Machine learning is a form of artificial intelligence and can help advertisers to achieve more effective reach. But Moloco’s Yu-Sien Low tells our Asia Editor Rica Facundo why some marketers hesitate to fully embrace AI despite the clear benefits.
WARC: In simple terms, how do you explain machine learning to marketers? What are the key benefits and use cases for marketing?
Yu-Sien Low: Machine learning (ML) is a form of artificial intelligence. It is designed to mimic human tasks and is inspired by how biological neurons interact in human brains. When it comes to digital advertising, ML helps advertisers to be far more effective in reaching the right target audience, at the right time, with the right advertisement.
Moreover, ML helps advertisers with insights on users who engage with their ads and carry out further actions, like app installations or purchases. Such insights and access to these valuable user segments helps marketers make decisions about how much to pay for an ad, what ads to show to users and even predict the actions that users will take, ultimately helping advertisers get better at targeting the right people with the right ads. All these things are done in real time with very low latency.
While there are clear benefits, what areas do you think make marketers hesitant about fully embracing AI?
A key reason why advertisers are hesitant to embrace ML models is the lack of understanding about data and the recent industry changes towards privacy. With third-party data going away and the increasing concerns around data privacy, first-party data has a larger role to play in performance advertising, which in turn will impact how advertisers use ML models.
And while ML has become quite a buzzword (much like “generative AI”), it’s important to understand that not all ML is created equal – it is not a silver bullet that alone enables amazing product experiences. Companies have to build ML models, often from the ground up, for specific use cases, and then integrate these models into their products to create customer value. To harness these advantages, companies must assess their marketing channels meticulously to ensure the right machine learning fit.
What are some key actions that marketers can take to address these concerns?
It’s important to check and ask your partners the right questions, such as what kind of practices these companies or media channels are putting in place to protect user privacy.
For example, when first-party data is being passed to media channels, it is non-personally identifiable information (i.e., it’s not the name, gender, demographic, etc.)
Partners can ensure that they are encrypting the data when it is in the midst of being sent over, so that by the time it reaches the other side, the data is already encrypted.
Another thing that marketers can do for peace of mind is to check what kind of steps their partners are taking for compliance, whether that’s GDPR, CCPA, PDPA, etc. Or check whether they’ve been accredited by any renowned organisations such as the IAB. When it comes to retail, to preserve user privacy, marketplaces need to build data management systems that follow privacy-by-design industry best practices like NIST Privacy Framework or the cloud vendors’ architectural guidance. For instance, a fundamental way to minimise data exposure is by sharing only necessary data with people or machines on the principle of least privilege.
Marketers should also be asking and ensuring that their partners are not sharing the information they are giving to other advertisers. In the case of Molocco, we process the data for our clients but we don’t own the data.
How does the rapid development of AI and machine learning exacerbate the issue of trust needed for more transparent and effective data party practices?
Trust becomes more of an issue when you don’t have the right understanding of the technology. So with the rapid growth of machine learning, the technology that is being used by the media partners could seem very alien to the marketers.
The problem arises when marketers don’t know how to ask the right questions to clarify. This is an instance where knowledge is power. You really need to make the effort to educate yourself, to find out more, ask questions, ask for accreditation, ask for proof, etc., to have peace of mind.
There are many different types of machine learning models. How do marketers know what type of solution is right for them?
Agreed. Every adtech platform boasts its use of machine learning. But are some machine learning models better than others? What makes machine learning good? Or bad? And where does machine learning fit in the changing regulatory environment? Marketers should also be asking these types of questions and more from their partners.
Some people say machine learning and AI have been around for a long time, but it’s still considered a black box. That’s why it’s important for partners to help marketers better understand the different models of machine learning. For example, operational ML is one of the most sophisticated and arguably the most difficult ones to build. It requires a very highly skilled engineer and technology to build. As alien as the technology may sound, asking providers the type of model they are using is very helpful. This is because not all companies can build such models. It is not just the complexity that needs to be considered but the speed, scale and infrastructure needed to process the data within milliseconds, all extremely challenging from an engineering perspective.
Another factor to consider is capacity. Since we are dealing with large amounts of data, advertisers need to consider a model which can process the large amount of information and be updated in real time. Linear machine learning models are limited, as opposed to operational ML, which can be updated on an hourly basis and will probably just take two to three weeks to learn and deliver results without incurring any costs. So it is a zero-cost approach to learning.
At the same time, there are some other solution providers where the machine learning model requires marketers to start running the campaign, which means that you incur costs for it to learn over two to three weeks. So actually, the first two to three weeks are actually meant for learning, not delivering the results.
After a marketer decides what kind of model best suits their needs, what are the next criteria?
It’s important that advertisers understand and align on the objectives of the campaign from the very beginning, especially with the creatives. Creatives are the very first thing seen by the user. So, providing a variety type of ad formats is very important, such as banner, videos and so forth, to ensure that you can cast the net as wide as possible. [This is] because there are some publishers out there, for example, who will maybe accept 15-second videos, but not 30-second videos; but there are other publishers who can accept 30-seconds videos. So, a variety of creative formats is actually very important.
How do you think AI will help marketers be more effective? What success metric should they be focusing on?
There’s a need to always think in advance. Of course, digital marketers will be aware of what’s going on in the industry and the changes in the landscape. They should be asking their partners what they are doing about these trends? What’s coming up in their product roadmap so that they can help deliver the performance to the next level?
These are all the things that they can look out for, in addition to just looking at the day-to-day performance deliverance.
