Introduction to AI in Paid Media

Today, paid media is on the verge of astounding transformation, primarily driven by large language models such as the widely popular ChatGPT. These and similar generative AI services are predicted to dramatically reshape not only paid media — but marketing, public relations, customer services, journalism and in fact, all corporate communications — as companies scramble to reoptimize their push-and-pull communications for the new conversational nature of brand-customer interactions. These interactions are increasingly occurring in real-time, resisting brand throttling of media engagement and more closely mirroring human-talking-to-human dialogue. What will emerge as a result of these shifts is commonly referred to as marketing at the speed of AI, with the core of every marketing layer — from paid media to email to website to customer journey to social media presence to PR — being AI-generated.

With AI’s ability to enable rapid alterations of all creative content, plans and strategies, corporate communications can now be digitally sculpted and instructed to be engaging, optimized or bluntly raw, for almost any target audience, regardless of what stage of brand engagement, customer journey motivation or intention, they happen to be at that instant. And the payoffs from these evolutions could turn out to be phenomenal: reducing timelines of creative content production for paid media, website, social, PR and customer journey optimization from days and weeks to mere seconds or minutes; allowing each element of corporate communications to be audience-mode specific or personalized; and allowing rapid in-motion alterations of creative messages in response to real-time reports of success or failure gathered from integrated data analytics systems. The possibilities are endless.

2. The Evolution of Paid Media

While it may seem that paid media today is merely a tool for brands to buy visibility or user data, it used to be the only way to control content dissemination, and owned channels did not have the capacity for creating a substantial impact. However, media spend has been decreasing over the past decades since the emergence of the Internet. Owned and earned media have gained importance alongside the declining cost of content creation and distribution, thus giving rise to integrated media strategies that try to combine the strengths of various media types. Other than the categorization of media spend phases that we describe in a coming section, we identify two major stages of evolution in paid media strategy on the basis of the internet’s advertising effect on paid media. The first stage extends until 2000 approximately. In this stage, paid media are not consulted during decision making by the consumer, nor are they used for assisting conversion during online shopping, which is usually preceded by product research.

The second stage is the current phase, which has not had a substantially different period as of yet in the past two decades. In this phase, consumers consult paid media when making purchase decisions, and paid media help conversion during research and shopping. An important development is the proportion of the media channel in digital marketing strategy that is accounted for by paid media. In our opinion, this proportion has actually been declining drastically, due to a variety of factors. Digital marketing has become almost an altogether paid environment that any brand endeavoring to be visible has to engage in, making marketing activity account for a substantial part of the sales of established companies and startups. Nevertheless, we realize that marketing budgets are mostly stagnant and what is left for paid media strategies is rapidly declining budgets.

3. AI Technologies in Advertising

To allow paid media to perform well with less and less human management, cutting-edge ad products are integrated with a variety of AI technologies that are now heavily used in ad functionality, and AI-conceptual products are being born in quick succession. Among them, especially in the leaps and bounds of progress that is made possible by large language models, the innovation is likely to have a huge impact on advertising. In the paid media industry, it is interesting how popular AI will be used to create ads, and how ad reviews will change due to image generation technology and large language models. Technology is progressing in the right direction although it is debatable if whether large-scale language models that are currently available can actually understand the concept. Three technologies that have been in use in advertising before the recent AI boom are machine learning, natural language processing, and computer vision. Machine learning, a subfield of AI, has been used to detect which advertisements are likely to convert a user by estimating the probability of conversion, optimizing the delivery of ads to maximize conversion, and automating the write and design ads with a lot of ad variation while optimizing ad performance. From the point of view of ad management, machine learning technology can find the optimal setting that breaks away from the limitation of human resources and the nature of the algorithm and rules of the advertising platform. For example, at the beginning of the ad delivery period, ads are delivered by matching a combination of ad settings prepared by the campaign manager, but after being observed for a period of time, machine learning technology automatically selects the combination of settings that is most likely to deliver the best performance.

3.1. Machine Learning Algorithms

Machine learning describes a range of programs that use algorithms to analyze large amounts of data and interpret the results, often to make recommendations or predictions about future behavior. The technologies coming under this umbrella have been used for many years in many industries. As advertisers have become able to gather and analyze larger and larger amounts of consumer data, and as consumer choice becomes more fragmented, driven by the growth of large retailers and new platforms, drawn both by their consumer marketing and ad modeling tools, advertisers have increasingly turned to machine learning technologies to help them make sense of this data and this unpredictable postmodern consumer.

A major use of machine learning in advertising relates to improving the efficacy of marketing campaigns by refining the targeting of display advertising in paid media. Successful targeted display advertising depends not only on finding consumers who look like likely purchasers of a brand’s product, but also on getting them to see relevant ads at the right moment and to take the desired action. Machine learning can help with all aspects of this process. It can help brands assess the predictive value of the various sources of data they have on their likely purchasers and refine their audience segments based on that predictive value. Marketers can use machine learning for lookalike segments and other predictive segments. It can help with all the predictive modeling used for campaign planning. And once the ad campaign is underway, algorithms can help determine the advertising spend more accurately for every publisher and moment of the campaign. Companies can more accurately predict short-term outcomes like in-store visits or match a particular campaign’s predictive model against real-time behavior to make campaign adjustments and send relevant ads to consumers.

3.2. Natural Language Processing

Natural Language Processing or NLP refers to the use of techniques from the fields of artificial intelligence and linguistics to analyze and process human languages. It is a subfield of AI working to make it possible for machines and humans to communicate with each other using natural language instead of formal languages, like programming ones. This field covers many different areas, including speech recognition and generation; the automatic translation of one language or dialect to another; extraction of meaning, semantic structure, and relationships from text; information retrieval, storage, and organization based on linguistic content; and support for human language learning. NLP is often combined with other AI technologies, such as knowledge representation or computer vision. However, its most recognizable uses are in the areas of machine translation and automatic speech recognition.

