Summarized from this W3C document.
This document provides an overview of key advertising use cases that depend on cross-site data sharing.
Advertisers want to know how many times people saw their ads.
Sometimes, an ad is returned from an ad-server, but never actually enters the viewport. Sometimes an ad might enter the viewport, but only for a few milliseconds. Sometimes an ad might enter the viewport, but fail to load any images or video before it leaves again. Sometimes there might be other elements drawn over the top of the ad. These cases illustrate "non-viewable" ad impressions. Advertisers only want to pay for "viewable ad impressions". The reasoning is simple, an ad cannot possibly have an effect on someone's future purchasing behavior if they never saw it in the first place!
Advertisers want to show ads to humans, not to bots and scripts. The industry has devoted a great deal of time and energy towards identifying invalid traffic (IVT), so that it can be filtered out of impression reports.
When a measurement vendor (ad server, publisher, etc) reports a certain number of impressions (presumably viewable impressions, with invalid traffic filtered out), how is the advertiser to know if this number is accurate? Advertisers would prefer not to just accept these numbers on faith. For this reason, the industry has established a number of independent, third-party measurement vendors that run their own verification scripts to provide advertisers with the peace of mind and confidence that the number of impressions they are being billed for are accurate, and measured in a consistent way across all of their ad buys.
Not all products are suited for all consumers. For example, in the United States sales of alcohol are restricted to people aged 21 and over, and there are even regulations limiting the extent to which advertisements for alcohol can be shown to people under 21. Thus advertisers of alcohol have not only an economic interest but also a legal obligation to avoid advertising to teenagers.
Advertisers need to know how many conversions happened as a result of their ad campaigns.
Ideally advertisers would like to know how many conversions were caused by their ad campaign (i.e. would not have happened were it not for this ad campaign). This is variously refered to as "causality" or "incrementality" measurement. The gold standard measurement approach to answer this question is “Lift Measurement”. This involves a “test group” who is shown ads, and a “control group” who is not shown ads. These groups should be randomly selected prior to running the test to ensure they are well balanced.
The total number of raw conversion events is counted in both groups, and the difference between the two is called the “lift”.
Much like direct response advertisers want to know how many incremental conversions were caused by their ad campaigns, brand advertisers want to know whether their ad campaigns succeeded in making consumers incrementally more aware of, interested in, or favorable towards their brand, product, or service.
Because these attitudinal outcomes are not linked to a discrete observable action (like an online purchase), they typically must instead be measured through surveys that ask questions like "to what extent do you believe that Volvo makes the safest cars on the market?"
Lift is measured by comparing the prevalence of a given answer among an "exposed" group of respondents who have seen a particular ad campaign compared to the prevalence of that same answer among a control group who have not seen the campaign.
The alternative to lift measurement is to use some heuristic to “attribute” conversions to ads. The two most popular heuristics are “click-through attribution” and “view-through attribution”. “Click-through attribution” gives credit to an ad if the person had previously clicked on the ad prior to the conversion event. “View-through attribution” gives credit to an ad if the person had previously seen the ad prior to the conversion event.
Both attribution heuristics come with some concept of an “attribution window”. For example, a “one-day view-through” attribution window would only count conversions which happened within one day of an ad view. A “28-day click-through” attribution window would only count conversions which happened within 28 days of a click.
In practice, advertisers run many ad campaigns simultaneously across multiple platforms. People often see (or even click) on multiple ads for a given product prior to making a purchase. When a person interacted with multiple ads, across several websites before eventually making a purchase, who should get the credit for the conversion?
One popular approach is “Multi touch attribution”. This is a system for allocating partial credit to each of the ads the person interacted with prior to a conversion.
The alternative is called “Last touch attribution”. This approach grants 100% of the credit for the conversion to the last ad the person clicked on prior to the conversion. This approach may give too much credit to search ads, as illustrated in the following example.
People often make a purchase from a different browser (or even a different device) than the one where they saw an ad. A few common scenarios for this include:
Ad shown on a Smart-TV. Person opens their phone to interact with the advertiser.
Ad for a big-ticket item shown on a smartphone. Person completes the purchase on their laptop later after doing some price comparisons / web research.
Ad is shown within a mobile app. A click on the ad opens a mobile website either within a “Webview” or within the built-in mobile web browser. Person completes the purchase in the browser while the ad click is within a mobile app.
Ad is shown within a mobile app. A click on the ad opens a different mobile app, but the person does not commit to purchasing then, as they are travelling. Later at their computer, they open a browser and further consider the product. They then switch back to their phone (where they perhaps have Apple Pay or Google Wallet set up) and purchase quickly and easily.
It is very important to provide measurement for these flows given how common they are, and how for certain platforms, they are essentially the only way conversions happen.
