Inside Rakuten AI: Neil Primozich on Rakuten AI-powered ads empowering merchants

In this series, we sit down with Rakuten AI leaders for a deep dive into the stories behind this transformative technology and the inspiring individuals driving Rakuten’s vision of AI for all. Watch this interview and others in the Inside Rakuten AI series on our YouTube channel.

Digital advertising is a dynamic field where artificial intelligence is already adding value by driving greater personalization and better return on investment for advertisers. The diverse Rakuten Ecosystem, which includes more than 70 different services, offers plenty of digital touchpoints for advertisers to reach customers.

We met with veteran industry expert Neil Primozich, General Manager, Merchandising and Advertisement Department, to learn about his unique journey from advanced research to practical tech. He offered an in-depth look under the hood at the AI-powered platforms driving Rakuten’s advertising success.

From particle physics to pricing power

Primozich’s professional path is anything but conventional, charting a course from the highly specialized realm of physics to revenue-driven retail and advertising. “I chose physics in college because I thought it was cool,” he recalls. “But by the time I was in graduate studies, literally only 25 people in the classroom would understand the topics. And it’s not because it’s super genius, it’s just very specialized.”

Soon he began to think about a career with more tangible impact. “Progress in physics is very slow. You’re working on a very small piece, and you don’t see the big picture,” Primozich shares. He shared the moment when it all hit him. “I saw some construction workers building a highway that everybody sees, knows and can appreciate,” he recalls. “I wanted to build something like that, to show my kids instead of only publishing a paper in a journal that nobody will ever read.”

A turning point came when Primozich saw a job posting for retail price optimization, a nascent field at the time. This was “before AI was a thing,” he notes. “I joined a 25-person startup that quickly grew, eventually being acquired by SAP. We used quantitative modeling to predict sales and inventory, helping retailers optimize their pricing.”

Over a decade of innovation at Rakuten Ads

In 2010, Primozich joined Rakuten where he’s been at the heart of the company’s evolving data and advertising strategies. “I began in the Next Generation Search team, which later grew into the Big Data Department,” he said. “My early work involved price and inventory optimization and then I pitched an ad platform for one-on-one coupons on behalf of merchants, known as Smart Coupons. This really got me into to ads.”

His chance to prove the power of ads arrived when Rakuten Group Chairman and CEO Mickey Mikitani challenged the development teams to compete with Google in A/B tests to bring up Rakuten’s search ads performance. “I raised my hand to propose a third alternative platform for search ads and put together a solution in two weeks with just two other people,” Primozich shared. “My goal was just to prove that Rakuten could compete with Google at their own game of search.”

“Our new platform leveraged an existing Search Quality Improvement platform that tracked the entire customer journey and search, tying it back to original queries,” he explained. “By using competitive statistics and grouping search queries, we aimed to gather more comprehensive signals to display relevant ads. This innovative approach allowed us to match Google’s own search ad system in that particular test with defined tasks and conditions.”

Impact of AI on search

Rakuten’s advertising landscape primarily consists of two types of ads: Search Query Ads and Personalized Display Ads. Search Query Ads are straightforward: “A user can enter, Nike shoes and we will show Nike shoes and similar shoes from whoever is paying the most. Personalized Display Ads, on the other hand, are more proactive,” Primozich clarifies. “This involves showing ads for items that may or may not be actively searched by that customer and prominently displaying them on Rakuten Ichiba’s top page.”

The advantage of specialized AI models in advertising boils down to efficiency and cost. “Running an LLM for tens of thousands of requests per second can quickly become prohibitively expensive,” he explains. “Specialized models, particularly for retail, are far more cost-effective. If a smaller model, trained to identify a specific item like summer coat finds the exact item, that smaller model is worth a lot,” he notes.

These models are also nimble enough for iterative learning through reinforcement learning, a process too costly for large LLMs. “In general, a small model can try and fail billions of times. It doesn’t cost so much over the long-term and each time it iterates it gets a little bit better.”

In practice, multiple specialized models often work in concert. “A search model might be composed of eight or nine smaller models, to determine both item relevance and click probability. The combined cost of these multiple smaller models is drastically cheaper than one large one, making the system more efficient as well,” Primozich commented.

Empowering merchants with Rakuten ad platforms

Rakuten’s advertising prowess is significantly bolstered by two key platforms: the Rakuten Promotion Platform (RPP) and the Price and Inventory Optimization Platform (PIOP). “RPP is the merchant-facing tool where they can enter their budget, and then we show their items on Rakuten Ichiba’s search page, top page and various locations throughout,” Primozich explains. “PIOP manages selling products and pricing for selling, including inventory management, while RPP handles the advertising aspect. These two platforms empower merchants by automating complex decisions, allowing them to focus on their core business.”

“PIOP analyzes historical sales data to predict how much each item will sell at various price points in the future,” Primozich shares. “This is crucial for inventory management, helping merchants decide how much stock to purchase for sales events or how to price items to clear excess inventory. The platform also tracks budget distribution across different channels and ad locations, optimizing for return on ad spend. The ultimate goal is to constantly try to optimize for the merchant given their ad budget.”

The success of Rakuten’s AI-driven advertising is measured by a clear and impactful KPI: Return on Ad Spend (ROAS). “We take a lot of pride in making sure that, by far, the return on ads spent from Rakuten is the highest in all of Japan,” Primozich proudly states. “We have an average ROAS of 1,000% This means that for every yen a merchant invests in advertising, they can expect a return of nearly ten times that amount in sales.”

Beyond ROAS, Primozich’s team focuses on ad consumption or how much of the ad budget is utilized. They also tackle the cold start problem for new merchants or items with no past sales history for predictions. In such cases, the models predict click probability to give exposure for that merchant, ensuring that even new entrants have a chance to succeed.

A future of ads that bridge digital and physical with generative creativity

Looking ahead, Primozich sees exciting ad trends that benefit both online and brick-and-mortar merchants. “The vision is to help physical stores optimize pricing and inventory, and then use digital ads to match ads with the people that would be interested in that sweater you’re selling. It’s kind of a win-win for everybody.”

AI is also accelerating and improving existing systems. “The one pain point for many merchants has always been which kind of image do they show to the customer” Primozich identifies. “Traditionally, this was a very time-consuming and manual process. However, AI LLMs are really gifted at quickly making diverse ad creatives for A/B testing and faster deployment.”

Primozich’s journey from theoretical physics to building AI advertising with tangible results at Rakuten is a testament to his adaptability and passion for empowering merchants. With advanced AI platforms, specialized models and strategic KPIs, his team is commited to supporting merchants in achieving the highest return on their ad investment, fostering a thriving ecosystem for all.