NLP has a wide range of applications in advertising: predicting stock price fluctuations based on sentiment analysis for financial ads, improving search marketing through concept expansion, ad targeting, and keyword selection. While the techniques and methods we apply to NLP tasks continue to evolve—NLP is driven both by theoretical advances in the AI and ML fields, as well as advances in the availability of data to both learn from and apply the models to—there are a number of tasks that we frequently need to solve. These task areas can be generally represented under the three broad headings of categorization, extraction, and generation. In the categorization area, tasks include text classification, document categorization, and genre detection; in the extraction area, tasks include named entity recognition and bioinformatics; and in the generation area, tasks include speech synthesis, dialogue system, and machine translation.

3.3. Computer Vision

Visual content is massively shared, viewed, updated, modified, discussed, and consumed each second through social media sites. If text were the only means of communication, there would be no social network except for the ancient letters exchanged with wax tablets. This gigantic amount of visual content is responsible for the evolution of neural networks and deep learning. Actually, the family of artificial neural networks called Convolutional Neural Networks exists to process image data, following two simple principles: local correlation and dimensionality reduction. No other human-made technology has been able to internalize and learn the visual world as deeply as CNNs.

Using computer vision models to search for information in websites or on e-commerce platforms, companies can perform tasks to enhance the understanding of digital media, expanding insights on how customers search for and interact with brands, products, and services in digital touchpoints. Computer vision can become a groundbreaking solution for brands and paid media strategists in two important ways: understanding how digital media is populated by visuals and exploring new ways of using digital visuals. The first way is based on image recognition tasks that use CNNs, requiring that the brands properly label or tag their own images in their owned media. The second way makes use of models that understand, segment, and predict cross-modal aspects of digital visual content. The tools being developed for cross-modal processing enhance the analysis and understanding of digital visuals and have the potential to transform the way paid media strategies are developed.

4. Targeting and Segmentation

4.1. Behavioral Targeting

Paid media has traditionally relied on crude psychographic and demographic categories. In the digital world, more sophisticated behavioral targeting techniques have emerged that assess user activity prior to visits to provide a more accurate reflection of interests. Data companies aggregate large numbers of behavioral data points across the web to build databases of user interests that can be harnessed to target consumers. Behavioral targeting is particularly useful in biddable media environments.

While targeting based on user activity has proven effective, it does have its limitations. Most prominently, it is not deterministic; that is, it cannot predict what activities consumers will perform in the future after they see the ad. Because user behavior is constantly changing, interest levels can shift, rendering knowledge of past actions relatively uninformative. Predictive targeting applies statistical analysis to past behavior data combined with a variety of other data sources to formulate probabilistic models of future behavior. These models can help marketers identify which prospect groups are the best segments for targeting, and which specific individuals are most likely to convert. Predictive modeling works by analyzing statistical patterns in data on existing customers to understand the factors that drive conversions. The model is then fitted to a wider pool of potential customers to get scores on who is most likely to purchase in the future.

4.2. Predictive Analytics

Predictive targeting is a more sophisticated approach that goes beyond past behavior to allow marketers to make probabilistic statements about the probability of future behavior. At its most basic, predictive modeling uses historical data to identify which characteristics are the best indications of likely future buyers, and then builds a statistical model that can be generalized to the entire population. Predictive analytics can be based on a broad variety of inputs beyond prior behavior, which can include data from model building experts, customers, partners, and internal databases. However, predicting future behavior is a much bigger challenge than simply clustering customers with similar traits together.

4.1. Behavioral Targeting

Behavioral targeting aims to deliver ads tailored to consumers, which can increase ad response, clicks, conversion, and ROI. Different targeting methods choose different characteristics to define the segments to focus on, and how the characteristics influence ad efficacy. These characteristics include demographics, past purchasing behavior, site visits, and ad response. Targeting is crucial for paid media optimization because paid media marketing budgets are extremely large, and the ad is served to a paid audience who did not initiate the interaction. Thus, ad relevance needs to be considered at a larger scale beyond the clicked-on audience. Paid media marketing are dependent on both marketers and advertisers. Marketers willing to spend money for advertisement are expecting profits that exceed the spending; therefore, advertisers should present ads that cater to user preferences and are expected to convert. When managed properly, such interaction of the two parties leads to a win-win situation, and paid media channel can achieve outstanding performance. Behavioral targeting can be performed by advertisers or by trusted third parties specialized in such function. Behavioral targeting can achieve high performance because it uses real-time data on user activities. The definition of real-time, as well as how much information would be enough to perform behavioral targeting effectively, is determined by the lingua franca of online advertisement sharing the query.

Behavioral targeting is a form of predictive analytics that leverages all engaged parties’ knowledge, voluntarily or involuntarily, so that advertisers can serve the right ads to users during their activities on a different yet related platform, such as a display or video ad served by a publisher platform. The metrics used by advertisers and marketers are campaign click-through-rate, display conversion, video view-rate, video interaction, and video conversion rates. Conversion and interaction refer to offline conversion or interaction, video interaction and conversion refer to online interaction, and display interaction and conversion refer to cross-channel attribution. Overall, behavioral targeting, when backed by sufficient user data, can deliver a targeted experience that is much closer to the user goal than that of conventional contextual targeting alone.

4.2. Predictive Analytics

Predictive analytics assess a population based on the likelihood that each individual in that population will complete a purchase or take a desired action. Predictive budgets answer the question, “What will happen?” and make recommendations on the best way to allocate resources to achieve the desired outcome. Predictive models are usually limited to transactions that have occurred.

Predictive models based on probabilities work for prospecting and recency models are primarily used for budgeting. But recency models can help marketers understand the act of mail prospecting. Probability models can help you answer the question, “What will happen?” Predictive response models identify optimal segments for targeted marketing programs to achieve response, incrementality, and revenue; constrained budget and channel allocation models predict optimal allocations by channel; revenue prediction models predict gross predicted spend per person; product posting models uncover faster selling products and prohibitential marker-making.

