Pioneered in the early 2000s, programmatic advertising generally refers to the use of software and data to purchase and/or sell ad space, as opposed to more manual and direct forms of buying and selling. While nearly any form of ads can be transacted this way, programmatic is now the dominant means of digital ad buying and selling.
This is the practice of attempting to gauge the impact of each individual interaction a customer has with a brand, ideally determining which ad or channel had the most impact toward driving the desired outcome, such as a sale. For example, an apparel retailer may run ads on Google search, Facebook, YouTube, and display ads across the web. Attribution modeling could help that brand determine the efficacy of each.
Each programmatic ad auction is governed by different rules and buying volumes which can affect how buying and selling play out. For example, whether winning bids are based on first price or second price, the number of advertisers, the competitiveness of the inventory, whether some brands employ pricing floors or not, the availability of the data, the level of transparency – all these variables impact how an individual auction functions. All of these will influence what kind of bidding strategies brands will employ in an individual auction.
Auction Models in Programmatic
Various companies employ different auction models, such as first-price auctions (where the highest bidder wins and pays the price it bids), second-price auctions (where the winning bidder pays a price closer to the next highest bidder), and so on. Some auctions, such Google’s search auction, factor in ad prices and potential click-through rates.
This tactic refers to a publisher helping a brand reach its audience on other sites across the web, as well as lookalike audiences on other platforms.
Historically, when ad space was bid on by multiple bidders in an auction, the winning buyer would pay only a penny more than the price offered by a runner-up (the second price, in this case). However, more recently, programmatic exchanges have shifted to a first-price auction model – not unlike an auction at Sotheby’s – in which the winner simply pays whatever the winning bid price was. However, that can end up costing brands more than they had in mind. Via bid shading, ad-buying platforms can automatically help brands adjust their bids in real time so that they win without overpaying. They might split the price difference, for example, or pay a lower price based on historical rates.
This is theoretically good for brands, but it can also limit the transparency they have in the prices they pay. In other words, the buyer gets a discount, but with some DSPs, they don’t know how that discount gets calculated. However, some platforms, including FreeWheel’s Beeswax, have begun making these fees transparent.
This is information that is passed in the bid request. Given that DSPs and SSPs help execute millions of transactions, they are able to see numerous macro trends regarding a given ad market. Some companies collect this data, analyze it, and use it to build their own data pools or even products. For example, while executing thousands of buys for multiple home furnishing companies, an adtech company could collect bidstream data and put together a package for advertisers looking to reach highly responsive furniture shoppers.
A product released by the adtech company IPONWEB in 2013, Bidswitch serves almost as a universal adapter for DSPs and SSPs/exchanges. The idea behind Bidswitch is that brands or publishers can plug into multiple sources at once without having to repeat integration steps while centralizing targeting parameters, data flow, and reporting.
Channel Attribution Modeling
This form of attribution modeling attempts to gauge the value of different marketing channels – such as search, social, mobile web, and OTT – and how they collectively and individually impact consumers.
Ads that rely on third-party cookie data for targeting purposes.
This is a fraudulent activity typically used to game affiliate marketing companies into paying commissions. For example, a person may buy something online, and a fraudster will make it appear as if they’ve made that purchase right after visiting a site with an affiliate link for that product. Fraudsters place cookies for those sites on consumers’ machines without their knowledge by using pop-ups, shady ads, and other mechanisms.
As third-party cookies are gradually eliminated from digital marketing, brands are being forced to use a variety of other signals to target consumers with ads; increasingly, those signals include first-party data, broad anonymous data sets, as well as contextually based targeting techniques.
As cookies are gradually phased out of digital marketing, brands are having to adjust how they use adtech buying tools such as DSPs. In the absence of cookies, brands can employ their own first-party data, segment data, or macro audience data to bid on ad space using DSPs.
As the digital advertising world braces for the elimination of the cookie over the next few years, there are various cookieless targeting initiatives underway aimed at helping digital media retain its targeting power. Many in the industry are advocating for the use of more direct identifiers to replace cookies, such as hashed email addresses or wireless numbers. Yet on the other end of the spectrum, many are predicting that advertisers will begin targeting much broader groups of web users via probabilistic data and more use of sophisticated data science.
In addition, companies such as LiveRamp, The Trade Desk, and Google are attempting to develop new cross-industry cookie alternatives.
CTV vs. OTT
Definitions across the industry vary, but generally speaking, OTT (over the top) refers to video content that is delivered to TVs without a cable box. For example, when a cable subscriber uses a streaming service to watch a show, rather than their set-top box, they are going over the top. CTV (connected TV) commonly refers to content consumed via apps on smart TVs or through devices like Roku and Amazon Fire. There is some debate over these definitions.
