From his days at Google and Snowflake to his current role as the Chief Machine Learning Officer at Moloco, Tal Shaked has been at the forefront of integrating machine learning (ML) into the tech landscape. In this exclusive interview with Marketing in Asia, Tal shares his valuable insights on how ML has evolved within tech companies, the strategic application of machine learning in marketing, and the critical ethical considerations that companies must prioritize as they deploy these technologies.
In your transition from leading projects at Google and Snowflake to becoming the Chief Machine Learning Officer at Moloco, what fundamental shifts have you observed in machine learning’s role within tech companies? How do you believe companies should evolve their approach to machine learning to stay ahead in today’s competitive landscape?
One of the key changes I’ve noticed is the growing accessibility to technologies that enable more companies to develop Machine Learning (ML) powered products. The increasing amount of open source ML algorithms and models combined with better Cloud computing environments and specialized hardware has enabled many more companies to develop and deploy various flavors of ML with significantly fewer resources and technical expertise.
Making ML capabilities more accessible has been a game-changer, especially for smaller and medium-sized tech companies. Businesses that may have previously lacked in-house ML expertise or budgets to build complex ML models from scratch can now leverage what is available in open-source, Cloud, and other pre-built, off-the-shelf solutions, to drive meaningful business impact.
When it comes to the future, companies should focus on the core business problems they are trying to solve, rather than getting swept up by the industry hype. The key is to identify areas where ML can have the most meaningful impact on their business and target customers, and then strategically apply the technology to those specific use cases.
For example, while chatbots powered by natural language processing may be a trendy application of ML, they may not necessarily be the best solution for every company’s customer service needs. Similarly, the hype around self-driving cars has led many companies to explore this technology, but the reality is that it remains a complex and challenging problem to solve.
The most successful companies in leveraging ML are those that can clearly articulate the specific business challenges they are trying to address and then carefully evaluate how ML can be applied to solve those problems. This requires having the right people and expertise on board – whether that’s building an in-house ML team or partnering with specialized providers like Moloco.
Drawing from your analogy of chess and machine learning, could you elaborate on the strategic moves that marketers need to adopt to leverage machine learning effectively? Specifically, how can machine learning navigate the complex challenges of data privacy and user control while ensuring personalized and impactful ad experiences?
In marketing, much like in chess, success requires foresight – you’re making a move now that may eventually lead to you winning the game many moves later. Similarly in advertising, machine learning empowers advertisers to make more strategic choices. By showing ads to someone now who is more likely to become a high value user in the future, you are setting yourself up for long-term success. Showing the right ad to the right user at the right time for the right price requires learning patterns from very large datasets, which is similar to leveraging experience in chess to look many moves ahead and accurately evaluate any position.
In order to learn this we need to leverage large amounts of data from other users that we and advertisers have engaged with to distinguish high value from low value users for specific campaigns. There are many challenges in obtaining and using first-party and other data sources in a privacy compliant manner, to optimize for showing relevant, personalized ads to users and high-value outcomes to advertisers.
The privacy – personalisation paradox is an issue that’s been around a while. With the delays in phasing out cookies and subsequent adoption of robust and compelling alternatives, most marketers are uncertain about how to proceed. Some are postponing their approach to see how the landscape evolves, while others are anxious and experimenting with whatever is available today or under development to stay ahead and be prepared for anything.
Additionally, there are companies sitting on a goldmine of first-party data that is collected with consent, and is readily available to fulfill a privacy-first advertising strategy. Many of these companies lack the infrastructure, expertise or compliance mechanisms to harness the power of first party data effectively.
This is where ML comes in to help marketers maximize the ROIs of their campaign. Through ML, companies can effectively learn from their first-party data to gain deep insights into customer preferences and behaviors. This enables them to build models and tailor their marketing efforts with precision in a privacy compliant manner.
Also Read: Elon Musk Proposes Posting Fee for New Users on X to Combat Bots
In a future increasingly shaped by machine learning and AI, what ethical considerations should companies prioritize to ensure their AI developments are beneficial and not detrimental to society?
We are part of a society that is evolving, constantly churning out new innovations and building products. The same goes for ML and AI – as technology marches forward, it opens up opportunities for good and unfortunately, misuse by bad actors, despite the good intentions of the innovators. In this day and age, people are increasingly aware of the dual nature of these technological advancements and therefore there are organizations and regulators which help to identify potential loopholes and pitfalls which can help provide developers with guardrails and awareness.