Predictive models that are segment-specific and only imperfectly give the best prediction results, or “real equity.” Decision tree models are an effective means to build segments. Recency models look at the expected number of purchases in the next inter purchase period. Latent variable models use additional demographic or psychographic information to help explain purchase likelihood. Transactional data provides a rich set of purchase behavior information that can be filtered down into a large number of segments; these segments can then be assigned probabilities or averaged over to build joint prediction segments.

5. Creative Optimization

Artificial intelligence is helping marketers understand what content drives results, allowing them to deliver messages that resonate with their target audiences. AI achieves this in two main ways, the first being Dynamic Ad Creation. While various techniques of ad personalization have existed for a long time, this task usually required lengthy and costly processes. Since the dawn of the digital age, brands have been using customer data to create segmented paid media strategies based on people’s age, gender, or online behaviors. But what if marketers could develop entirely different campaigns for dozens of different audiences with little to no human input? Enter AI Dynamic Ad Creation. Brands can feed AI model databases with multiple creatives, copy including calls to action, and past campaign data for dozens of audience segments. AI analyzes which combinations of visuals and text achieve the best results for different demographics, then automatically selects the best creatives, displays them to the right segments, and optimizes their performance in real time.

The second way in which AI is facilitating creative optimization is A/B testing automation. By automating A/B testing, a process that has proved vital for successful campaigns but is tedious, time-consuming, and often brings suboptimal results, marketers can spend less time on testing and learn more. Marketers have traditionally tested some variable for a certain amount of time based on their impressions, clicks, or conversions in order to conclude which version has performed the best. Then they would take the winning ad and do it all over again, this time testing a new element. But natural fluctuations and customer habits can easily bias results, leading to inadequate learnings. What’s more, with multiple campaigns running at the same time, testing even one element can take an unreasonably long time. With an endless range of combinations of visuals and copy, parallel testing ads can be unrealistic too and often lead to wasted budget and suboptimal performance.

5.1. Dynamic Ad Creation

Dynamic Ad Creation

Paid media creative is traditionally a labor-intensive process that takes considerable time, effort, and people to develop a large suite of assets. These assets may include static, animated videos and development of microsites. With all of its problems, creative development was relatively insulated from any automated process, save, perhaps, procedural tools for templated page layouts. Not any more. Powered by a plethora of mechanisms for generative AI, it is now possible to create digital assets with a fraction of the resources and time that would normally be required, all while minimizing the risk of attribution fraud via click to conversation. Be it digital images or videos, marketing copy, or responsive long-form landing pages, generative AI has the capability to create ads that in terms of quality and quantity would have traditionally been impossible. Why? Because generative AI can quickly create and refine assets based on audience feedback and conversion rates.

A Dynamic Ad Solution Use Cases
Many luxury brands use paid media to drive in-store sales. But what happens when consumers walk through the doors? The luxury retailer invests significant resources developing the ambiance of its physical space. What are ad wins when a consumer who conducts an online search for a Gucci bag is then served an ad on an ad network with either an image or video of that handbag, coupled with copy bragging about its limited availability, secure checkout, fast shipping, and exclusion of copycats and fakes. For a small investment of time and money, this short funnel revs up.

The above images exemplify some of the simpler implementations of generative AI, but what if a lead-generating paid file conversion has produced a long list of email addresses to which you are sending three or four newsletters a week? It would be even easier to keep your subscribers engaged by using audience interests and viewing data to dynamically test new content in near real-time statements and suggestions: for motivation and feedback are a win-wingunaical messages, product display ads, or promotion emails that target an individual subscriber’s historic or current topics of interest- Would boost email open rates and retention and cut the unsubscribed.

5.2. A/B Testing Automation

Paid search advertisers have used A/B tests to optimize ad campaigns for as long as paid search has been around. Companies have made it easy for advertisers to poke ad creatives into an A/B testing tool, wait for recommendations, and then generate millions of dollars in extra profit. But these recommendations are still based on predictions of performance over a fixed period rather than understanding of how to dynamically serve the “winner” most of the time a user will see the ad. Besides, A/B tests can take a substantial length of time to reach conclusions. The computationally heavy nature of these two types of problems—exploiting estimated performance and serving the ad that is expected to perform best at a given moment—has hampered how much these companies can automate PPC ad copy testing.

Automating A/B testing of text ads is hard. In practical situations, text ad delivery is a stochastic process in which an ad is shown to a user depending not just on advertisers’ requested bids shown to the user, but also on the bidding algorithms used by the PPC platform. There is also no closed-form solution for determining optimal ad exposure frequencies. Another problem is the time required for estimate precision; the more times an ad is shown to users, the more precise the estimate, but less ad exposure means less precision. Learning estimated conversion differences might also require concurrent A/B tests measuring different keyword groups, with possibly mismatched group sizes. While it is important to exploit knowledge gleaned from these A/B tests to serve the best ad, it is also essential to explore less-exposed ads if the estimate is uncertain, and of different text ad memory or keyword groups, to collect better and more precise estimates.

6. Real-Time Bidding and Programmatic Advertising

In the world of paid media, ad placements, and the media that forms the various touch-points of any marketing campaigns are bought when a campaign is created by a marketer. These ads are then handcrafted and sent over to the various ad platforms for approval, while the platforms decide where to show the ads to deliver the best results for the advertisers. Programmatic or Automated Marketing or Ad-Tech takes this a step further by automatically placing the ads using AI-based algorithms which can decide the bidding price for the ad placement in real-time considering various factors such as the advertiser, the advertiser’s history and budget, the probability of success of the ad, past CTR and other characteristics of the web pages, and supply-demand attributes. Real-time bidding is one implementation of programmatic advertising involving auctioning off the rights to show an ad to an audience on a specific webpage at the exact moment that the webpage is loading by the user.

