avatar image

Machine learning has been around for a long time. First coined in 1959 when the technology was used to detect speech patterns and sonar signals. Now, almost a century later, from physics to farming, its applications are still expanding. One of the fields where it has found its best fit is marketing.

We sat down with Rahul Verma, Director of Product Management at Moloco, to discuss the differences between machine learning and AI, best practices, and much more.

Can you please introduce yourself and tell us about your role at Moloco? What are your responsibilities?

"Thank you for the introduction, and it's a pleasure to be here today. My name is Rahul Verma, and I’m a Director of Product Management at Moloco. I’ve been fortunate to work in the technology industry for over 20 years, building adtech and machine learning capabilities at companies like Microsoft and Meta.

As the industry evolved, I was increasingly drawn to the intersection of adtech and machine learning. That’s where I’ve dedicated my efforts for the last eight years, continuously seeking opportunities to drive innovation and results by leveraging data and algorithms. It was during this time that I joined Moloco. At the time, they were experiencing hyper-growth. I lead the initiatives to advance our platform’s performance for advertisers via machine learning and privacy-safe solutions."

Machine learning and artificial intelligence are two terms that are often used interchangeably. How do you understand the differences between the two?

"I see Artificial Intelligence (AI) as a big umbrella term. It refers to intelligent systems that can simulate humans in terms of reasoning, learning, and making decisions to perform tasks in the real world. This includes machine learning. 

The key attribute of machine learning systems is the ability to make decisions based on datasets to identify patterns, make predictions, and most importantly, self-learn and improve."

Machine learning has been a part of digital and programmatic advertising for nearly a decade. Why is machine learning such a good match for advertisers? And how is Moloco using it?

"Machine learning has fundamentally transformed advertising. In the early 2000s, it was exclusive to big tech companies like Google and Meta. However, within the next decade, it became available to more companies with the adoption of cloud computing - which democratized access to the massive computational power required by advanced machine learning. By automating everything across the adtech stack including targeting, bidding, and optimization, machine learning freed up advertisers to focus on their marketing strategy. Nonetheless, it takes incredible skill and infrastructure to build and run models at the low latency that’s required in adtech.

In programmatic advertising, DSPs need to process billions of bid requests to understand which bids are worthwhile and how much to bid based on a myriad of signals. To be successful, machine learning models need to be able to make predictions in real-time, and this challenge is magnified on the open internet.

By automating everything across the adtech stack - machine learning freed up advertisers to focus on their marketing strategy

As our CBO Sunil Rayan discussed last month, the open internet is a large — and largely untapped — opportunity. It’s an increasingly vast and complex ecosystem where advertisers need to navigate obstacles like fraud, first-price auctions, privacy regulations, and many other hurdles. These challenges all require advanced machine learning models to be successful. This was the very principle on which Moloco was founded — since the beginning, we’ve built our business based on operational machine learning, which enables real-time, autonomous decision-making at a massive scale. Our number one investment area has been hiring world-class machine learning engineers to develop and refine our algorithms and run at extremely low latency. This investment has enabled us to help mobile app developers acquire high-quality users, for the right price, across billions of bid requests from all major exchange partners.

We’re focused on enabling our advertisers to succeed through advanced machine learning. When they succeed, we succeed."

What would you say to others looking to leverage machine learning? What best practices would you recommend to get the most out of it?

"Let’s face it — “machine learning” has become quite the buzzword. That can be super overwhelming and challenging for advertisers to distinguish what different solutions are capable of and which ones are most useful for them. I recommend three key practices for advertisers to get the most out of machine learning: 1) ask for product deep dives rather than take offerings at face value, 2) make sure you’re passing holistic 1st-party data in real-time, and 3) be open to testing with different partners. 

Not all machine learning is created equal, you need to test different partners to select the platform that maximizes outcomes and best aligns with your long-term business goals

Everyone says they do machine learning, but not all machine learning is the same. To find a solution that solves unique challenges, advertisers need to look under the hood. That includes asking questions about the machine learning technology, infrastructure, and how their models are differentiated.

Second, the adtech ecosystem is constantly evolving, and what matters most is how you adapt. To solve real-time problems and reach users effectively, you need real-time data. For advertisers, this means sharing real-time data with your partners and demanding transparency and insights so you can apply these learnings to your overall business strategy.

Finally, since not all machine learning is created equal, you need to test different partners to select the platform that maximizes outcomes and best aligns with your long-term business goals."

What are the biggest mistakes you’ve seen advertisers make when getting involved with machine learning?

"Looking across all of our advertisers, we’ve seen two common mistakes: 

  1. Expecting the same ramp-up period for different goals. In fact, different bidding types require different learning periods based on the available data for model training. It takes longer for models to ramp up on deeper-funnel goals like ROAS since trainable data becomes more sparse. As a generic example, it can take 1000 installs to get 5 user in-app actions, which means these in-app actions take more time to learn. In short, it’s important that advertisers have the right expectations ahead of launching a campaign. 
  2. Not sharing real-time, first-party data. Like we discussed earlier, data is vital to performance because models are only as good as the data they train on. By not sharing robust first-party data, advertisers won’t be able to capitalize on the full benefits of machine learning."

In your opinion, who is using machine learning in an exciting way?

"Adobe is one of many players supercharging their business with machine learning. Its Sensei solution automates time-consuming tasks for creative professionals. It enhances Adobe’s Creative Cloud Suite and other products with predictive analytics and personalization features. Most recently, Adobe integrated the power of generative AI in design through Generative Fill, which speeds up workflows and lowers the barrier to entry by enabling users to create and edit designs with simple text prompts."

What are you most excited about for the future of machine learning and its applications?

"Adapting to privacy rules and regulations is an immense opportunity. In the post-ATT world, user data is harder to come by, so advertising success will hinge on large models that are able to uncover and leverage additional signals to make accurate predictions. To preempt these headwinds, Moloco is working closely with our ecosystem partners to adapt our capabilities to privacy-first frameworks such as SKAN and Android Privacy Sandbox."

Do you have any concerns about the future of machine learning? What are they?

"On the flip side, data and privacy is a growing area of concern; however, this is where advanced machine learning can help. Machine learning models will need to adapt to using fewer user-level signals and supplement this loss with contextual or campaign data. Only advanced machine learning models will be able to successfully adapt. There’s no doubt that machine learning will only become more powerful and useful. Leveraging it in the right way will hinge on advertisers clearly defining their goals, testing various partners, and extracting insights that can be applied to their overall business." 

Fill in the blank. “When it comes to mobile marketing, bet on  ________.”

"Companies that invest time and resources in advanced machine learning technology."

Let's put these tips to good use

Grow your app business with ironSource