advertisers and publishers. Its main purpose is to be a medium for connecting the two entities together.  Thus ad network will go out and buy impressions from the publishers and run campaigns for advertisers on those impressions. They share the payment received from the advertisers with the publishers.

How it works?

Most of the ad networks manage their advertiser and publishers using one or more ad servers. They setup each publisher site and sub-sections within that site in the ad server. They then manually classify these sub-sections of the site in a pre-defined list of channels. Generally there are about a few dozen channels. A few examples are – sports, health, finance, blogs, fitness, arts, entertainment etc.  Once this setup is complete, the ad network exports out the publisher tags for the publishers to deploy on their web pages. Each section of the website might get a different publisher tag.  As soon as the publisher tags are deployed on the website they are ready to do business.

On the advertiser front, the ad networks get creatives (banner, text or flash ads) from the advertisers to run the campaign. These creatives are uploaded in the corresponding advertiser space on the ad server.  A campaign is then created by assigning a set of creatives to it and targeting it to a set of sites, sub-sections within a site or channels within a site.  Other targeting parameters might also be set depending on the advertiser insertion order (IO). These might include things like – day part targeting, geo targeting, frequency capping per user etc.  Once the targeting is setup the network is ready to roll.

As soon as the publishers start receiving traffic from their users, the publisher tags loads and makes a request to the ad server. The ad server checks to see if there is an eligible campaign for these impressions. If there is one then the ad from that campaign is displayed and recorded. If there is none, then a default ad is shown. If more than one campaign qualify for the impression some ad servers can perform an auction to get the best price for that impression.

Why need an ad network?

 So why does one need an ad network? The publisher can talk directly to the advertiser and cut their deals and eliminate the middle man. The problem is that the ad networks have relationships with a lot of publishers and advertiser. This let them provide a wide verity of publisher impressions to an advertiser and a wide verity of advertiser campaigns to the publisher. It might be impossible for a small publisher to achieve this just by themselves. Ad networks are also very secretive about their advertisers and publishers for this very reason.

Most ad networks have a certain group of publishers and advertiser. For example, you might find ad networks which only have travel related publishers and advertiser.  If you were an advertiser looking to run a travel related campaign you might just go to them.

Future

The online ad industry is growing and evolving at a rapid pace.  People used to run campaign and manage them using Excel, now there are machine learning algorithm running inside ad servers which are trying to optimize an ad placement based on a dozen parameters. Being just a middle man who controls the relationship between an advertiser and publisher might not be enough. The ad networks will have to redefine themselves such their value proposition to advertisers and publishers is much more than just connecting them together. They should be able to provide a better price for the campaign for the advertiser and higher prices of impressions for the publishers.

Contextual vs Behavioral Targeting

Contextual and Behavioral are two very different types of targeting methodologies employed by ad servers and ad networks to deliver advertiser campaigns.  Depending on who you ask about what they think is the best form of targeting you might get different answer.

Why is contextual targeting better?

People who support contextual targeting believe that the context of the page is a good indicator of what a user is interested in. Targeting based on this information is valuable and can provide lift in performance of advertiser campaign.  The idea here is that if you show ads that are contextually relevant to the context of the page then the chances of a user clicking on them is higher compared to traditional methods. Companies and research firms have done tests to prove that this form of targeting indeed provides an improvement in campaign performance.  Google’s adsense and adwords are example of such targeting and is supposed to be very successful.  

Why is behavioral targeting better?

People who support behavioral targeting believe that context of a page is very momentary and does not provide accurate and enough information about the probability of a user to click on a given ad.  They believe that one needs to ‘follow’ the user across the internet and ‘observe’ what they are looking at before you can tell what they might be interested in. Behavioral targeting claims that you can draw precise conclusions about the likely hood of a user clicking and converting on a given ad. They can make correlations between various user activities for a given period of time and the chance that a user is interested in a given advertised product.

Which one is really better?

This is a tough question and like I mentioned before, based on who you ask this question you might get a different answer. These are two very different forms of campaign targeting which seem to be quite successful. There are companies doing well in both these domains.

There are so many different ways of targeting an advertiser campaign that isolating the impact of one particular criterion is very difficult. For example, how can you be sure that when you did a contextual or behavioral targeting for a campaign the fact that the user was a certain age, from a certain geographical location, using the system at a certain point of time etc did not have any role to play in them clicking on the ad. Some advanced ad servers can also perform yield optimization for advertiser campaigns. This means the ad server can make a judgment about an impression based on all the above parameters for a given user before it places an appropriate ad. In this case, all the above parameters go into a machine learning system which predicts the probability of a user clicking on a particular ad from a particular advertiser campaign.

Even though it is hard to compare whether behavioral targeting is better than contextual targeting, one thing, in my opinion, that can be said about these forms of targeting is that they may perform better if they were used in conjunction with each other than them being used in isolation.  For example, if the system could ‘remember’ the contexts of the pages a user has been on for duration of seven days and draw some kind of correlation based on this information, then it might be more precise because it is a blend of both contextual and behavioral targeting. The contextual data can go into the learning system which was earlier using just the simple data available like ZAG (zip code, age, geo), day part, frequency etc.  Contextual data is in most cased richer in information that the traditional form of targeting data. Thus, a system learning on contextual data gathered for a user over a period of time will be able to do a better job of predicting a user behavior than one which is devoid of it.