A growing number of brands have either partnered with adtech companies or worked on their own to develop customized demand-side platforms. These buying platforms often contain features that are built specifically for individual advertisers, and even their own custom algorithms that can be used to bid on particular audiences or to identify unique patterns or behavioral segments that advertisers may care about.
Data Management Platform (DMP)
A DMP is a centralized platform that allows advertisers to create target audiences based on a combination of in-depth first-party and third-party audience data.
Demand-Side Platform (DSP)
This refers to software and tools developed exclusively for the buying side of the ad business to purchase inventory programmatically. Demand-side platforms can essentially help advertisers sift through thousands, if not millions, of potential ad impressions across sites and apps across the web – all in one centralized platform. The term “DSP” can refer to the software itself or the companies that sell and manage the actual software.
Exchange bidding is similar in concept to header bidding in that it levels the playing field among potential advertisers bidding on inventory. But, instead of the bidding taking place on an individual site, the auction occurs within an ad server. This is often seen as Google’s answer to header bidding, as the company now handles all the transactions between
SSPs and Publishers
In a perfect world, this arrangement reduces the complexity – and resulting latency inherent to header bidding. However, some industry insiders are uncomfortable with the idea of Google hosting such auctions, while the company is also commonly a bidder itself via the Google Display Network.
Typically, first-party data refers to registration data collected by marketers, or login data collected by publishers. However, publishers can also employ first-party cookies, which allow them to create profiles of audience groups, such as auto or tech enthusiasts.
When impressions go up for bid in a first-price auction, the advertiser, via a DSP, ends up paying the full price of whatever they bid, regardless of whether they could have paid a much lower price to beat the competition. If you bid $5 for an ad impression, and everyone else bids $1, you pay the full five bucks.
Header bidding for publishers opens the ad marketplace to more bidders, and no one single bidder has any clear advantage – theoretically. Would you rather have one person making an offer on your house, or 10? However, the flip side is that putting every single bid up to multiple advertisers causes a lot of volume for the publisher and its adtech partners to manage at any given moment – potentially driving up costs while taxing sites.
Historically, when various adtech companies would bid on potentially running an ad on a publisher’s site at a given moment, each one would essentially have to get in line, as auctions took place within an ad server (usually Google’s) and bids would go out to one potential buyer followed by another. Companies such as Criteo and Google could arrange deals with publishers that would assure they’d be first in line in this process.
By the mid-2010s, companies began to develop a solution called header bidding. By plugging in code on a publisher’s header, each outside source of ad demand was placed on equal footing. The bids take place before there is a call to an ad server. Companies ranging from AppNexus to Index Exchange have built commonly used header bidding “wrappers” to facilitate demand.
Short for “Identity for Advertisers,” IDFA is a randomly generated device identifier employed by Apple for devices using iOS. IDFA was designed to help app developers determine which ad channels are generating leads, downloads, etc. But third parties such as attribution companies and ad networks had been taking to collecting IDFA data and using it to build their businesses. This use case was not exactly what Apple had in mind. Apple phased out IDFA, which impacted the entire mobile app marketing ecosystem, from gaming companies to third parties to Facebook. Without IDFA, brands and app companies have been left in the dark when it comes to measuring the specific efficacy of mobile ads.
Incrementality is a tactic which aims to help advertisers isolate the impact of different ad budgets and partnerships, and then helps them determine the value of adding more budget and/or new partnerships. For example, a heavy social media advertiser might want to calculate how many more people it would reach by adding a new platform to its buy – and what that new reach would cost relative to the rest of the media plan.
This is revenue that is driven by an ad campaign that can be determined to be additive to a company’s typical business during a period.
Instream video ads appear prior to or within streaming video, including pre-rolls, midrolls, and post-rolls.
In-House Programmatic Buying
A growing number of marketers have elected to build their own internal teams of programmatic buying experts and analysts. These marketers can work directly with adtech companies, using tools such as DSPs and DMPs, and buy media on their own, without needing an ad agency. More advanced advertisers have even developed custom software and proprietary algorithms.
Several companies use IP addresses to target people in specific locations – also known as “geofencing.”
This is the act of targeting consumers using IP addresses, which are anonymous identifiers tied to specific devices, including desktop computers and smart TVs.
IP Targeting Cost
IP targeting generally costs more than broader-based digital ad targeting, since brands use this technique to target very narrow, specific sets of people, and typically the performance of these campaigns is more effective than normal.