The auction is held in a few milliseconds which is enough time for the algorithms of the bidders to analyze the ad spot, the user, and all other multi-dimensional factors that can influence the CTR and decide on the bid price. The biggest benefit of RTB is that the advertisers get access to the best ad spots converting at the lowest rates, while the release of ad booking inventory becomes more efficient for the publishers. Other technological advantages of RTB include heuristic learning, targeting ads to specific audiences, and working on a bigger scale with more ad spots and advertisers. There are dozens of ad-tech startups and businesses leveraging RTB to continuously optimize ad conversions improving on the decades-old advertising formula moving digital dollars from above-the-line to below-the-line.

6.1. Overview of Programmatic Buying

The advertising method known as programmatic buying revolves around the automated purchase of ad placements through technology systems. The processing of orders generally happens by making ad placements via an advertising exchange. The most common model of digital advertising, display ads contribute to a substantial revenue of online advertising. The evolution of digital advertising has brought about a two-fold growth model where mobile ads and programmatic buying of display ads are the key drivers. Mobile advertising has become a critical industry just as print advertising and, for the first time, social networks and native contain advertising have also arrived on the scene.

Programmatic buying allows advertisers to save time which would otherwise go into negotiating ad placement deals since the process is automated. Automated ad networks offer advertisers a dashboard system through which they select their campaigns along with targeting parameters. The dashboard also provides real-time metrics monitoring. Advertisers do not have the responsibility of keeping up with all the websites that attract their target consumers or fret over how often their target consumers actually see their ads. With programmatic buying of display ads, ad placements are optimized and the advertisers’ resources are used in the most efficient way possible. Programmatic purchasing enhances the behavioral targeting capabilities of advertisers as well as their ability to link their display campaigns to on-the-ground sales figures. It provides the capability to establish a real-time link between online consumer behavior activities and occasional offline purchases of products. Improving testing capabilities are becoming increasingly important as advertisers elect to spend more of their budgets on programmatic display buying.

6.2. Benefits of Real-Time Bidding

Programmatic advertising has grown into a multi-billion-dollar industry, mostly thanks to the development of RTB, one of eight forms of programmatic advertising. Several unique properties of RTB make it particularly appealing for advertisers. These properties of RTB help to accomplish most of the objectives cited in the previous section. One is the efficiency of being able to set a single campaign online and then let RTB distribution take care of the rest. The RTB translator also ensures that the campaign meets any reach objectives—high or low at the discretion of the advertiser—requesting a minimum number of unique impressions per day either globally or by subnet. Timeliness is another advantage: online purchasing allows for specific ads to appear at very short notice soon before someone embarking on a business trip to the city of the ad’s destination. RTB also has the advantage of rich targeting capabilities, allowing for highly specific audiences to be reached. For all these reasons, ad agencies have implemented RTB for many of their online and offline media campaigns, including out-of-home ads associated with mobile activity, especially when business travel in that city is of primary importance. Most advertisers view the adoption of RTB positively, and research has shown ad performance improvements in using RTB. One of the key elements of successful RTB utilization is the linking of display ads to promoted business products and services. Most of the general negative feeling about RTB has to do with the dishonest advertising practices that have characterized display ads in the past. Brand advertisers find great utility in using RTB to promote a cultural event associated with a bottom-up community event. Display ad block rates have come down from levels above 50% to around 25%, due to the increased targeting accuracy provided by RTB.

7. Data Privacy and Ethical Considerations

The marketing and advertising industries widely utilize consumer data to reach the appropriate audience, at the right moment, with the ideal message on the perfect channel. However, as technology becomes rapidly more developed and advanced, so does the utilization of data on a macro-consumer scale. This evolution has given way to growing concerns for data privacy, data misuse, and the negative impact that a pandemic, economic recessions, and wars have globally on the human psyche. To this end, it is requisite and warranted that digital marketing fields honor and preserve the rights of consumers, in the digital and analog worlds, by being compliant with existing standards while also implementing ethical practices in the use of artificial intelligence tools and technologies.

More specifically, there are regulatory bodies present across the globe whose primary focus is to ensure that companies are compliant with privacy standards that advocate for the ethical treatment of consumers regarding their identifiable data and information. In the European Union, this is primarily the General Data Protection Regulation and is enforced by government-appointed Data Protection Authorities. The regulation goes into effect when a company or entity conducts business in the EU or if it offers goods or services that are accessible to persons residing in the EU. In the USA, the California Consumer Privacy Act serves a similar function as it pertains specifically to California residents, and other states have been in varying phases of drafting and passing similar data privacy legislation. Third-party are prohibited from collecting and selling consumer data without providing proper notice and attaining affirmative consent from consumers. The ethical use of AI in advertising requires the use of unbiased technology so that all iterations of AI advertising methods deliver equal and trustworthy messages to all consumers.

7.1. GDPR and CCPA Compliance

As discussions around data privacy and collection provoke societal scrutiny, privacy laws enforce compliance. Advertisers and brands that leverage first-party data for retargeting customers via paid search or social, email marketing, or DSPs must prioritize partnerships with ad tech vendors that comply with these laws.

In adopting these regulations, the European Parliament sought to protect data subject privacy and streamline the UI associated with digital marketing. For example, these laws give users control over how companies use their data, such as to show them targeted ads. To effectively customize, brands must collect, retain and leverage first-party data that may include email addresses, payment records, shopping preferences, behavioral patterns, etc., which are all considered sensitive personal information. With the enforcement of legislation like these, companies must ask for and obtain consumer consent before tracking their online activity with paid media. As a result, websites must include a cookie notification banner on the home page. Once consumers accept the terms employed by a company for tracking their online behavior, a cookie will be placed on their web browser. Cookie banners typically provide a link to a detailed cookie policy hosted on the brand’s website.