Machine learning, similar to many other tools and technologies like media and explosives, when used the right way can create a lot of benefits. However, occasionally those same technologies can be used either intentionally or unintentionally in ways that might cause harm to people. As machine learning and AI evolve, we need to take great care to try and ensure that they are designed in a way that a reasonable human will get the right outcome that is beneficial to them. Furthermore, if a bad outcome occurs, ideally that intent can be mapped back to the human operator rather than the technology itself.
With your extensive background in both machine learning and chess, how do you perceive failure and learning from it in both domains? Can you share a personal anecdote where a setback in either field provided you with significant insights or a breakthrough?
Losing, in chess or otherwise can be a humbling experience, but it also presents a valuable opportunity for growth.
Each time I was outplayed in a game of chess, I would analyze the game and review each move and the decision-making processes that led to my losses. This exercise revealed gaps in my understanding of many positions, as well as areas where my emotional state had influenced my decision-making. I would aggregate these analyses and find patterns that led to the mistakes and figure out ways to improve my decision-making.
This is also how machine learning works in advertising. The computer tries to find patterns that distinguish valuable users from non-valuable users. Whenever the resulting models make predictions on new data, the system later gets back the responses from the users. Those can highlight mistakes or bad predictions which then get fed back to the ML system so that it can continue to improve over time.
Developing high performing ML systems is very complicated, and at a minimum involves designing features, model architectures, and optimization algorithms that can best map patterns in the data to models for making future predictions. Along the way there are usually many bugs and failures such as incorrect assumptions about the domain and poorly constructed objectives. It can take many iterations to work through those challenges, just like it can take trying out many different moves in chess to properly evaluate a position. And through trial and error, you collect data and experience to help you make decisions on which moves to play in certain positions.
Whether in chess or machine learning, the path to success is rarely linear. Setbacks and failures are inevitable, but it is how we respond or learn from each iteration that truly matters.
As machine learning technologies continue to evolve, what emerging trends or innovations do you foresee playing a pivotal role in transforming the advertising landscape? How should companies prepare to integrate these advancements into their strategies?
As machine learning technologies continue to evolve, I’m excited about the emerging trends and innovations that will shape the future of advertising.
One particularly important area of innovation is getting better data to measure and model the outcomes of advertising campaigns. Showing an ad to a user is often the beginning of a long journey in converting that user to a happy, paying, and high-value customer. The better we understand that journey and can measure the outcomes from showing ads, the more we can learn and optimize future ads we show. This has become relatively simple to do for shallow events such as showing an ad that a user immediately clicks. However, a user that clicks on an ad is not nearly as valuable as a user that clicks and makes a purchase a day later. Similarly, a user that makes 10 purchases over the next month is more valuable than a user that just makes one purchase. Being able to tie all those pieces of data back to the ads can be challenging. For example, imagine a user that is shopping for a car. They might click multiple ads for cars, visit many websites, go to some dealerships to test drive a few cars, and eventually make a purchase a few weeks later. As technology enables more of those longer term outcomes to be mapped to specific ads, the more ways ML can be used to optimize for those outcomes.
Another exciting development is the rise of hyper-personalized advertising. Traditional advertising may target broad demographics such as users who are white, male, and between the ages of 18-24. People within those broad demographics can actually have very different interests and respond differently to the same ads. The more data we have about users, the better we can personalize ads to their interests and specific needs. We can see this on sites like Amazon and Netflix that have our complete purchase and viewing history which are used for future recommendations. Generative AI is opening up many new possibilities here. Traditional search has been done in a text box with a few search terms. Instead people pass in whole paragraphs to chatbots expressing their interests and much more nuance. That can lead to a much more personalized ads experience in the longer term. Another example is shopping for clothes online. Imagine being able to view a model in your likeness as you consider different outfits..
It’s an exciting time for advertisers as ML technology opens up new possibilities for connecting with consumers on a deeper level.
As we conclude our enlightening conversation with Tal Shaked, it’s clear that the journey of machine learning is far from static. With his deep understanding of both the technical and strategic facets of ML, Tal has provided a roadmap for companies looking to harness the power of AI responsibly and effectively. As machine learning continues to reshape the advertising world, Tal’s perspectives will undoubtedly aid marketers in navigating this evolving landscape, ensuring that innovation aligns with ethical standards and consumer benefits.
