Enhancing Product Recommendations

Product recommendations play a critical role in helping customers discover relevant items and drive engagement and conversion. Even small improvements in recommendation quality can compound at scale, especially in large retail platforms. In a recent tech blog, data scientists from CVS Health shared how they enhanced their existing “You May Also Like” recommendation module by integrating large language models into the system. - At a high level, the recommendation approach is grounded in “similarity”. For any given product, the goal is to identify other products that are similar based on rich product attributes. The overall workflow follows a familiar pattern: generate product embeddings, measure similarity between products, and surface recommendations accordingly. - The key challenge, however, lies in data quality. Not every product comes with a well-written title or detailed description that can be used to generate meaningful embeddings. Some products have extremely short or uninformative metadata, which limits the effectiveness of traditional embedding-based methods. - This is where LLMs add value. For products with sparse or low-quality text, the team leveraged ChatGPT to generate a roughly 200-word product summary, expanding on the product’s purpose, usage, ingredients, and key attributes. These enriched summaries provide higher-quality inputs for embedding generation, improving both coverage and recommendation accuracy. While rebuilding an entire recommendation engine around GenAI can be appealing, this case is a good reminder that incremental, well-targeted improvements often deliver real impact. Integrating LLMs to strengthen weak points in existing systems can be both practical and powerful—and this example is a nice one to keep in mind. #DataScience #MachineLearning #GenAI #LLM #ChatGPT #Recommendation #IncrementalImprovement #SnacksWeeklyonDataScience – – –  Check out the "Snacks Weekly on Data Science" podcast and subscribe, where I explain in more detail the concepts discussed in this and future posts:    -- Spotify: https://lnkd.in/gKgaMvbh   -- Apple Podcast: https://lnkd.in/gFYvfB8V    -- Youtube: https://lnkd.in/gcwPeBmR https://lnkd.in/g7Xv3tG4

AI products like Cursor, Bolt and Replit are shattering growth records not because they're "AI agents". Or because they've got impossibly small teams (although that's cool to see 👀). It's because they've mastered the user experience around AI, somehow balancing pro-like capabilities with B2C-like UI. This is product-led growth on steroids. Yaakov Carno tried the most viral AI products he could get his hands on. Here are the surprising patterns he found: (Don't miss the full breakdown in today's bonus Growth Unhinged: https://lnkd.in/ehk3rUTa) 1. Their AI doesn't feel like a black box. Pro-tips from the best: - Show step-by-step visibility into AI processes - Let users ask, “Why did AI do that?” - Use visual explanations to build trust. 2. Users don’t need better AI—they need better ways to talk to it. Pro-tips from the best: - Offer pre-built prompt templates to guide users. - Provide multiple interaction modes (guided, manual, hybrid). - Let AI suggest better inputs ("enhance prompt") before executing an action. 3. The AI works with you, not just for you. Pro-tips from the best: - Design AI tools to be interactive, not just output-driven. - Provide different modes for different types of collaboration. - Let users refine and iterate on AI results easily. 4. Let users see (& edit) the outcome before it's irreversible. Pro-tips from the best: - Allow users to test AI features before full commitment (many let you use it without even creating an account). - Provide preview or undo options before executing AI changes. - Offer exploratory onboarding experiences to build trust. 5. The AI weaves into your workflow, it doesn't interrupt it. Pro-tips from the best: - Provide simple accept/reject mechanisms for AI suggestions. - Design seamless transitions between AI interactions. - Prioritize the user’s context to avoid workflow disruptions. -- The TL;DR: Having "AI" isn’t the differentiator anymore—great UX is. Pardon the Sunday interruption & hope you enjoyed this post as much as I did 🙏 #ai #genai #ux #plg

