Analyzing Ecommerce Website Traffic

Killer graph. Out of the £130 billion online non-food purchases we make in the UK, £27 billion of them get sent back to retailers. Our research with ZigZag Global shines a spotlight on the significant challenge online returns cause in the industry, focusing on those consumers who consistently and intentionally over-order - the "serial returners". Key stats ➡️ Around 11% of online shoppers are serial returners (frequently over-ordering with the intention of returning many items) ➡️They account for 24% of all online returns ➡️Serial returners send back, on average, £1,400 worth of online orders per year, compared with an average of £650. ➡️ This amounts to £6.6 billion of returns. ➡️ Almost three-quarters of serial returners are under the age of 45, and they return more than 42% of all their orders. A 1/4 of serial returners admit to over-ordering just to reach a minimum order value (often to trigger free delivery) only to return goods they had no intention of keeping. The same proportion also said they had returned items after finding them cheaper elsewhere or on promotions. While 18% admitted to returning items having already used them for a short period. There is no silver bullet here that is going to fix this issue for retailers. A nuanced understanding of specific triggers and barriers is essential to effectively target returners through pricing and returns options. 💥 For many boardrooms debating whether they should charge for returns, my thoughts are: 💥 The returns equation transcends simple binary choices between free or paid. Retailers must architect differentiated returns propositions that align commercial realities with customer lifetime value. Smart retailers will segment their returns strategy by customer profitability metrics, leveraging AI to identify purchase patterns that predict long-term value. This enables dynamic returns pricing that protects margins while fostering relationships with truly valuable customers. The goal isn't to punish returns – it's to price them according to their true cost to serve, while rewarding profitable shopping behaviours. There's also a paradox at play where customer acquisition costs are optimised but customer profitability is compromised. Many retailers are essentially subsidising unsustainable shopping behaviours at the expense of margin, unknowingly targeting customers they could do without. The real opportunity lies in leveraging returns data as a predictive indicator of customer profitability. By applying advanced analytics to returns patterns, seasonal purchasing behaviours, and cross-category browsing and mining deep behaviour insights, retailers can enable proactive intervention before profitability erodes. This shifts the conversation from universal policies to personalised solutions that can turn returns from a pure cost centre into a strategic lever for customer engagement and loyalty. Full research is available to download here ⬇️ https://lnkd.in/e5paRNWC

Your buyers decided on their phone. Your website was signed off on a 27-inch monitor. This is one of the most expensive blind spots I see, and almost nobody clocks it. Think about how a website gets approved. Designers build it on big, beautiful monitors. Stakeholders review it on their laptops, on fast office wifi. The executive gives the final nod on a large screen, in a meeting, with the site looking immaculate. Everyone in that chain signed off on a version of the site most of your buyers will never see. Because your buyers are on a phone. On mobile data. One-handed, half-distracted, on a train or between meetings. That is where the cracks show. The hero that looks stunning on desktop pushes the call to action below the fold. Tap targets sit too close together. Text reflows into a wall. The page that loaded instantly on office wifi crawls on 4G. The form that was easy with a mouse becomes a battle with a thumb. None of it shows up in the boardroom, because nobody in the boardroom is using the site the way a buyer does. Here is the uncomfortable part. For most businesses, a large share of the traffic, and often the conversions, happen on mobile. So the experience that actually drives the revenue is the one version nobody reviewed. You did not optimise your website. You optimised the version your team looks at, and ignored the one your buyers live in. When did you last go through your own site on your phone, on mobile data, the way a buyer actually would?

