Hi everyone. I am Ankita Saxena, senior project manager at Amazon and alumni of Carnegie and University. I work on AI driven products and focus on recommendation systems and how these are shaping both customer experiences and business strategy. Today, I will walk all of you through how these systems have evolved over time and from just, you may also like feature, it has become a core infrastructure of multiple industries, powering their business decision making and improving their customer engagement. So we interact with recommendation system almost every day when Netflix suggests you a show based on your viewing pattern. Or Spotify starts building your playlist or website, highlighting what we may like the next. These all feel very simple features, but there's lot of intelligence behind them in transforming industries, and these are, that INT intelligence system is now empowering the core, core business infrastructure of. Different industry areas. So we'll today we'll talk about how this sim how from simple suggestion to strategic infrastructure this has changed the four different industry and what is its application across different industries. We'll talk about e-commerce platforms, streaming services, digital retail and enterprise marketplace, the same technology, how it is adopted across very different industry. So let's go deeper into the each industry and how the recommendation system works and how it is impacting the different aspects of the business decision making there. Let's start with e-commerce platform first. In E-commerce. Now the system, the recommendation system has evolved beyond the production. For example, if customer is frequently buying a what certain kind of reusable water bottle, along with a fitness care, the systems forced that pattern leading to. Leading retailer to bundle them or increase supply or actually create their own private level brand. That's the business inside. Now, the system gives you the engine, and this way the engine is shaping both the customer experience and inventory decision making. Let's go deeper into how the entire technical architecture works in e-commerce for the recommendation. So the recommendation to architecture in commerce starts with data collection where real time capture of user behavior, transaction and contextual ness gets collected when customer is interacting on the platform. And then use with the collaborative filtering, the item to item analysis happens, identifying what are the purchase pattern, what is the product affinities for the customer. Which gets feed fed into the personalization engine, which is like dynamic recommendation generation, tailored to individual user profiles. And the aggregation of these insights give gets fed into the business intelligence which helps businesses into inventory planning, supplier management. Even the product development decisions. So just think about if be shoppers, view a jacket, but don't buy it. That system will flag it as a friction, and the team may respond by either adjusting price, improving the images, or testing new messaging of the jacket to the customer to improve the conversion ratio. Now let's move into the next industry, which is a streaming platform. Streaming platform is focused on hyper personalized discovery, unlike the e-commerce platform. And it is more complex because it not only look at collaborative filtering, it looks at MicroGen genre metadata tagging, and. Multimodal analysis using visual element audio characteristics, narrative structure, plot content, and many more. And what this recommendation supports and helps with it guides the investment decisions the production decisions, regions acquisition targets for licensing deals, and help platform understand which genres and formats will resonate with specific audiences and sector. Yes. So just think about if the millions of your customers are viewing or binge watching light-hearted comedy drama content platform will definitely invest more in that styles because the data reveals there is a long term demand and which can eventually convert into a long time customer lifetime value. So the recommender drives both user engagement because customer. The person, the customer is seeing what they want to see. And obviously business investment because now customer business is super clear on where to invest to get more value out of the customer. The most interesting piece of this is how the multi-model analysis recommendation system works and what is the detail of it. So when you look at multi-model analysis in streaming. It does collaborative filtering. It has a collaborative filtering layer, which where it analyzes viewing patterns across millions of users to identify content with similar audience appeal. Then it goes to micro genre metadata, like which is grammar tagging system, creating thousands of content categories beyond traditional genre. For example romantic is a traditional genre, but it'll go more deeper into. Romantic with multiple couples and a Christmas team. That's the granularity. It goes on. Then it analyzes video and visual and audio and signals which uses computer vision and audio processing and attract extra characteristics like cinemagraphic style, pacing, and mood. Just think about, if the system sees a viewer prefers a warm color balance and slower pacing, it recommends visually similar shows. Even if the story is different, it goes far beyond clicks to understand aesthetic preference of the customer, and that's where you bring in lot of engagement from the customer. And then the narrative structure mapping na, natural language processing analysis. Plot elements, themes, and character ask for deeper matching. That's how the entire multimodal analysis happens, which gives a hyper personalized recommendation to the customer. Now, my most favorite, the digital retail deal, and I will talk more about the fit uncertainty. Because that's the biggest pain point as of now in the digital retail industry. And obviously a misfit product ends up into a poor customer experience, lesser customer engagement, and obviously higher return rates impacting the company's revenue. So this recommendation system. Where we use visual similarity algorithms and ar try-on capabilities is surely helping to solve this problem. So just think about if a shopper uses virtual try ons for sneakers, the model will suggest styles that match their first shape and past certain behavior. This reduce uncertainty, the number of decision points a customer has to take. And obviously in town reduces the returns and the impact of this recommendation system on digital retail is phenomenal. It improves the conversion rates, obviously