Can Recommender Systems Actually Know When They're Wrong? Researchers from Tsinghua University have developed a breakthrough approach to help recommendation algorithms become "self-aware" of their prediction quality before any user interaction occurs. The Core Innovation: List Distribution Uncertainty (LiDu) Traditional uncertainty methods focus on individual item predictions, but recommendations are fundamentally about ranking lists. LiDu addresses this by calculating the probability that a recommender will generate a specific ranking order based on prediction distributions of individual items. How It Works Under the Hood: The system models each predicted score as a Gaussian distribution with both mean (expected score) and variance (uncertainty). For any two items, it computes the probability that one ranks higher than another using these distributions. The overall uncertainty becomes the negative likelihood of the most probable ranking the model generates. Technical Implementation: Three uncertainty quantification methods were tested: - MC Dropout: Uses dropout layers during inference with multiple forward passes to estimate variance - Deep Ensembles: Trains multiple models with different initializations  - Variational Bayesian: Replaces the final layer with a Bayesian weight matrix that outputs both scores and prediction variance Key Findings: Testing across six real-world datasets (Amazon, MovieLens, Douban, XING, Yelp) with five different recommenders (BPRMF, LightGCN, SimpleX, SASRec, TiMiRec) revealed strong negative correlations between uncertainty and performance. Higher uncertainty consistently indicated lower recommendation quality. Practical Applications: This label-free performance estimation could enable data augmentation for sparse positive samples, user-specific recommendation strategy adjustments, and model selection without requiring user feedback - potentially bridging the gap between offline and online evaluation. The work establishes an empirical connection between recommendation uncertainty and performance, opening pathways toward more transparent and self-evaluating recommender systems.

You click "play" on Netflix. In 200 milliseconds, a recommendation engine just processed millions of videos. Most ML engineers know these systems exist. Few understand what's actually running under the hood. I spent the last 6 months building a complete deep-dive series on production recommendation systems — from first principles to the exact architectures running at YouTube, Spotify, and TikTok. Here's the complete roadmap: 🎯 Foundation Layer 1️⃣ RecSys Fundamentals — Content-based, collaborative filtering, and hybrid approaches that power every modern recommender 2️⃣ How Recommendation Systems Learned to Think — The evolution from matrix factorization to transformer-based generative agents ⚡ Retrieval & Ranking Pipeline 3️⃣ The 3-Stage Funnel — How two-tower models, vector databases, and cross-encoders work together at scale 4️⃣ How YouTube Finds Your Next Video in Milliseconds — Two-tower retrieval, in-batch negatives, and the engineering tricks that make it work 5️⃣ Vector Search at Scale — IVF, PQ compression, and making 100M+ vector search actually possible in production 6️⃣ From Candidates to Clicks — The complete ranking stack: from 1,000 candidates to the one item you actually tap 🔧 Production Reality 7️⃣ Solving the Cold Start Problem — Contextual bandits, meta-learning, and LLMs for new users and items (how Spotify, TikTok, YouTube do it) 8️⃣ Beyond Ranking — How diversity, freshness, and business constraints turn a ranked list into a product-ready feed Every post includes: → Production architecture diagrams → Real code examples (PyTorch, Faiss, ranking models) → Case studies from actual systems → The engineering tradeoffs that matter Full series: https://buff.ly/GKEvulv If you're building RecSys or joining a team that does — this is your blueprint.