Meta Ads Analytics: Measuring What Matters Navigating the Meta Ads Analytics Dashboard Get acquainted with the Meta Ads Analytics dashboard. Focus on key areas such as the overview tab, offering a snapshot of your campaign performance, and the detailed breakdowns providing insights into specific metrics. Tailor your analytics view to align with your campaign goals. Whether you're tracking conversions, engagement, or reach, make sure to monitor the metrics most relevant to your objectives. Key Metrics to Monitor These encompass likes, comments, shares, and video views. High engagement rates typically indicate that your content resonates well with your audience. Click-Through Rate (CTR), Cost Per Click (CPC), and Conversion Rate are crucial for understanding how effectively your ads drive action. Return on Ad Spend (ROAS), a vital metric for measuring profitability, informs you about the return generated for every dollar spent. Audience Insights Examine the age, gender, location, and other demographic data of the individuals interacting with your ads. This information aids in tailoring future campaigns to better target your audience. Understanding how different segments interact with your ads, such as the time of day or device used, can optimize ad delivery for maximum impact. A/B Testing Results Utilize Meta Ads Analytics to scrutinize the results of A/B tests. These insights guide you on which creative elements, ad placements, and audience segments work best. Based on A/B testing data, make informed adjustments to continuously enhance your campaigns' performance. Conversion Tracking Ensure you've set up conversion tracking to measure the actions users take after clicking on your ads. Understanding the path that leads to conversions provides insights into the customer journey and identifies the most effective ad elements in driving sales. Leveraging the Data for Campaign Adjustments Apply the insights gained from Meta Ads Analytics to make strategic decisions. This may involve shifting budget allocations, adjusting target audiences, or tweaking ad creatives. Regularly check your analytics to stay on top of campaign performance. Ongoing monitoring allows for timely adjustments to optimize your campaigns. Reporting and Strategy Development Develop comprehensive reports based on analytics data to share with your team or clients. Use the trends and patterns identified in your analytics to inform your broader advertising strategy. This can include budget planning, seasonal adjustments, and long-term targeting strategies. #facebookads #MetaAds #Analytics #DigitalAdvertising #digitalmarketing #CampaignOptimization #DataAnalysis #AudienceInsights #ABTesting #ConversionTracking #StrategicDecisions #ROI #SocialMediaMarketing #MarketingStrategy #AdvertisingPerformance

In e-commerce, organic recommendations are designed to maximize relevance, while ads are designed to generate revenue. When these systems operate independently, placement is often dictated by business rules rather than the actual value each item brings to a shopper. In this tech blog, engineers at Flipkart describe how they re-architected their “Similar Products” widget to integrate sponsored and organic products into a unified ranking system. Instead of reserving fixed positions for ads and recommendations, they built a framework where both compete for placement using a common ranking approach. The team developed a unified ranking model that explicitly accounts for the different data distributions of sponsored (i.e. ads) and organic products, allowing the same behavioral signal to be interpreted differently depending on context. They also introduced a score normalization layer that maps outputs from separate ranking services onto a common scale, enabling true merit-based competition for every slot. Recognizing that production systems are only as strong as their weakest dependency, they further added safeguards to prevent service failures from unintentionally flooding the experience with ads. The result was a system that allocates each position based on predicted value rather than predefined rules, leading to a 1.3% increase in organic orders and a 3.4% increase in ad revenue. What I like about this work is that it approaches the problem from a user experience perspective rather than treating sponsored content and organic recommendations as separate optimization targets. Users do not care whether a surfaced product is an ad or an organic recommendation—they care whether it is relevant. By creating a framework where different types of content compete fairly for user attention, the team was able to improve both engagement and monetization simultaneously, which is often a sign that the system is making better decisions overall. #DataScience #MachineLearning #AI #Analytics #RecommendationSystem #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/eMcmGMDk