Under both laws, companies must disclose how they track user behavior with cookies in their cookie policy and also provide a full disclosure of, as well as an opt-out of, users’ data being sold to third parties. Unlike one law, which applies to any businesses that consume data about residents, regardless of where the business is based, another is legislation that covers only businesses that meet specific thresholds.

7.2. Ethical AI Use in Advertising

While many organizations focus on the life-saving and life-enriching aspects of the development of AI technologies, there are areas in the development of those core algorithms — as well as their use in practice — that need more and deeper works with much stronger fairness constraints. Research has shown that the inputs given to machine learning systems, the reviews of online products, the pixels in a photo, can themselves reflect human biases and stereotypes. The algorithms digest these inputs and copy their patterns into the decisions they make, perpetuating biases that exist in the real world. These negative repercussions of AI use pertain to data sets and to algorithm and model design. In the context of advertising, there are ethical implications in the choice of audience segments for an advertising campaign, as discriminatory choices provide different groups of people with differing access to critical information in life (free housing, job opportunities, etc.). While using AI to inform advertising choices, advertisers should necessarily practice moral values, such as decentralization, maximization, fairness, and privacy.

At the same time, AI use in advertising has an impact on small content creators and small businesses. Overviewing the topic, a framework to characterize and classify generative content creation based on key values is presented. By this framework, the role or responsibility of AI for content creation can assume three different levels: AI-assisted, AI-augmented or AI-centered. Further, a small business-oriented social network app that has the potential to internally use generative AI (specially with an AI-augmented level) is explored and it is concluded it is not still the time for such technology to be implemented there for a number of reasons.

8. Case Studies of AI in Paid Media

The field of AI in digital advertising has demonstrated remarkable successes as well as notable failures in the short history of machine learning deployment in paid media. Recent years have also witnessed the emergence of a number of interesting AI-powered applications that directly tout value propositions for addressing the issue of workforce asymmetries across levels of analytical sophistication. And that claim to provide tools to support campaign creation, execution, or evaluation, regardless of the background and expertise of the practitioner. In this section, we summarize some select campaign-level instances of both successful and unsuccessful applications of AI in the digital advertising context.

Successful Campaigns

AI has been applied successfully to a variety of digital advertising objectives, fueled by the growing abundance of real-time econometric data and rapid advancement in algorithmic power. To address issues related to the complexity and resource intensiveness of manual campaign optimization, ad networks have integrated AI into their platforms at scale for purposes of automating the optimization of bidding and ad delivery across campaigns and ad groups. Similarly motivated by the underlying high-dimensional optimization problems, data analytics companies have developed auction bidding tools that incorporate various advanced techniques. Recent years have also seen the appearance of powered tools that allow advertisers to build more effective campaigns with little-to-no machine learning expertise enabling practitioners to access machine learning anywhere along the marketing technology vertical. Several innovative companies now offer tools for creatives that eliminate the guesswork involved in not only developing the appropriate messaging for each audience segment at the ad level, but also selecting the assets within an ad format that actually increased effect and engagement.

8.1. Successful Campaigns

The search marketing program for Home Depot was that they partnered with an advanced advertising technology company to treat their SEM as a profit center, rather than just the top of their marketing funnel. They were processing sales attribution data from their e-commerce platform every hour and then making real-time bids based on both this data and historical and predictive analytics for marketing attribution and customer lifetime value across both search and display. Their campaign was able to tie PPC spend to incremental online and in-store sales with better and more timely information than other marketing channels contained. The firm’s search marketing program included deep optimization of PPC bidding strategy for conversion attribution, considering the impact of customer ad blockers and click fraud on accurate estimates of conversion data.

The advertising technology company had built a simple wrapper that used its API, allowing clients to display ads on social media and search engines, only when its algorithms forecast customer conversions – it operated on a pay-per-user-conversion model. This technology was used to optimize ad viewers for an article on the company’s IPO, increasing CTR to over 3%, compared to 0.06% that all other ads on that page were achieving, and driving a third of the article’s daily traffic. This showcase allowed clients to utilize optimization push- and pull-buttons, engaging the services only on their blockbuster campaign launches. They were tempted to ask why the ad experience, viewed by a billion-plus search engine users and an ever-increasing number of social media users, had not been guided by smarter, more predictive, and more customer-centric algorithms long before.

8.2. Lessons Learned from Failures

The following section examines several more artificial intelligence-enabled paid media initiatives that had rocky starts or ended on unplanned notes. While these blemishes may overshadow the successes in some eyes, they also provide lessons to inform future attempts at technology implementation. The sad irony is that some of the projects detailed here were announced with great fanfare, yet were quietly retracted after failing to achieve intended results or creating unintended consequences.

Not everything that advertises well is well-lived. A startup scam advertising ephemeral deepfake technology captured the imagination of mainstream advertising and tech media and won an award as an “AI for the Humanitarian Good” startup. A year later, while still largely ineffable as a product set, it had become the foil of the existing deepfake startup ecosystem, announcing new use cases as it poached its more product-age firm “neighbors.” Evidently, achieving a consumer-facing version with product-market fit would take longer than the founders anticipated. Ten years into the fad that is influencer marketing, criticisms about the integrity of influencer accounts and the apparent tenuous relationship between influencer-generated content and precise actions of audiences lurked just below the surface. One firm promising a solution to provide the type of hard and verifiable metrics marketers crave offered something akin to digital forensics at scale. For marketers, verification was a lot of sifting through the data before it meant anything. Then it was acquired and shelved. It didn’t have the sustained interest to survive beyond a “quick and easy” moment.

A new experimental creator-driven publisher team built a narrative early in its story that focused on education, encouraging audiences to see the inherent value of becoming active participants in the process of narrative curation that constitutes advertising for others: the sponsored post influencer’s art. The team curtained off its content development and had decided to focus on product-market fit. That was in the fall of 2022. Apparently, it was more important to keep other creators from becoming red herrings in its strategy to monetize the publications they would be setting up to help grow than establish a long, ongoing narrative around it as a topic of inquiry with wide relevance across the industry.