Most companies are using AI for efficiency. Some are accelerating value creation. A great case study is how Colgate-Palmolive is driving innovation. Here are specific ways they are embedding GenAI across innovation processes to substantlly improve research and product development. These come from an excellent article in MIT Sloan Management Review by Tom Davenport and Randy Bean (link in comments). 💡 AI-Driven Product Concept Generation Accelerates Ideation By linking one AI system that surfaces consumer needs with another that crafts product concepts, Colgate-Palmolive can swiftly generate creative ideas like novel toothpaste flavors. This AI-augmented workflow produces a broader product funnel and allows rapid iteration, enabling more employees to participate in the innovation process under guided human oversight. 🔍 Retrieval-Augmented Generation Enhances Data Reliability The firm’s use of retrieval-augmented generation (RAG) integrates company-specific research, syndicated data, and real-time trends from sources like Google search data. This approach minimizes the risk of hallucinations and ensures that responses are deeply grounded in verified, internal content—delivering more accurate market analysis and trend detection. 🤖 Digital Consumer Twins Validate and Refine Concepts Moving beyond traditional focus groups, the company has developed “digital consumer twins”—virtual representations of real consumer behavior. These digital twins rapidly test hundreds of AI-generated product ideas. Early evaluations show a high level of agreement between virtual feedback and actual consumer responses. This innovation speeds up early-stage concept validation and reduces reliance on slower, more limited human panels. 🔐 Democratizing AI Through a Secure Internal AI Hub Colgate-Palmolive’s AI Hub provides employees with controlled access to advanced AI tools (including models from OpenAI and Google) behind corporate firewalls. Mandatory training on responsible AI use, including guardrails and prompt engineering best practices, ensures that employees harness these tools safely and effectively. Built-in surveys and KPI tracking further enable the company to measure improvements in creativity, productivity, and overall work quality. 🌐 Bridging Traditional Analytics with Next-Gen AI for Measurable Impact By integrating traditional machine learning with cutting-edge generative AI, Colgate-Palmolive is not only boosting operational efficiencies but also driving strategic growth. This seamless blend supports tasks ranging from market research and innovation to marketing content creation—demonstrating a holistic, value-driven approach to adopting AI that is a model for other organizations.

🚀 How do you ensure your customers see what they want to see — not just what you want to show? With AI and ML becoming core to ecommerce (both B2B and B2C), product discovery is getting a lot of attention. And rightly so. But here's the truth: most recommendation engines fail not because the models are bad, but because the first two steps were never right. Let me explain. Many product managers (especially in fast-paced orgs) jump into building rec engines with a "let's plug in collaborative filtering and see how it goes" mindset. But without clearly defining what type of recommendation makes sense for your use case — and how it ladders up to a business metric — you're setting yourself up for rework. Here's how I approach it when working with teams: Step 1: Business Understanding: Start with the why before touching the how. ◾ What are you recommending? Products? Content? Users? Services? ◾What does success look like? Higher CTR? More revenue? Better retention? ◾Where will it show up? Homepage, PDP, cart, email, app banner? ◾What constraints exist? Does it need to be real-time? Can it be batched overnight? Without alignment on this, even the most advanced ML model will fall flat. Step 2: Choose the Right Recommendation Type: Now comes the how — but it should be tailored to your product + user journey. ◾Content-based filtering: “You liked this, so you’ll like these similar items.” ◾Collaborative filtering: “Users like you also bought this.” ◾Hybrid models: The best of both worlds — widely used in ecommerce and streaming. ◾Knowledge-based systems: Rule-driven, useful when personalization is constrained (e.g., insurance, banking). Let me make this concrete with a simple example: Imagine you’re building a recommendation module for a first-time visitor on your site who hasn’t logged in. If you apply collaborative filtering, it’ll fail — there’s no past data to compare. But if you use content-based filtering on the item they’re browsing and pair it with trending items, you instantly make the experience better. It’s not about which model is smarter. It’s about which makes sense for the scenario. Let’s be honest — your recommendation engine’s success doesn’t start with machine learning. It starts with product thinking. #AI #ProductManagement #Ecommerce #Personalization #RecommendationEngine #ProductStrategy I write about #artificialintelligence | #technology | #startups | #mentoring | #leadership | #financialindependence   PS: All views are personal Vignesh Kumar

