Machine learning is among the most exciting developments in modern advertising. It changes how businesses collect and analyze data and what they can do with that information.
The technology also raises questions: what is it, why use it and how does it work? Let's take a closer look.
What Is Machine Learning in Advertising?
It is the process through which advertising technology takes in data, analyzes it and uses the results to improve a task. In simpler terms, this is how the technology learns to become a better advertising specialist.
What it learns depends on the system. The task could involve media buying, mapping customer journeys, identifying audiences or another aspect of advertising.
The more relevant data it processes, the more it can learn about that task and the better it can become at completing it, much as a person learns through experience.
Do not confuse machine learning with artificial intelligence.
The Difference Between Machine Learning and AI
The two terms often appear together in discussions of advanced technology. They are related but not interchangeable.
Machine learning describes a particular process: machines use data to learn and improve their functions. AI is broader, covering technology that performs tasks traditionally requiring human intelligence. Machine learning is therefore one aspect of AI, not a synonym for it.
Why Machine Learning?
From digital transactions and retail stock to server-room temperatures, modern businesses can track almost everything.
Reports suggest that more than half of available data goes unused. This is often called dark data. The main obstacles include:
A lack of tools, incomplete records, excessive amounts of unusable information and disconnected systems. Behind these barriers is a simple but substantial problem: people cannot do everything. There is too much to identify, collect and process.
The particularly striking gap concerns collecting and analyzing data: the lack of suitable tools is cited at 85%.
Advertisers already hold terabytes and petabytes of demographic information, including personal details, locations and interests they can use to reach potential customers. For 95% of advertisers, much of this information remains poorly understood. AI offers a way to bring it under control and make greater use of it.Orchid Richardson
Using AI to understand data and take it further may sound futuristic, but it has been happening for years.
Examples of Machine Learning in Advertising
Its role is not always visible because complex processing takes place in the background. Some of your favorite tools already use machine learning to deliver their insights.
Here is what advertisers can do with it:
1- Discover Unexpected Opportunities
Good advertisers use targeted data. The way they obtain that data, however, may be imperfect.
You may want to analyze every piece of information about an offer, but a limited budget forces you to prioritize what appears most important. That means collecting less information overall.
Less obviously, assumptions about the product and audience can restrict advertising decisions. If you sell a video game, you might target young gamers and their middle-aged parents without considering grandparents. Those assumptions can cost you opportunities.
At the 2018 AI Conference, Unity's vice president of advertiser solutions, Julie Shumaker, described just this kind of problem:
An advertiser may have a precise goal, such as selling a $17 game installation to a 22-year-old player, and may never consider a 65-year-old woman. Machine learning can reveal that she might spend around $3.99 over three days. If acquisition costs 75 cents, the return can be as attractive as higher-value targets among conventional audiences.
By processing large datasets cost-effectively, machine learning can uncover revenue opportunities that budgets and human assumptions would otherwise obscure.
2- Improve Your Advertising Creative
Audiences respond differently to media, fonts and calls to action. These creative choices affect whether people click or tune out.
Machine learning is often associated only with numerical data, but predictive analytics can help improve creative work as well.
A system can analyze previous advertisements and campaigns to identify what is likely to work next. Brands can better understand how messages, fonts, colors, images, button sizes and formats affect overall performance.Rajiv Bhat
This is not hypothetical. Mobile app marketing automation platform Bidalgo offers a service called Creative AI. It analyzes visual media to identify creative approaches likely to succeed.
The system can assess which advertisements have a positive effect, even considering details such as how people pose in images. It then produces an analysis-based creative brief for content teams.
A related study described in the Journal of Consumer Psychology matched images to personality types. Researchers used algorithms to identify 89 visual characteristics, including hue, saturation, color variety, detail and the number of people shown.
The study asked 745 participants to rate images on a scale of 1–7. They also completed a personality assessment across openness, conscientiousness, extraversion, agreeableness and neuroticism. Researchers then examined which images appealed to which traits. They found, among other things:
- Extraverted people preferred simpler images and images featuring people.
- Open-minded people preferred pictures without people and with cool colors such as blue and black.
- People with higher neuroticism preferred calm scenes with minimal stimulation.
A follow-up study found that participants preferred advertising images aligned with their personalities. More importantly, the algorithm identified a relationship between personality, image type and product interest. People expressed more positive attitudes toward the matching brands and greater purchase intent, rather than merely liking the pictures.
3- Increase Contextual Relevance
In theory, excellent creative should be enough to generate an excellent response. In practice, it is never that simple.
The advertisement must also appear on the right platform, at the right time and with appropriate targeting. Machine learning can improve this process.