9. Future Trends in AI and Paid Media

As AI continues to transform the paid media landscape, advertisers must stay ahead of the curve and embrace the emerging technologies that will shape the industry in the years to come. Innovations in generative AI and other areas will drive pace and change, creating more opportunities for creativity, more data and media to work from, and better and faster decisions to be made. Planning will become more dynamic, based less on historical models, allowing for real-time changes to plans across the evaluate and execute steps in the media planning and buying process. This will allow advertisers to allocate budgets more accurately to separate insights and trends, adapting tactics while measuring the direct impact on performance.

Through the means of carefully matched incentives, AI will shift more budgets and content creation to more authentic and meaningful channels that help solve for consideration and preference along the path to purchase. The advantages gleaned from applying the new AI tools to paid media planning will be swift and impactful. Bringing other functions, especially optimization, closer to the planning table will deliver better performance and faster execution, while creating greater strategic value. Data will drive more different and better-informed discussions, resulting in better alignment and synchronicity across the business. And more investment and belief in creative storytelling will translate to media, comms/creative share shift, amplifying the halo effects that paid media is capable of creating, ensuring that all players can share best practice learnings around the potential for more humanistic communications that AI amplifies.

9.1. Emerging Technologies

Cutting-edge technology is expected to have an increasing impact on core innovations in the media space. Key emerging topics in technologies such as Generative AI, 5G, Metaverse, Web 3.0, Cyber Boths, Neural Processing Units, Computer Vision and Augmented Reality/Virtual Reality are contributing to fundamental changes in how content will be distributed and how the consumer will want to interact with it. In addition, topics such as Blockchain are reinforcing consumer trust by helping secure ownership rights. This new composable environment will open major new media revenue streams. While the threats affecting many sectors clearly derive from traditional technology issues such as protection from hacking and malware, cyber-crime, the obsolescence of infrastructures, coupled with new parameters such as the transition to Cloud, which is making enterprises more exposed, and the growth of mobile technology, which is making the access to corporate data easier, at the same time, accelerated technological development has taken the Advertising & Media sector to a different stage. On the one hand, directly productive activities have access to new innovative solutions to optimize their strategies: Big Data, Cloud solutions, Data-Based Creativity, Blockchain. On the other hand, the changes in consumers and their habits are so sudden that the sector players are forced to adapt rapidly to the evolution of the demand. This paper proposes some ideas to navigate a sector that is now more uncertain than it was a few years ago, even if exciting, and to enhance the transition towards a new normality. In order to seize and explore the business opportunities coming from the new environment. This document draws from our engagements and experience to help companies around the world address these challenges.

9.2. Impact of AI on Consumer Behavior

While some worry that increased automation will reduce jobs for humans, others argue that changes in consumer behavior will prompt businesses to respond by also changing their business transparency, principles, and procedures, to keep appealing to consumers. The changing attitudes of consumers, from wanting to maximize personal gains to personal advantages, have been noted. Also, fueled by the COVID-19 pandemic and the growing role of artificial intelligence tools in their lives, Gen Z and Gen Alpha are expected to prioritize purpose-driven purchases and corporate social responsibility (CSR) and corporate charitable initiatives higher than earlier generations, and thus companies should embrace the value-driven model. Previous research also confirmed that younger generations are more alert to initiatives directed towards helping the natural environment or creating a more sustainable business model, for instance, using biodegradable materials. Companies will have to be transparent about the use of AI and ensure that all aspects of their product or service delivery, not just marketing, are coherent and meaningful.

On a related note, the uniqueness of an individual’s data trail, from exploration to exit, that is generated in interactions with a website or platform can now often be harnessed by businesses with the help of AI algorithms and robotics for the purpose of persuading that consumer to take a desired action. Exceeding informed consent, a new wave of behavioral advertising calls for sending nudging-style advertising messages to the mobile devices of users while they are physically en route to stores that sell those products. Borders are further tested by a new business model termed social influence marketing wherein influencers endorse products and services while displaying a certain lifestyle, and advances in data mining techniques enable the identification of product influencers in a network and prediction of product sales in that neighborhood. Accordingly, companies can identify those individuals whom ordinary consumers trust and are more influenced by, and reward these individuals for encouraging their peers to act in a given way.

10. Measuring Success and ROI

In every business conversation there is some discussion related to ROI, which is only natural given that business is all about making money and any initiatives you engage in must have some justification in terms of future revenue contribution and profitability. Digital paid media is often lauded for its direct ROI so it is also accepted that you should have the ability to measure what results you get from the efforts you engage in. Given the technology behind digital media, measuring success has become an easier process and of course, the key driver for this is ubiquitous tracking that is enabled by tracking technology at a browser level and tracking pixels at page level. The two enable you to measure how many of your messages users see and how many times they respond, either by visiting your website or completing a desired action, thus enabling you to compute conversions and assistance conversions along different cohorts by time to conversion. This information on action activity can then be used to measure actual revenue earned on the site as well as other business metrics such as customer acquisition costs, customer lifetime value and gross margin and from this easily determine ROI.

Digitally enabled user tracking does come with a caveat though – you’re mostly using tracking technology to track users, but that assumes users are accepting tracking and that companies are not blocking assortment of tracking at pixel level. In fact, browser manufacturers are now keen on blocking such tracking because they claim the ultimate reason for this is improvement of user experience. While this will have an impact on the capabilities you have for measuring success and often lead to inaccurate under-reporting of conversions and rollback of attribution windows, the fundamental question that you can track still and the driving factor in your communication strategy is whether or not your program succeeds in enhancing users’ willingness to pay for your product or service offerings.