Lately, I’ve been thinking about why choosing something to watch often takes longer than actually watching it. You open a streaming app, scroll for a while, watch a few trailers, switch genres, and sometimes end up rewatching something familiar. For years, most recommendation systems have been optimized around predicting what a user is most likely to click or watch next. In large-scale systems, this usually involves generating candidate content using embeddings and retrieval systems, ranking those candidates using machine learning models trained on engagement signals, and then presenting results to the user. But even well-optimized systems struggle with something fundamental: 𝐡𝐮𝐦𝐚𝐧 𝐢𝐧𝐭𝐞𝐧𝐭 𝐜𝐡𝐚𝐧𝐠𝐞𝐬 𝐪𝐮𝐢𝐜𝐤𝐥𝐲. A person’s watch history reflects what they liked in the past, but it does not always capture what they feel like watching in the moment. In reality, discovery often feels less like ranking and more like a conversation. Preferences are 𝐜𝐨𝐧𝐭𝐞𝐱𝐭𝐮𝐚𝐥, 𝐟𝐮𝐳𝐳𝐲 and 𝐞𝐯𝐨𝐥𝐯𝐢𝐧𝐠. This is where 𝐬𝐞𝐪𝐮𝐞𝐧𝐜𝐞-𝐛𝐚𝐬𝐞𝐝 𝐦𝐨𝐝𝐞𝐥𝐢𝐧𝐠 𝐚𝐧𝐝 𝐠𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐯𝐞 𝐀𝐈 start to change how recommendation systems can be built. Instead of treating every interaction independently, modern approaches can model user behavior as a 𝐬𝐞𝐪𝐮𝐞𝐧𝐜𝐞 𝐨𝐟 𝐚𝐜𝐭𝐢𝐨𝐧𝐬 𝐰𝐢𝐭𝐡𝐢𝐧 𝐚 𝐬𝐞𝐬𝐬𝐢𝐨𝐧. Transformer-based models are particularly well-suited for this because they can learn patterns across sequences of behavior. These systems can begin to understand how preferences shift during discovery rather than simply predicting the next click. In production environments, this often leads to 𝐡𝐲𝐛𝐫𝐢𝐝 𝐫𝐞𝐜𝐨𝐦𝐦𝐞𝐧𝐝𝐚𝐭𝐢𝐨𝐧 𝐚𝐫𝐜𝐡𝐢𝐭𝐞𝐜𝐭𝐮𝐫𝐞𝐬 that combine retrieval systems with generative or reasoning models: 💡Real-time user events feed feature stores that support both offline training and low-latency inference. 💡Embedding-based retrieval systems (e.g., two-tower models + ANN search) reduce millions of items to a few hundred candidates in milliseconds. 💡Ranking models score these candidates based on click probability, watch time, and completion likelihood. 💡Session-aware embeddings and sequence models capture short-term intent shifts during browsing. 💡Re-ranking layers enforce diversity, freshness, and exploration under strict latency constraints. Recommendation systems are gradually moving from 𝐩𝐫𝐞𝐝𝐢𝐜𝐭𝐢𝐧𝐠 𝐛𝐞𝐡𝐚𝐯𝐢𝐨𝐫 to 𝐮𝐧𝐝𝐞𝐫𝐬𝐭𝐚𝐧𝐝𝐢𝐧𝐠 𝐢𝐧𝐭𝐞𝐧𝐭, and from 𝐫𝐚𝐧𝐤𝐢𝐧𝐠 𝐜𝐨𝐧𝐭𝐞𝐧𝐭 to 𝐠𝐮𝐢𝐝𝐢𝐧𝐠 𝐝𝐢𝐬𝐜𝐨𝐯𝐞𝐫𝐲.

ChatGPT eCommerce drop: Part 3 (foundational Q&A) Q: Why should eCommerce leaders pay attention to ChatGPT’s shopping assistant? The way consumers discover and decide what to buy is fundamentally shifting, from keyword search to conversation. If your product content isn’t optimized for AI discovery, you're lagging. Q: How is this different from Google search or traditional marketplace discovery? Old-school search engines return a list of links or paid ads. ChatGPT returns curated, context-rich product suggestions with images, pricing, reviews, and direct buy links. Difference is that AI models understand intent, not just keywords. Instead of “best sneakers,” a user may ask, “What’s a comfortable walking shoe for traveling through Europe in the summer?” ChatGPT understands that nuance and recommends accordingly. Q: What powers ChatGPT’s product recommendations? It’s a mix of structured product data and contextual intent signals. Product metadata (titles, descriptions, tags, inventory) Real-world reviews with specific use cases or outcomes Signals of trust (brand credibility, availability, content quality) Integrations with platforms like Shopify and product feed partners The AI model then uses this data to recommend products that match the why, not just the what. Q: So what changes for brands now that AI is in the shopping flow? Discovery is an earned visibility game. You can’t just outbid, you have to out-relevance. Generic content doesn’t work; rich context wins. Volume of reviews matters less; specificity and clarity matter more. The brands showing up in ChatGPT’s results are the ones with deep, well-structured content and high-context product storytelling. Q: What are the key elements brands should focus on to stay visible in AI-driven shopping? Priorities: 1. Structured Data Implement schema markup across product pages. Use tools like Shopify’s native integrations to feed product info cleanly. 2. Contextual Product Descriptions Who is this for? What does it solve? What makes it different? 3. High-Context Reviews Prompt users to share how and why they used a product. 4. Review Accessibility Make reviews public, crawlable, and visible next to your products. 5. Feed Accuracy Keep product data synced: availability, pricing, variants, and descriptions. Outdated info will kill your ranking in AI. AI models favor reviews that mention specific use cases, emotions, and product outcomes. A single thoughtful review like “Perfect for marathon runners with flat feet” now outranks 50 vague 5-star ratings. I’m excited for this AI eCommerce era. More to come from The Other Group #ai #ecommerce #commerce