reduced reduction rate. We have seen a reduction of almost five to 8% in return rates. Wherever we are using the visual similarity or virtual try on features. And in turn, this has improved our conversion rate by 15 to 25% on average across the different digital retailers, which, and it made customer really happy because now they are able to get better fitting, more relevant alternatives without much. Multiple returns, and obviously this improves the engagement. A very interesting engagement data. We, what we have seen is because customer is seeing a lot of visually similar item, it ended up engaging more and the session duration has gone up about 20, 30% wherever the visual similar. Recommendation system is being used, and obviously more session time directly converts into increased order value, which is in the tune of eight to 12%, which is a huge impact to the p and l. And then the last, the fourth one, the enterprise marketplaces, which is a two-sided optimization. My optimization system, and I will say it is the most complex of. Any recommendation system because it has to work in a multi-agent reach for reinforcement learning environment. It, the agent has to not only optimize for the buyer, it has to optimize for the supplier as well, and that is where the complexity comes in. Let's dig deeper into how this two-sided optimization system works. And what is the benefits of using this? So this multi-agent reinforcement learning architecture works on four different components. There is a seller agent, which works on maximizing visibility to qualified buyers while managing inventory efficiency. Then another one is a platform agent, which balances marketplace health, liquidity, and long-term sustainability. And then the re regulatory agent, which ensures compliance within regional laws, logistics and payment requirements. And the fourth one, the buyer agent optimizes for relevant supplier discovery and best value matches. One of the great example you can think about is if a wire wants low shipping cost, but a seller wants to clear the access inventory. The system learns to pair the buyer with the nearby supplier to satisfy both sides, which is like optimizing for the both the agents and the agent balances competing, goes very intelligently benefiting both and creating a win-win situation for both s seller and the buyer. That's the impact of the recommendation system. So another example is if the system deducts there is a rising interest in sustainable products, then it gives a recommendation to the product teams to expand that category, and marketing shifts to the messaging and the insight flow across the entire system. That's how the entire recommendation system impacts the businesses. So I would say, the entire recommendation system has multiple business impacts across operational excellence delivering a lifetime customer value and revenue impact. So you may think about when it comes to operational excellence, the recommendations system helps supply chain optimization. When it comes to demand forecasting, inventory management, which is all driven by recommendation patterns, same for the resource allocation, like where we should invent for invest for the content, if it's a streaming services or what should be the green field areas for the new product development, where we are seeing a huge demand, but lower conversion and the patterns, there are common product feature patterns under each demand. And where should we expand the market Next, what should be our new next private level? And obviously, and the most important is efficiency gains. We with the recommendations digital system, we remove all the dependency on manual search and all the friction and lowering the transition cost. And then if we focus on the customer value with the. Better recommendations and the products which customer is looking for. You bring a lot of retention. You improve customer retention and loyalty and creates a personalized experiences for the customers, which gives you a lot of competitive edge and definitely the revenue optimization. You can do cross selling, upselling. You can discover the new revenues work revenue streams. Using the recommendation system. For example, if two different products are bought all the time together, you can bundle it and sell it, cross sell it to the same customer, and under the recommendation system, and the most important, the market intelligence, it gives very deep insights into the customer behavior, what customer is looking for, and what's the new product we should be building, which customer? Think of, but not finding as of now. So to summarize the recommendation system is redefining the personalization as a core foundation layer of all the businesses. So it's no longer just a small feature recommendation. It has become a fundamental architecture. Powering business operations and it is a cross-functional integration. So once we have this foundation up and running, it not only impacts one section, supply chain, it, or marketing plan, it impacts the insights of this recommendation system flows across all the sites of the business, marketing, supply chain, product development. Business strategy, customer ag engagement, all of the functions of the business. And once you start working backwards from the customer, obviously it creates a competitive differentiation. Superior personalization creates a defensible mode through a network effect and advantages the same percent. It becomes a foundation and not a feature. And I would say companies that understand behavior at scale, deliver more relevant experiences, creating an advantage, that new customer can't easily copy that it, and that insights become a strategic mode because you are the one who knows your customer the best. To end with AI powered recommendation system represents the convergence of customer experience excellence, and a strategic business intelligence organization. And the organization that mastered this convergence will define the competitive landscape of their industries in the near future. So the strategic imperative is clear. Recommendation system must be viewed. Not as isolated technical implementation, but as integrated business platform that drive decisions across entire organization. From e-commerce inventory strategies to streaming content, investment from digital retail conversion optimization to enterprise market marketplace security. These systems are reshipping how industries operate and compete. Thank you so much for your time and I hope this showed how recommendation system act as strategic in engines, not just customer features. And thank they. Have a great day. Thank you so much.