Web advertising is not the only contributor driving purchases made on an advertiser's website. Email campaigns, social media posts, call center interactions, promotions internal to the advertiser's site and other channels may also have influenced the user's purchasing decision. In order to properly allocate budgets to the channels that influence conversions, advertisers need to understand the relative importance of the various channels and how they work in sync to drive conversions.
Fraudsters will attempt to send fraudulent conversion events. Possible motivations include:
Disrupt the ad measurement of a competitor business
Disrupt the ability of a publisher to provide ad metrics to advertisers
(When the ad was served by a 3rd party ad network) appear to have a higher conversion rate in order to get higher ads revenue from the ad network
Fraudsters frequently engage in “click flooding” or “click spamming” attacks. The aim is to steal credit for conversion events that the click-source had no part in causing.
Fraudsters need scale to perpetrate large attacks. One of the easiest ways to reach large scale is to distribute a malicious browser extension.
Another common way that fraudsters achieve scale is by distributing malicious mobile apps. These apps may well contain webviews that can simulate web traffic
It is also feasible for fraudsters to modify the source code of a browser and compile a malicious version of it. Malicious browsers might be used within a single base of operations (for example a “click farm” with a dozen employees), or could potentially be distributed to real people in a scaled way (for example, people who accidentally downloaded the malicious software while searching for free copies of otherwise copyrighted programs.)
Some web-pages may contain graphic imagery, or might discuss controversial topics, or include news about disaster or tragedy. Advertisers care a great deal about the types of concepts that people will associate with their brand. For this reason, advertisers might not feel comfortable having their advertisements run alongside these types of content.
For this reason, the industry has developed a lot of "Brand Safety" controls. Advertisers commonly provide ad-servers with either "whitelists" of the websites where they are happy to display their ads, or "blacklists" of websites where they are not happy to have their ads shown. Some ad-servers also support "keyword blocking", where the actual text on the page is crawled, and if certain keywords are present, the ad should not be shown anywhere on the page.
Transparency and trust in billing data, which must be auditable and produced by an accountable party. This reporting should be accurate for advertisers and publishers of all sizes and allow for reconciliation mechanisms between the two parties in case of mismatch.
Targeting is the selection of a specific audience to see an ad. There are a few types of targeting that will be particularly affected by the loss of 3rd party cookies.
As an advertiser, I want to be able to build the audience for my campaign using any combination of either:
List of users I can provide
Users that have had interactions with my website or store over a certain period, or not. E.g.: "abandoned cart users", "users that have bought baby diapers between J-30 and J-15 but not since then"
I want to be able to:
get a understanding of the reach of my audience while defining it, and through the life of my campaign
my audience to be updated in real-time. E.g. if my audience excludes people that recently made a purchase on my site, I want to stop displaying ads to any user making a purchase in near real time.
A very common theme in user feedback about ads, is that people strongly dislike seeing ads for products they have already purchased, or mobile apps they already have installed. Such ads are not just an annoyance to people, they also drive no value for advertisers.
This is implemented in much the same way as retargeting ads, by keeping track of the browsers / devices who have fired a particular conversion event, and preventing particular ads from being shown to them.
A common problem all businesses face is finding new customers. One highly effective approach is to try to find people who are similar to the business’s existing customers. These people are significantly more likely to be interested in the business’s products than a randomly selected person. This use-case is especially critical for small businesses that do not have large ad budgets, and cannot afford to spend money for a long period of time on broadly targeted ads until the system learns which type of people are most likely to engage.
One common way this is achieved today is with 3rd party cookies. Before starting to run ads, a website owner can instrument their website with code that generates a conversion event each time a customer engages with their website. In this way, an ad-tech platform can learn about the aggregate properties of the current customers of the website, and utilize this information to generate a “Lookalike Audience” of people “similar” to them. Ads can then be targeted to these “similar” people, who are more likely to find this website relevant.
A common practice in the industry today is to run ads that are shown to previous visitors of a website. While there is a lot of negative sentiment related to “ads that follow you around the internet”, this capability forms a significant fraction of digital advertising.
As an advertiser, I want to be able to select on which type of environment my ads are displayed (web, mobile web, app) and what's the target environment for my ads.
People dislike highly repetitive ads. There is ample customer feedback to support this point. Not only that, advertisers do not want to spend their ad budgets showing the same ad to the same user over and over.
While frequency capping tries to cut off the long-tail of an ad being over-delivered to a specific user, there is also the opposite side of the delivery curve where an ad is often under-delivered too. When you look at when people convert, it's often after they've seen the ad several times. There's typically a sort of "sweet-spot" range that corresponds to the best conversion rates.
Marketers sometimes choose to rapidly pause their ads in all media after an adverse news event such as a product recall or airplane crash. Publishers of any web property on which the marketer's ad might appear will need to apply the pause to any ad placement on their sites.
Publishers participate in the ad ecosystem to monetize and customize their content.
For websites that do not require a user to log-in, comparatively little (or perhaps no) information is known about the person browsing. This makes it extremely challenging to show a relevant ad. In certain cases it may be possible to show “contextual ads” that are reflective of the surrounding content, but in other cases this might not be possible.