Can LLMs improve product recommendations with re-ranking? Fascinating new paper from Meta on applying LLMs to recommendation systems. The domain discussed in the paper is content re-ranking, but I don't see why this couldn't be applied to ads. Re-ranking takes a ranked list of candidate items (following retrieval and, sometimes, pre-ranking) and updates the ordering to better optimize some objective function (eg., purchase). The authors describe how they utilize an LLM to re-rank candidates with a number of novel innovations: - Instead of building the LLM vocabulary from item embeddings, which would likely be too large to be useful, they decompose each item embedding into a sequence of "tokens" produced from a K-stage quantization process (with RQ-VAE). This process accepts the item embedding at k=1, calculates the residual vector from the nearest of C learned centroids for that step (called "codebooks"), and passes that output to k=2, and so on to K. This produces a Semantic ID (SID) of length K. - Because re-ranking must be done quickly, it requires a smaller LLM (the paper uses 8B parameters). So in training, they prompt a large model (Qwen-32B) with the user's history, the candidate items produced in ranking, the SIDs, and instructions to reason through its process of re-ranking these items. That model produces a reasoning trace and a re-ranked list. The authors use rejection sampling to retain only the outputs (reasoning traces + rankings) for which the ground-truth item is ranked sufficiently high. The 8B student model is fine-tuned via SFT on that distribution, learning P( reasoning trace + ranking | prompt ). - Finally, the authors fine-tune this model with RL on the outcome, using the ground truth's location in the list as the reward. This aligns the model's policy with the reward, enabling more thorough comparison across candidates (versus reasoning collapse). The authors make the point that LLMs can introduce additional product context, scalability, and "world knowledge" to RecSys, turning ranking into a structured reasoning task rather than a pure scoring task. The paper is quite dense but worth reading in full; link below.

Revolutionary Breakthrough in Recommendation System Efficiency: Request-Only Optimization (ROO) Researchers at Meta have introduced a game-changing approach that's reshaping how billion-user-scale recommendation systems handle training and inference. The Request-Only Optimization paradigm addresses a fundamental inefficiency that's been plaguing the industry for years. The Core Problem: Traditional recommendation systems suffer from massive feature duplication. When a user makes a request, multiple content items are served as impressions, but each impression duplicates identical user features across separate training samples. For systems generating 4-7 impressions per request, this creates enormous computational waste. The ROO Solution: Instead of treating each impression as a separate training unit, ROO captures all impressions from the same request in a single training sample. This elegant redesign separates Request-Only (RO) features from Non-Request-Only (NRO) features at the data source level. Technical Architecture: - Data Schema Redesign: ROO uses request-level joiners that buffer user-item interactions by unique request IDs, creating compact training examples with one copy of user features (RO part) and arrays of item features (NRO part) - Computation Deduplication: User-side embedding lookups and all-to-all communications are reduced from O(B_NRO) to O(B_RO) complexity - Unified Training-Inference Pipeline: The same data format works across both training and inference stacks, eliminating complex server-client optimizations Under the Hood Innovations: The system leverages TorchRec's variable-length batch sharding to handle different batch sizes between RO and NRO tensors. Advanced architectures like Hierarchical Sequential Transduction Units (HSTU) and UserArch can now scale dramatically because their computational costs are amortized across multiple impressions. Production Impact: - Training sample volume increased by 43-150% with same storage - Training throughput improvements: 220-570% for retrieval models, 125-266% for early-stage ranking - Model FLOPs scaled 7x with same compute budget - Significant offline and online metric improvements across multiple billion-user products

95% recommendation hit rate. 15% lower long-term retention. We called it the Accuracy Trap. And it almost broke the recommender system of an e-commerce startup I advised. Here's what happened, and the two metrics that saved us. A user clicks on a navy blue t-shirt. The model generates six more navy blue t-shirts. Hit rate? Near perfect. User experience? An echo chamber. Most recommender systems don't fail because the models are weak. They fail because they become boring prediction machines. Traditionally, engineers optimize for: Precision, Hit Rate, and NDCG. The SOTA systems also optimize for: Discovery, Long-term LTV, and Serendipity. The difference? User intent is exploratory, not just predictive. If your model only learns historical similarity, it slowly collapses. The good news: "Serendipity" is measurable. Two metrics I consider non-negotiable in production: ① Intra-List Diversity (ILD): Average cosine distance between recommended items. ILD → 0 means your feed is redundant. ② Novelty: Negative log probability of item popularity. Are you actually surfacing niche preferences, or just lazily recommending global bestsellers to everyone? One more layer: GenAI explanations. → If you're generating personalized recommendation explanations with LLMs, BLEU and ROUGE are no longer enough. → Semantic similarity matters more than token overlap. That's why BERTScore often outperforms both. The hardest lesson I learned building ML systems at Amazon and Twitter: * Your evaluation framework IS your product strategy * Models aggressively optimize whatever objective you give them - even when it actively hurts users. ───────────────────── This wraps up my Generative RecSys series. Past posts: → Semantic IDs: https://lnkd.in/g75sgtxd → Vision RecSys: https://lnkd.in/gg5efmzf → LLM Enhanced RecSys: https://lnkd.in/gBPUMSZa → Encode and Generate Paradigm: https://lnkd.in/g9MmWZaN → RecSys Context Management: https://lnkd.in/gs3nSKSc Follow along as I have interesting deep dives coming up. ───────────────────── Curious: What's one metric your team uses to keep your recommender from becoming a boring prediction machine?