10.1. Key Performance Indicators

Paid media is usually marketing a specified product or service with a particular goal in order to increase the sales of that product or service within a specific time period. The KPIs used to measure success for paid media campaigns are typically cost-based, conversion-based, speed-based, and value-based. Cost-based KPIs analyze the costs associated with paid media campaigns. The metric specifies the cost of each successful conversion based on the total ad spend and the number of conversions achieved, while provides information on how much clients spend for every click on their ads. Conversion-based KPIs use conversions as their basis for measuring success. The conversion volume specifies the number of conversions generated by an ad campaign while expresses the percentage of visitors taking the desired action. Speed-based KPIs point out how fast a certain goal is accomplished. The metric measures the average time users take from clicking the ad until they convert while the measures the time they take from clicking the ad until they complete their purchase. Finally, value-based KPIs provide insights into the overall value generated by a paid media campaign. The most important value-based indicator expresses the ratio of revenue generated by the ad campaign in relation to the costs of the ad campaign. Another important value-based KPI specifies the predicted net profit associated with the entire future relationship with a customer.

While the above can be seen as distinct KPIs through which the advertiser can obtain an overview of campaign effectiveness, they can also be contrasted and analyzed against each other in order to identify hidden opportunities or to determine the best KPIs to align for optimizing paid media campaigns.

10.2. Attribution Models

The value of touch points and interactions in marketing become clear when combined with a budget. It becomes possible to see how much each click costs, making it easy to project upsell and cross-sell revenue figures. Measurement, however, gets complicated beyond the first interaction before a conversion. There are a few paid media channels that make attribution straightforward. Digital paid ads are found in calendars and on specific pages of certain magazines and newspapers and in a limited fashion on TV. But most digital tracking is done to track a person who has clicked on a paid ad for a day or up to 90 days after the click and who makes the conversion on a landing page with the same pixels as the impression pixel of the ad network during that time window. It becomes notable to evaluate paid media across the major ad networks due to current and emerging tracking solutions.

It is possible to give credit for a conversion to all channels though many times a marketer or agency is only interested in evaluating the performance of a single channel. Modeling, whether a last-click model or first-click attribution model, addresses this. For large businesses, there are pretty sophisticated algorithms as are available for smaller size businesses and companies. The last-click will always see high event numbers and conversion values since that is what is used to assess campaign budgets and objectives. To appeal to more verticals and use a multi-touchpoint attribution model, common vertical models like fractional are often applied. After all, campaign management is often a hot potato that changes monthly. If there is a disagreement about spending priorities, using a multi-touchpoint model at least creates a starting point that recognizes importance across many parts of the customer journey.

11. Challenges and Limitations of AI in Paid Media

There is no debate that artificial intelligence (AI) is a powerful set of technologies. Today, AI is changing and improving many aspects of how paid media is designed and executed. However, it is important to be mindful of the problems, challenges, and risks of working with AI in such a high-stakes area. Many industry veterans continue to express skepticism of AI or have reservations about its use, importance, or advantages. These doubts are valid, given the many challenges that AI currently faces as an area of exploration and implementation. These challenges affect the capabilities of AI-based systems in both small- and large-scale applications. In today’s rapidly evolving paid media environment, AI must be undertaken deliberately and with great care.

The technical limitations of AI represent the most obvious area for concern, as they impact the risk and potential for financial losses. The economic problems with AI are equally significant, having to do with the broader questions of how quickly and thoroughly the market will adapt to it. Will the apparent advantages and efficiencies AI can bring through automation, prediction, optimization, and other areas be realized in practice? Or simply be more hype than substance? In the following paragraphs, we address each of these areas in turn, discussing the limitations and challenges they present.

11.1. Technical Limitations

AI algorithms have several technical limitations. The complete elimination of manual interventions is too risky, in particular, when it comes to high investment campaigns. Thus, if paid media executives should be able to intervene manually in the campaigns to optimize the results, why not rely entirely on their expertise and knowledge of their audience and products? This brings us to the second consideration regarding the deployment of AI: advanced algorithms need a huge volume of data to learn from; as data leakage and overfitting are ever-present threats, AI requires companies’ databases to be considerably larger than traditional machine learning techniques. Consequently, AI algorithms are sensitive to data quality, and can even lead to disastrous consequences without a strict data supervision. Such a requirement can practically come to an obstacle as the database often cannot exploit minute features for extreme customer segmentation, as there might be only a few observations for each possible combination of the relevant features.

A third reason has to do with the difficulties that AI systems have regarding causation: what happens if we changed some of our advertising levers because of AI recommendations? Will the effect of changing some of them have different effects compared with previous ones? Or if we are in a seasonality induced scenario, do we have confidence that AI will recognize the seasonality? Last but not least, even if AI may be able to discover some very complex relationships among the input variables, the complexity of the resulting algorithms might affect advertisers’ performances, either because they have no insight into system operations or because they are even more prone to overfit than simpler algorithms would.

11.2. Market Adaptation

In terms of paid media, market adaptation means ensuring everybody involved in the media planning, buying, and optimization processes is up to speed on the new capabilities. Publishers must adapt their media offerings and technical environments to enable marketers to take full advantage of the new capabilities and maximize the benefit obtained from advertising investments. Marketers must make budgeting, tracking, and optimization processes more fluid and adaptive. Agencies must integrate the potentials into more strategic planning processes, and tools must impart that intelligence. The AI capabilities on offer must also be relatively easy for most ecosystem players to learn about, use, and optimize for.

Whether one is from the publisher side, agency, or client, if one closely identified a pet AI capability and invested in its exploration, it would be wise to consider its accessibility to the rest of the extended team — the tracking and measurement processes, the collaboration processes, the technical environment supporting the investment — as fundamental elements of the experimentation stage of initiating that one AI capability. No human-in-the-loop system is optimal if only certain human ones are specifically trained for it. Automating a stage of a media process can be powerful, yet fall into a strange pit — it can produce weird results, positive or negative — that destined the need to intervene and adapt the rest of the media process to be particularly fragile.

