In a poll conducted by Integral Ad Science (IAS) 69.0% of agency executives said that ad fraud was the biggest hindrance to ad budget growth, compared with more than half (52.6%) of brand professionals who said the same.
How much is ad fraud costing advertisers? Nobody knows, but with estimates ranging from $6.5 billion to $19 billion, there’s a lot at stake. Marketers are becoming more assertive in their demands for better fraud prevention measures and they are seeking to increase their knowledge of different fraud types – from bots to unauthorised domain reselling – and wider technology adoptions to drive their Marketing strategies overall. Ad tech providers will need to adapt their technology and techniques to meet this demand.
Over the last few years, artificial intelligence (AI) and machine learning (ML) has been making major headlines because they are transforming every facet of our personal and professional lives. Machines that understand text and speech, and can sense the environment around us, virtual agents and robots that change the way we communicate and work and autonomous vehicles are becoming the reality around us. Digital advertising is no exception, as we use AI and machine learning to help in the fight against ad fraud.
But before we explore how AI and ML are used to fight ad fraud, it’s helpful to distinguish between the two, as the terms are sometimes used interchangeably even though they are not the same.
Artificial Intelligence: Artificial intelligence (AI) has existed in one form or another for decades, but recent algorithmic advances, the exponential increase of processing power and storage, and incredible growth in the amount of data generated by online activity has brought this technology to the mainstream. AI refers to a range of big, potentially world-changing technologies, like natural language processing, virtual agents, robotics, autonomous vehicles, computer vision, and machine learning.
Machine learning: Machine learning is a more specific technology that uses artificially intelligent computer systems to autonomously learn, predict, act and explain without being explicitly programmed. Simply put, machine learning eliminates the use of preprogrammed rule sets – no matter how complex. Machine learning techniques are extremely effective in combing through the millions of impressions generated by digital advertising campaigns and detecting patterns in these streaming data sets that are indicative of fraudulent activity, such as non-human traffic or other bot-driven automated activity.
Data is key when it comes to building machine learning systems. More data equals better models - and that holds true when it comes to fraud detection. Practitioners need their machine learning platform to scale as data and complexity increase. Although academic tools often work well with thousands of records and a few megabytes of data, real-world problems are measured in gigabytes or even terabytes of data.
In an August 2018 survey of 250 senior-level marketing decision makers worldwide conducted by Forrester Consulting and Accenture Interactive, 26% of respondents said that when it comes to innovation over the next 12 months, one of the key elements of the CMO role will be driving a new technology strategy. This was tied for the most popular response in the survey, indicating how technology is commanding attention from C-suite marketers.
How AI and Machine Learning help fight ad fraud
Machine learning powers two critical pieces of our fraud detection technology.
Behavioral and network analysis: Use of big data to distinguish real user behavior from bot behavior by looking at anomalies within site visitation patterns. Cohorts of bots tend to visit the same cluster of domains over and over because their behavior is automated. Detecting these patterns can allow us to surface bots based on their behavior. Most humans don’t visit the same sites, in the exact same order, multiple times per day. If these cohorts have only visited specific domains that indicate a pocket of bot activity, we track these patterns and mark this traffic as fraudulent.
Machine learning techniques also help also identify patterns in traffic that aren’t immediately obvious to human analysts.
Large consumer brands like Uber and Amazon are leveraging big data and machine learning to power major technological innovations, from driverless cars to drone delivery services. We believe that marketers deserve access to the same predictive technologies to protect their digital investment.
Browser and device analysis: Machine learning allows advertisers to identify invalid traffic sources by matching browser features to the user agent. This type of determination is often mistakenly labelled deterministic, but it would be impossible without employing machine learning methodologies to detect patterns within large data sets. Applied correctly, and powered by sufficient data, this method of detection can help to weed out entire bot networks.
As technology continues to evolve, fraudsters are not only becoming more skilful and tech-savvy, but they are increasing in numbers. Organisations that want to defend themselves against fraud need to have a superior, faster-learning solution that can constantly evolve yet is easy to use and maintain.
AI for brand safety: Artificial intelligence should be used to automate the brand safety solutions. Through machine learning which can be informed by extensive data, the solutions are able to continually improve their understanding of the digital landscape at scale, and can automatically assess and determine the appropriateness of new pages without requiring explicit programming to do so. Companies must use AI to scale brand safety offerings and not forget to also apply a human lens to continuously audit and enhance their models to keep up with the constantly evolving landscape of risky content.
What’s next for AI and ML in their uphill battle?
AI-driven automation - AI will lead to automate routine processes, which will help optimise high volume, rules-based work and enhance scale for solutions provided to the advertisers. This technology will help integrate cognitive capabilities so that business flows to better analyse a given situation, learn from it, and reason through it without intervention from a human analyst.
Next-generation machine learning – Companies must constantly work to expand their machine learning capabilities, especially through the adoption of deep learning and reinforcement-learning techniques based on neural networks. These two technologies depend less on existing domain knowledge and ensure optimised results through self-correction of the fraud prevention and detection models.
Ultimately, artificial intelligence will continue to drive change throughout industry and society. For digital advertising, much of that transformation will be powered by AI-driven automation and machine learning that can help us grapple with the massive data sets produced by large, fully scaled, digital campaigns. Like our peers across the industry, we see tremendous potential in AI and machine learning to continue transforming the way we do business. While it’s too soon to say we’ve identified all opportunities, it’s clear that these technologies could be applied to many facets of media quality. However, it’s important not to think of these technologies as far-flung possibilities. As the examples above demonstrate, AI and machine learning already need to be a part of our business today.
Laura Quigley is managing director of South East Asia at Integral Ad Science.