How we've leveraged insights from LinkedIn Ads data to narrow in on our ICP When we first launched Impactable's LinkedIn ad service, our approach was "the broader, the better." Like most early startups, we believed that anyone with a business online was a potential client. Over time, the demographic data told a different story. Targeting everyone diluted our impact. Analyzing click-through rates, conversions, and client intake forms began to paint a clearer picture of who was truly benefiting from our services. This insight was a game-changer. We began to pivot, concentrating on sectors where we saw the most traction and intentionally stepping back from markets that, while initially appealing, didn't align with our strengths as we scaled. This strategic shift wasn't just about cutting out less profitable sectors; it was about doubling down on where we could make the most significant difference. By niching down based on data insights, we could tailor our services, hone our expertise, and ultimately, deliver more value to our clients. The shift wasn't just about who we chose to serve but about becoming the best at serving them. #DataDrivenDecisions #MarketingStrategy #LinkedInAds #NicheMarketing

The returns data domino effect is real inside of brands. Many brands attribute high return rates purely to customers ordering multiple sizes. The reality is more complex. By diving deep into returns data, you uncover insights that will change how you think about returns. And what strategies you implement to reduce them. Common trends include: 👉 SKU-Level Reporting: Identifying products with excessive return rates can reveal issues with sizing, quality, or design. For instance, if a specific dress size is returned 65% of the time, it's a clear signal to revisit sizing charts or quality control processes. 👉 Stock Optimization: With returns taking up to 21 days to process, especially for international orders, having visibility into "stock in transit" is crucial. This data can help prevent unnecessary order cancellations and improve inventory management. 👉 Customer Behavior Analysis: By analyzing return patterns, brands can distinguish between serial returners and genuine product issues, allowing for more targeted solutions. 👉 Accurate Financial Reporting: Layering returns data over sales data provides a true picture of margins and profitability, essential for making informed marketing and business decisions. 👉 Cost Calculation: Understanding the average cost of returns, including courier fees and processing time, is vital for developing effective return policies and pricing strategies. We go deep on how to use data to analyse your return data in part 4 of our Paid Returns series in the Commerce Thinking newsletter. Read and subscribe below: https://lnkd.in/etvrkvWY

Stop Wasting Your Google Ads Budget on "Unknown" Demographics & How Advanced Bid Adjustments Can Skyrocket Your ROI! I've seen too many advertisers bleed money on vague targeting. Enter advanced bid adjustments, the unsung hero of precision advertising. The game changer in Google Ads: Demographics like age, gender, and income often include an "unknown" bucket: users Google can't categorize. Bidding blindly here? You're essentially gambling on conversions. By leveraging advanced bid adjustments: - Boost bids on high-performing demographics: Crank up bids by 20-50% for proven converters (e.g., 25-34-year-olds in your niche) to dominate auctions. - Slash or exclude "unknowns": Set negative adjustments (-100% effectively excludes) to avoid funneling ad spend into black holes where data is scarce. - Layer with audience signals: Combine with RLSA or custom audiences for hyper-targeted efficiency, often yielding 30-50% higher CTR and lower CPC. Real-world win: One client slashed CPA by 42% in Q4 by ditching unknowns and fine-tuning bids, all while scaling spend 2x. Pro tip: Always monitor performance reports weekly. Unknowns can represent 20-40% of traffic; ignoring them is like leaving cash on the table. What's your biggest Google Ads pain point? Drop it below, let's optimize together! Follow Sushil Dahiya for more marketing insights and actionable tips! #GoogleAds #PPCStrategy #DigitalMarketing #BidOptimization