Search engines are a very special case, where contextual ads make a lot of sense. The search term already indicates a strong interest in a specific topic. For many use-cases though, “contextual ads” will have exceptionally little contextual information to work with.
Today, publishers rely on 3rd party ad networks to serve relevant ads. These ad networks use 3rd party cookies to connect the person viewing the publisher’s website to some other source of information they have (perhaps from a logged-in website).
Irrelevant ads generally perform poorly (people do not engage with them), which will lead to reduced publisher revenues.
As mentioned above, most publishers rely on multiple 3rd party ad networks to serve relevant ads. One important use-case worth considering is how a publisher is to decide which ad network should serve an ad for a given opportunity.
Today, the industry is slowly converging on “real time bidding”. This is the optimal mechanism for ensuring fairness, equal access, and equitable treatment. In such a system, each ad network is asked to “bid” on an ad opportunity. The highest bidder wins, and is given the chance to show an ad. They must pay the amount they previously bid.
A lot of useful content on the web only exists because Affiliate Marketing provides the financial basis to support its creation.
Publisher websites provide “affiliate links” to merchant websites. If the person clicks through and eventually completes a purchase, the publisher receives a portion of the sale price.
There are many similarities with display ads, but the main difference is the payment scheme. Instead of being paid per impression or per click, the publisher is only paid when there is a conversion.
When a site/brand wants to customize the content of their web pages based off of the user having seen a particular ad or opened a particular email. This is often about a consistent, more straight-forward user experience - so if a brand shows someone a particular product/sale, and then that user shows up at their site, the user doesn't have to dig around to find out what they're interested in.
When someone clicks a link to a website, that URL might encode information that can be used to personalize the experience. It might be a link to a particular offer or product. It might be a link that is specific to a particular marketing campaign. It might be personalized to the person themself.
Publishers usually consider that some ads contain inappropriate imagery, promote inappropriate products or messages they don't want to be associated with. Each publisher needs to be able to maintain a "block-list" of specific brands, categories of brands, or specific pieces of ad creative.
Some ads may contain (or link to) malware. Publishers need a mechanism to block malware from being served to their sites' users and to take appropriate action against the source of the malware. In the event that a malvertising campaign is discovered, publishers require mechanisms to block affected ads from appearing on their sites.
Publishers use third-party malvertising detection services to prevent malvertising from serving to their users.
As a Publisher, I want to:
Have a daily detailed and accurate reporting of advertising revenues on my properties.
Have the ability to investigate and understand variations in my advertising revenues.
Have the ability to reconcile and investigate discrepancies between the reporting I get on my ad revenues and the revenues that will actually come from advertisers or ad networks.
Publishers need to be able to fulfill directly sold ad placements. Many publishers often secure higher-revenue advertising based on the ability to deliver guaranteed impressions. Publishers need to balance short term optimization of revenue with longer-term strategic client relationships, which means serving lower priced ads ahead of higher priced ads in certain cases.
Publishers need to set floor rates (a minimum rate below which an ad will not be shown, even if there is no ad at a higher rate) based on several criteria.
A publisher may not want an ad to serve at all, if it is below a minimum floor set for that page, section, or site.
A publisher may not want an ad from a certain brand, vertical or campaign unless it is above a minimum floor set for that brand, vertical, or campaign. For example, a publisher may set a "floor for automotive." Publishers choose to protect their direct sales business, by not allowing marketers to undercut their pricing by way of other channels.
Publishers need to honor contractual terms preventing competitors from appearing together. If the contextual system serves an ad for one brand in one slot on the page, the decision process in other ad slots needs to know not to serve direct competitor ads. The publisher needs to control which brands are considered competitors. (A pair of brands might be considered direct competitors in one publisher niche, but not another.)
One user need is to be presented with an accurate explanation of why a particular ad was shown. This is especially important in use-cases like re-targeting when people want to understand how personal information is shared. When no explanation is provided, people may imagine innacurate theories as to why they are seeing a given ad.
One user need is the ability to opt-out of all ads from a specific business. This could be for any reason, but examples might include a dislike for that particular company / brand, a lack of relevance, excessively repetitive ads from that business in the past, a prior bad experience with that business, or offensive ad imagery.
The compexity in serving this user need comes from the difficulty in recognizing which ads promote which businesses. A given business may run ads through many channels, through many different ad agencies, and might promote a variety of apps and websites. All of this makes it very difficult to successfully satisfy this user need.
As a user, upon request I want to easily be able to opt-out of ads provided by a specific marketing service.
As a user, I prefer to see ads that are relevant to me.
As a user, I do not want:
Ads to degrade my browsing experience, in term of navigation, e.g.: ads blocking access to content or slowing down page load.
Ads to be annoying, e.g.: get to many times ads for the same product, or ads for products I already bought.