Instead of relying exclusively on audience data, systems increasingly process page content in sophisticated ways.
Developments such as sentiment analysis blur the boundary between human and machine interpretation. Systems can identify opinions in text and classify an author's attitude toward a topic or product as positive, negative or neutral. This interpretation continues to develop as language introduces nuances such as sarcasm and idioms. Advances in deep learning also allow computers to process images and video in more human-like ways.Harmon Lyons, IAS Senior Vice President of Global Business Development
At a basic level, understanding a page helps advertisers and publishers deliver more relevant ads. More sophisticated applications open up further possibilities.
The Vodafone Example
Vodafone in the UK wanted to include the iPhone X in its advertising, but Apple's restrictive brand-use rules made it difficult to mention the product.
It turned to GumGum. Machine learning analyzed page images to find iPhone advertising and place Vodafone advertisements over those images.
The Jeep Example
Jeep, another GumGum customer, chose contextual targeting instead of behavioral modeling. The technology searched pages for images of Cherokee competitors such as the Toyota RAV4, then placed Jeep advertisements over those competing models.
Machine learning can also protect brand reputation. As programmatic advertising increasingly appears in unsafe contexts, such as alongside extremist content, it can help prevent a public-relations problem before it happens.
4- Reach a More Relevant Audience
Advertisers seek maximum relevance. Narrower audience segments bring them closer to the one-to-one personalization customers want.
This requires large amounts of data and a sufficiently advanced algorithm to turn it into useful decisions.
Major networks such as Facebook, Google and LinkedIn provide these capabilities. Their datasets include interests, hobbies, locations, job titles and other information that can refine targeting.
Businesses cannot easily gather all of this themselves. When their customer information is enriched by networks such as Facebook and Google, it becomes useful for identifying the people most likely to want an offer.
Facebook collects data and supports audience creation. It also uses machine learning to decide whom to bid for and who is most likely to complete an objective.
The strongest advertisements today offer personalized content that genuinely means something to the audience.
Each audience should receive creative informed by its data. The post-click page needs to continue that personalized experience.
An example comes from Abreva, which created 119 advertisements tailored to the content viewers were watching.
Starting with four core videos, Abreva used Director Mix to customize text for different interests, creating 119 versions. Viewers received relevant ads based on what they watched.
Someone watching celebrity gossip saw this sequence:
Message 1: You cannot get over celebrity breakups.
Message 2: But you can get over a cold in 2.5 days.
People watching beauty tutorials saw:
Message 1: Cold sores on your lips bring you down.
Message 2: Get rid of cold sores in 2.5 days.
This approach increased awareness and consideration. Abreva reported a 41% rise in overall ad recall and a 342% increase in search interest among its audience on Google and YouTube.
Read the Think with Google case study here.
Responsive search ads let you provide multiple headlines, copy variations and descriptions for Google to test and combine. Advertisers using the feature saw an average 15% increase in clicks.
5- Bid More Strategically
Not every programmatic impression is worth bidding for. Some opportunities deserve more investment than others.
Demand-side platforms remove some of the guesswork. Machine learning can make bids and optimizations that once required experienced media buyers.
Google Smart Bidding, for example, uses machine learning to optimize conversions or conversion value in each auction. This is known as auction-time bidding.
Google Describes Five Types of Smart Bidding Strategy:
- Target CPA: A Google Ads strategy that adjusts bids to obtain as many conversions as possible at or below a target cost per action. Advanced machine learning adapts bids to each auction.
- Target ROAS: Bidding based on a target return on advertising spend. It aims to generate greater conversion value or revenue within that target.
- Maximize Conversions: Automatically sets bids to get as many conversions as possible while spending the campaign budget. Machine learning optimizes bids for each auction.
- Enhanced CPC: Adjusts manual cost-per-click bids for clicks considered more likely to produce a sale or conversion, helping obtain more conversions from manual bidding.
The choice depends on objectives, budget and other factors. Smart Bidding continually analyzes information to identify the bids and impressions most effective for the chosen goal. These algorithms can consider performance variables more comprehensively than one person or team.
Auction-time bidding uses a wide range of signals: identifiable characteristics of the person or context during the auction. They include device and location, also available through manual bid adjustments, plus signals and combinations unique to Smart Bidding. More details are available here.
Google's Guidance
Smart Bidding can help businesses of different sizes. It can optimize using data from across campaigns, potentially benefiting new campaigns without their own history. To assess results properly, measure longer periods, such as a month or more, with at least 30 conversions.
Machine learning can have a substantial positive effect on campaigns. Bidding, creative and, above all, personalization can improve considerably when you find a model that suits your needs.
Comments