IP Targeting Software
Companies such as Canopy and Accelerant specialize in helping brands match up specific IP address data with ad targeting tools.
IP targeting in advertising enables brands to match up targeting data with a specific device or even physical address, usually anonymously. IP targeting can be used to target ads to specific households via cable boxes – such as households that have just purchased a new car, for example. Brands also often use IP targeting to reach people in a particular location, such as in the vicinity of car dealerships or quick-serve restaurants.
This is the general practice of using unique, anonymous device identifiers – often randomly generated numbers or triangulated data segments – to target individuals or households with ads.
This is a blunter form of attribution modeling which provides the entire credit of a successful ad conversion to the last ad a person clicked. This tends to discount other ad exposures, and heavy favors – or over-credits – search ads.
Publishers and advertisers often want more in-depth information on programmatic ad campaigns beyond just the number of bids won and the prices they pay. To better understand why various auctions are won or lost, or why certain campaigns perform better than others, both parties like to dig into log-level data. This often includes basics such as location data and time stamps. Log-level data can be provided by ad-serving companies or DSPs. However, in most cases this information is only available from a small subset of overall auctions. Some adtech platforms, including Beeswax, can supply customers with 100% of their log-level data. Overall, the driving idea behind exposing log-level data is to gain as much transparency as possible, particularly regarding where a brand’s budget is being spent.
Pioneered by adtech companies such as Quantcast, lookalike modeling seeks to take small sets of targeting data and anonymously identify larger audiences or sets of data sharing similar characteristics. For example, retailers might seek a larger group of tech enthusiasts based on a small set of recent shoppers.
Machine Learning Attribution Model
Traditional attribution modeling requires people to set rules. That often means that subjective humans set up rules regarding what matters and what doesn’t (search ads are worth X, ads shown over a week ago are less valuable, and so on).
Increasingly, adtech companies are employing machine learning (ML) for attribution. Theoretically, ML can help attribution models objectively figure out over time what criteria should be used to assign value to various signals – and machines can pull data from across the web to do so, meaning the intelligence employed is not limited to a particular campaign.
Starting in 2015, a few big media companies began launching live TV packages delivered via the internet. These so-called “skinny bundles” have been aimed at the growing number of consumers who are either opting to drop pricey cable packages or who have never subscribed to a cable bundle. However, OTT IPTV (internet protocol TV) has had mixed success; some products have shuttered, and cord-cutting overall continues to accelerate in the U.S.
This is video delivered to TV screens via the internet, often via apps. This can include non-ad-supported services as well as ad-supported services.
Companies such as Teads and Unruly Media deliver video ads to text pages. These video ads only stream when a person scrolls down to a particular part of a web page or app. The idea behind outstream ads is to create valuable video inventory within primarily-text environments since the supply of instream video can be limited.
Private marketplaces (PMPs) are not unlike an invite-only ad exchange, where a distinct set of advertisers might be invited to bid on inventory from a single large publisher. Usually, a pricing floor – a minimum ad price – is set, and then advertisers bid for space.
Programmatic Bidding Dynamics
Media buyers employ a variety of criteria to determine what price to bid on an ad and when – but in most cases, these prices are static or based on a definitive set of instructions (like say, “if a new mom who’s in the market for an SUV shows up on this site, bid $5”). Rather, in real time, bidding dynamics can change based on the availability of target audiences, competition for inventory, historical pricing, and ongoing changes in performance, among other factors.
Some marketers want to purchase specific ad space on specific sites but prefer to use programmatic tools. Often these deals include a fixed rate, rather than auction-based bidding. They are designed to be easier to manage and reserved for premium inventory.
While many marketers have handled some aspects of their advertising output without the help of ad agencies, there has been a decided push to bring more of this activity in-house starting in the mid-2010s. This has been driven in large part by the widespread availability of programmatic buying platforms as well as marketing cloud software – and the importance of customer data. Marketers increasingly don’t want to hand off their most precious consumer data to agencies that may come and go. Thus, many have built specialist programmatic buying teams and license software from DSPs or other companies. The most sophisticated brands even build their own algorithms for ad buying purposes.
This is similar to programmatic direct deals, but in this case ad buyers have the option to buy inventory at a negotiated price but can chose not to – as the inventory has not been reserved.
A few large firms on both the buying and selling side of the ad industry have built full stacks, i.e., sets of products geared to service multiple aspects of programmatic transactions. For example, Google has its own ad server, DSP, and multiple exchanges. Similarly, AT&T’s Xandr features multiple tools for both buyers and sellers. The theoretical advantage of stacks is that they are built to work together, using common data sets and processes. But they also theoretically give one company a great deal of market power and access to pricing data, causing potential conflicts of interest that have more recently come under scrutiny.