Pinterest’s UniPinRec and the benefits of unified retrieval and ranking "In May, Pinterest released a paper introducing UniPinRec, a system that unifies candidate retrieval and ranking for Pins within a single model while preserving the traditional multi-stage recommendation funnel. As such, UniPinRec is something of a hybrid approach between Meta’s Hierarchical Sequential Transduction Unit (HSTU) and Kuaishou’s OneRec. HSTU provides a common architectural paradigm for retrieval and ranking but doesn’t fully unify them; OneRec is a fully combined, end-to-end generative architecture that uses semantic IDs for exact item prediction. UniPinRec unifies retrieval and ranking by adopting one data input format for both tasks, but instead of predicting an exact item, it generates “query vectors” that are then used to fetch candidates with ANN search." https://lnkd.in/d3sYaXnH

𝑯𝒂𝒗𝒆 𝒚𝒐𝒖 𝒆𝒗𝒆𝒓 𝒇𝒆𝒍𝒕 𝒕𝒉𝒂𝒕 𝒆𝒗𝒂𝒍𝒖𝒂𝒕𝒊𝒐𝒏 𝒓𝒆𝒄𝒐𝒎𝒎𝒆𝒏𝒅𝒂𝒕𝒊𝒐𝒏𝒔, 𝒅𝒆𝒔𝒑𝒊𝒕𝒆 𝒎𝒐𝒏𝒕𝒉𝒔 𝒐𝒇 𝒘𝒐𝒓𝒌, 𝒆𝒏𝒅 𝒖𝒑 𝒔𝒊𝒕𝒕𝒊𝒏𝒈 𝒐𝒏 𝒂 𝒔𝒉𝒆𝒍𝒇, 𝒖𝒏𝒖𝒔𝒆𝒅 𝒂𝒏𝒅 𝒇𝒐𝒓𝒈𝒐𝒕𝒕𝒆𝒏? You’re not alone. One of the biggest challenges in #Monitoring, #Evaluation, #Accountability, and #Learning (MEAL), research is not producing #evidence, but ensuring that evidence actually informs #decisions. Why recommendations often fail? Many evaluation recommendations are not used because they are: 1. Too broad or vague 2. Not aligned with decision-makers’ priorities 3. Disconnected from available resources and timelines 4. Developed without sufficient stakeholder ownership This is a well-documented challenge in evaluation systems, including government-led evaluations. This short brief offers a concise and practical guideline on how to develop actionable evaluation recommendations that are: 1. Clearly linked to evidence and findings 2. Feasible, specific, and implementation-oriented 3. Co-developed with stakeholders to strengthen ownership 4. Designed to feed directly into improvement plans and learning processes The document also provides a checklist for assessing recommendation quality. If you work in M&E, #research, #policy, or programmes and want your evaluations to actually influence decisions, this brief is for you. Save this post for later and share it with colleagues who may find this brief insightful. #MonitoringAndEvaluation #MEAL #MEL #EvaluationUse #EvidenceBasedDecisionMaking #LearningAndAccountability #EvaluationRecommendations #ImpactMeasurement #DataForDevelopment