Automation, be it fully or in part, does not end the need for constant oversight of the media buying and optimization processes. It creates an additional dynamic aspect of these processes — AI capabilities should be constantly trained on the output that was the reason for a specific process to be automated. If teams cannot remove the pitfall aspect of the new capabilities, then they should be relegated to the experimental phase until that solution or others become more widely adopted and thus more accessible.

12. Best Practices for Implementing AI in Paid Media

Paid media strategies today often incorporate Client Relationship Management (CRM) functionalities, acting as a two-way communication medium linking advertisers and consumers. By employing a continuous feedback loop, companies can reinforce brand messaging, nurture prospects, and deepen relationships. Messages can be personalized, retargeted, or optimized using customer data, predictive modeling heuristics, or CRM systems, while actions can include controlled push, paid media advertising, organic push, exposure, and domain. Combining paid advertising with other channels reinforces the paid media’s role of making brands visible. The objective of this chapter is to offer practical insights into how the paid media functionality can be enhanced through the use of Generative AI or Machine Learning approaches. The proposed tools and methods include dynamic creative optimization, predictive modeling, and optimization. From a strategic perspective, all proposed actions are anticipatory, thus requiring the marketer to think and plan for the long-term consequences of the anticipated user’s reactions. Four different AI services are proposed: development of guardrails, creative messages optimization, predict user action, and optimize media. Implementation of AI in media enables an anticipatory approach, yet it could also lead to brand safety issues. Marketers must therefore make sure to have guardrails, recommendations on how to write the copy and on the creative elements.

12.1. Strategic Planning

With various applications of AI already in common use for paid media, from serving ads to optimizing budgets and targeting, now is the time to focus on making the most of what AI tools are capable of doing while being careful to use appropriate guardrails to avoid their significant pitfalls and limitations. Clients regularly ask about AI strategy for their paid media, and we have compiled a few best practices for planning, implementing, and refining how algorithmic intelligence plays a role in their digital activity. Here are a few guiding principles, which can help lead to successful paid media growth:

AI is a Tool, Not a Replacement

Specific use cases of AI writing and image generation tools are quickly creating fear that advertising copy and design will no longer be job security. But just as there is no copyright on the Mona Lisa, likewise computers will never create a piece of art as timeless and subjective as a beautiful billboard. This sentiment should be shared as well with what are the much-needed systems behind the AI-powered ad copy, which require years of marketing, branding, customer journey, eCommerce, and performance to develop. There is no shortcut there. AI tools implemented in appropriate ways can save the people in charge of monitoring ad performance time at particularly high volume and low value times, but the role of active evaluation and application of learned insight from thoughtful and ad performance altering input to AI processes should stay firmly in human hands.

Build Instead of Buy

It can be tempting to see all the vendors rolling out their own AI ad copy generation tools that say they can help you create the “best” copy and optimize it for performance for you. These solutions can seem especially appealing with the prompt fatigue of having to come up with countless different crowds to optimize product inventories across language and region. But the same model is being applied to various products and campaigns for a huge volume of advertisers. The custom audience targeting that is key to successful paid media campaigns comes from specific brand tasks that leverage a unique set of consumer insights. Having a display ad system powered by an engine that has a proprietary monopoly on batched AI optimized copy and hooks for each target audience will yield less performance compared to a custom-built solution that acts as a part of a much larger owned customer network.

12.2. Continuous Learning and Adaptation

In contrast to traditional constraint and performance optimization processes in advertising, the application of AI hinges on exploiting large amounts of user data to maintain an observation and recognition cycle of the customer journey. For advertisers, this means that implementing AI in advertising requires committing to an ongoing process of continuous testing and adaptation. We identify three best practices in this respect. First, advertisers are advised to work closely together with the AI vendors to jointly learn from the data at hand. Second, advertisers should maintain a close monitoring of customer behavior and algorithmic recommendations to react in real-time to changes of established behavior patterns. This requires a shift of creative input, content production, and campaign management budgets and schedules towards events at the borders between the frontword and backword stages of the advertising funnel. This impacts data availability, data quality, and timeliness of data analysis, triggering a risk of optimization on short-term goals of minor importance. Stockpiling headroom budgets for real-time events may help here. Third, advertisers should follow a Markov Chain approach to attribution modeling, balancing a data-driven detailed ongoing analysis of returns to advertising mix investments in different media channels and time frames with top-down budget allocations across different channels.

AI is not only suited for putting “creative creative” functions with the focus on exploring new ideas for creativity in advertising back into the hands of people. AI is also suited to recognize what works well in the short-run and to let humans explore long-term efficient outcomes in an ongoing cycled approach that can be further optimized over time by the algorithms using reinforcement learning.

13. Conclusion

The potential for AI innovation to reimagine the processes associated with paid media is immense. With its ability to analyze vast quantities of data, autonomously perform tasks requiring deep expertise in a timely manner, and create new digital touchpoints at scale, AI will address many of the efficiency challenges that have plagued digital media. This will ultimately lead to significantly improved media performance for brands and greater perception of relevance and quality for consumers. This is truly a win-win proposition.

There are myriad applications of generative AI to paid media strategy and execution. Some are foundational, providing a new layer of functionality that will be used by brands and agencies to optimize processes, while others have the potential to disrupt the industry. These innovations are much more than the basics of Smart Bidding and Performance Max campaigns. They will create new frontiers in the brand-consumer connection, enabling advertisers to rethink ad formats, creativity, timing, targeting, media mix, and beyond. However, these applications are just the beginning of the journey. The road ahead towards more frictionless and efficient digital media will require thoughtful progression of experimentation and investment in AI-driven methods and models that harness the true potential of this paradigm shift.

In summary, we present a solid framework for brand leaders and agency partners to understand the rising wave of AI innovation and to think about the best way to explore its potential for their paid media efforts by addressing some core questions. These fundamental questions want to understand the capabilities and use cases for AI generative solutions, how to experiment and embrace these innovations in the short term, how to evolve the model for success in the mid-term, and the potential of regulation and policy issues on agency partnership dynamics in the long term.