Balancing Privacy and Personalization With ML-Powered Advertising Solutions

Advertising spending has seen unprecedented growth since the COVID-19 lockdown. Pre-pandemic e-commerce digital ad spending in the US amounted to $12.5 billion. In comparison, it is estimated to reach a staggering $38 billion in 2024.

It might be hard to believe, but an average American is exposed to as many as 4000 ads daily. 

Most of the digital ads today are targeted and, in certain cases, even personalized.

It is no surprise that, according to a recent KPMG study, 86% of people feel growing concerns about their data privacy, and 78% worry about the amount of personal data being collected online.

These concerns don’t go unnoticed. Laws and regulations surrounding personal data are getting stricter while, at the same time, the reign of third-party cookies is quickly ending, with Google following in the steps of Apple and promising to phase them out this year.

As advertising is hard-pressed on every side, serving personalized ads needs a new solution. Fortunately, it is already available and comes in the form of web scraping combined with artificial intelligence (AI).

For years, advertisers have relied on behavioral targeting, collecting consumer data through cookies or other means and deriving insights from it to predict buying behavior and consumer needs.

Notorious for requiring massive amounts of data and capabilities to analyze it, behavioral targeting is expensive and often brings a questionable ROI.

A poll by Digiday showed that 45% of publishing companies’ execs saw no significant benefit from behavioral ads, and 23% said they actually caused a revenue decline.

Further complicated by GDPR and the demise of third-party cookies, behavioral targeting today keeps advertisers on their toes and exaggerates data management costs.

Many advertisers started to shift their focus to first- and zero-party data. Some, however, decided to walk down a less-traveled but more rewarding path and invest in contextual targeting.

Quite an old approach, contextual advertising serves creatives based not on the user’s past and present behavioral trends, but on the content of the media the user is browsing through.

Google AdSense is probably the best-known example of contextual targeting – it analyzes page content and places commercials in front of audiences who are more likely to get interested in them.

Analyzing context tells what users are interested in in real-time. Moreover, it can present more relevant insights into their most-pressing pains and needs than behavioral analysis.

Imagine a relatable situation: you browse through several e-commerce shops looking for sneakers and finally buy them.

The next day, you’re browsing, let’s say, vegan recipes, but you get sneaker ads even though you’re not interested in them anymore.

They follow you for a while, and it’s annoying. This is behavioral advertising –
sometimes, it works, but many times, it makes you desperately click on a
‘block’ button.

Now, imagine that you read an article about a severe cyclone approaching and, in the right-hand corner, there’s an ad for wind- and water-proof functional outfits. It is an example of good contextual targeting.

First, it creates a personalized experience without trespassing on the user’s feeling of privacy.

Second, it opens new opportunities for businesses that can’t form long-term B2C relationships and gather proper behavioral data, such as single-product companies or infrequent service providers.

Analyzing context allows advertisers to consider time-sensitive data – location, weather, and even the general emotion of the content that the customer is viewing.

Such information opens new ways to influence consumer decisions.

Moreover, it lets companies experiment with real-time dynamic pricing and respond to quickly changing market conditions.

However, to serve up-to-date and personalized creatives in today’s multimedia environment, advertisers need to analyze all types of information – texts, visuals, and sounds.

Furthermore, they have to target ‘sentiments’ like anger, joy, positivity, or anxiety. Such complex data analysis requires costly human resources.

Fortunately, due to recent technological advancements in web scraping, artificial intelligence (AI), and its subset – machine learning (ML), contextual targeting has become easier and more cost-effective.

With the help of the newest AI technologies, advertisers can employ deep-learning algorithms to ‘understand’ and analyze text, images, or sounds.

Also known as semantic contextual targeting, it mimics human behavior to interpret the given context and serve relevant ads.

Powered by Natural Language Processing (NLP), AI can process human language patterns and perform deeper, more nuanced page analysis in real time.

On the most basic level, the algorithm that serves ads is fed with specific keywords it must find and ‘negative’ keywords it must avoid.

Negative keywords are necessary to avoid awkward or unethical situations – for example, advertising artisan chef knives in an article about street stabbings.

The algorithm should also separate the homonyms, for example, Turkey (country) and turkey (bird), rose (flower) and rose (color), etc. To do this, it must analyze the surrounding context of the word.

Using automatic language processing, the ML algorithm crawls the content of the page, taking into account the keywords, the sequence of words, their surrounding words, and their probable meaning.

After this, the algorithm sorts the crawled pages into categories.

Proper ML-powered software can categorize content at scale and combine it with media performance optimization and brand safety, putting the creatives straight into the most favorable context.

For example, AI can analyze a catalog of over 150 million videos, covering more than 200 IAB categories and 500,000 topics and serving contextually relevant ads, in less than 100 milliseconds.

The same technology not only decides how to serve creatives in the most efficient way but also optimizes the automatic bidding process.

It is estimated that investments in contextual targeting will reach $376.2
billion by 2027. Studies also show that up to 52% of UK and US advertisers plan to increase their contextual targeting budgets, and 86% of media owners expect this approach to become prevalent.

However, contextual advertising is not without its own challenges.

Probably the biggest challenge for companies seeking to employ contextual targeting is writing and maintaining custom scrapers for each website.

Technically, it is quite difficult to ensure that scrapers adapt to constant changes in the sites’ structure.

To keep the data flowing without interruptions and errors, the most efficient way is to use the services of reputable web intelligence providers.

Yet, another significant contextualization challenge lies in developing ML algorithms for content categorization at scale.

After crawling millions of sites in real time and applying granular classification, NLP algorithms must be able to categorize separate audiences that are interested in specific topics and serve them the most relevant ads.

Such technological sophistication is still costly and hardly accessible for small and medium-size advertisers and publishers. Usually, those having smaller advertising budgets employ less-advanced keywords-based contextualization solutions. 

Offering a personalized advertising experience while respecting privacy sounds like an oxymoron. However, there’s no indicator that privacy-related concerns will become less important in the near future.

Therefore, AI-powered contextual advertising will probably prevail and develop as one of the best alternatives for marketers trying to make their promotions more engaging and less offensive.

Currently, technological accessibility is the most pressing challenge for contextual targeting, keeping many businesses wary of this approach.

However, with contextual marketing becoming more prevalent, the proliferation of contextual analysis tools will intensify, driving costs down.

It is also possible that we will see an emergence of an open-source content mapping solution that will allow the industry to develop a more unified approach and make contextual advertising more accessible to SMEs.