The growth in header bidding and open bidding has dramatically increased the number of potential buyers and sellers for each ad that is served, meaning that adtech companies are processing and paying for thousands, if not millions, of bids that go nowhere. Generally, QPS filtering is an agreement between SSPs and DSPs to limit the sheer volume of queries per second. In other cases, QPS filtering can be performed on a per-customer level as with products such as FreeWheel’s BaaS™platform.
A growing number of adtech companies are offering either white label DSP products or self-serve DSPs (also called “self-service DSPs”) that don’t require big contracts, custom integrations, or huge spends. These can be employed by big advertisers that have brought programmatic buying in house, or even small brands that only require DSPs on occasion. Facebook and Google’s self-serve ad buying products are technically custom DSPs.
When impressions go up for bid, the winning bidder pays a rate that is just above the second-highest bid. This theoretically saves advertisers from significantly overpaying.
Some companies, when licensing SaaS products, will require a custom set of tools and services – they are the “single tenant” of that piece of software. In other cases, multiple companies will employ the same software, i.e., a “multi-tenant architecture.”
Supply Path Optimization (SPO)
This is a practice via which media buyers attempt to take more control over their ad supply chain through demanding more transparency, cutting more direct deals with publishers, and overall looking to eliminate as many intermediaries as possible. In other words, media buyers attempt to utilize the most direct path to inventory possible. This is both an attempt to provide brands with more visibility and control over their ad placements, while also reducing needless fees and latency.
Brands typically employ supply path optimization to reduce the number of adtech partners they purchase from, eliminate unnecessary steps in an individual ad buy, and increase transparency. If the goal is to create as direct a path as possible to desired inventory, advertisers often designate specific supply partners for SPO, such as preferred publishers, exchanges, and SSPs.
While the goal of supply path optimization is ultimately to provide the most direct, efficient, and cost-effective path to ad inventory, there is potential for waste in this buying approach. For example, brands can still end up bidding on the same impression via different exchanges, even if they are actively trying to avoid duplicative queries and multiple intermediaries.
Supply-Side Platforms (SSPs) refer to software and tools designed for the selling side of the ad business, such as web publishers and large media organizations. SSPs can be used to corral and package inventory from across an array of sites and apps and make it purchasable for DSPs or individual advertisers. For instance, an SSP might pull together ad space featuring thousands of impressions targeted to recently married college graduates – all in milliseconds.
Third-party cookies are small pieces of code that are automatically placed on users’ desktop computers and laptops by web browsers and publishers, and are used to anonymously recognize and track these users. Cookies help publishers recognize registered subscribers, for instance, and help companies remember frequent customers – not requiring them to log in during each visit. Cookies are often frequently used in ad targeting. However, cookies are generally ineffective on mobile devices.
TV Attribution Model
The TV industry has spent the past several years trying to match digital media’s ability to track consumers, and more importantly directly prove business outcomes from TV ad campaigns. It’s inherently more challenging for a number of reasons – TVs don’t employ cookies, people don’t log into individual TV networks, people don’t shop via TV screens, etc. But the industry has been making big strides. Companies such as Data Plus Math and Neustar have built models designed to both connect digital user data with TV audience data, and to track the effect of TV ads. For example, packaged good marketers might use several of these partners to target frequent big-box retailer shoppers, manage ad exposures across multiple screens, and attempt to match up that audience information with real purchase data from loyalty cards used in stores.
Companies such as iSpot pull data from TV manufacturers which can tell advertisers exactly which ads individual households were exposed to – and that data can also be tied to other identifiers and conversion metrics.
For many years, digital ad impressions were tracked, bought, and sold based on the number ads served, regardless of whether those ads were completely visible on a web page or app, or if those ads were even fully served at all. But starting in the mid-2010s, advertisers began demanding to only pay for ads that are fully viewable, and several vendors emerged, such as Moat, promising to track how much of an individual ad is viewable on a web page. Eventually, the Media Rating Council established a benchmark for transactions in the industry – at least 50% of display ads must be visible for one second, or two seconds for video.
White Label Ad Exchange
While several tech giants manage large, cross-industry ad exchanges, companies such as Crystall.io, SmartHub, and Ubidex enable companies to create private, proprietary ad marketplaces for specific sets of marketers and publishers.
White Label Retargeting
While some adtech companies such as Criteo specialize in retargeting specific audiences across the digital ecosystem, companies such as AdRobin, StackAdapt, and SmartyAds enable ad agencies to build their own retargeting products via licensed technology.