How to Analyze User Behavior With Data

One of the biggest challenges in UX research is understanding what users truly value. People often say one thing but behave differently when faced with actual choices. Conjoint analysis helps bridge this gap by analyzing how users make trade-offs between different features, enabling UX teams to prioritize effectively. Unlike direct surveys, conjoint analysis presents users with realistic product combinations, capturing their genuine decision-making patterns. When paired with advanced statistical and machine learning methods, this approach becomes even more powerful and predictive. Choice-based models like Hierarchical Bayes estimation reveal individual-level preferences, allowing tailored UX improvements for diverse user groups. Latent Class Analysis further segments users into distinct preference categories, helping design experiences that resonate with each segment. Advanced regression methods enhance accuracy in predicting user behavior. Mixed Logit Models recognize that different users value features uniquely, while Nested Logit Models address hierarchical decision-making, such as choosing a subscription tier before specific features. Machine learning techniques offer additional insights. Random Forests uncover hidden relationships between features - like those that matter only in combination - while Support Vector Machines classify users precisely, enabling targeted UX personalization. Bayesian approaches manage the inherent uncertainty in user choices. Bayesian Networks visually represent interconnected preferences, and Markov Chain Monte Carlo methods handle complexity, delivering more reliable forecasts. Finally, simulation techniques like Monte Carlo analysis allow UX teams to anticipate user responses to product changes or pricing strategies, reducing risk. Bootstrapping further strengthens findings by testing the stability of insights across multiple simulations. By leveraging these advanced conjoint analysis techniques, UX researchers can deeply understand user preferences and create experiences that align precisely with how users think and behave.

🚨The greatest drop-off is from Product Details Page To Cart Page, so we must improve our Product Details Page! Not so fast ✋ In today's age of data obsession, almost every company has an analytics infrastructure that pumps out a tonne of numbers. But rarely do teams invest time, discipline & curiosity to interpret numbers meaningfully. I will illustrate with an example. Let's take a simple e-commerce funnel. Home Page ~ 100 users List Page ~ 90 users Product Display Page ~ 70 users Cart Page ~ 20 users Address Page ~ 15 users Payments Page ~12 users Order Confirmation Page ~ 9 users A team that just "looks" at data will immediately conclude that the drop-off is most steep between Product Details Page & Cart Page. As a consequence they will start putting in a lot of fire power into solving user problems on Product Display Page. But if the team were data "curious", would frame hypothesis such as "do certain types of users reach cart page more effectively than others?" and go on to look at users by purchase buckets, geography, category etc and look at the entire funnel end to end to observe patterns. In the above scenario, it's likely that the 20 cart users were power users whilst new & early purchasers don't make it to this stage. The reason could be poor recommendations on the list page or customers are only visiting the product display page to see a larger close up of the product. So how should one go about looking at data ? Do ✅ Start with an open & curious mind ✅ Start with hypothesis ✅ Identify metrics & counter metrics that will help prove/disprove hypothesis ✅ Identify the various dimensions that could influence behaviours - user type, geography, category, device type, gender, price point, day, time etc. The dimensions will be specific to your line of business. ✅ Check for data quality and consistency ✅ Look at upstream and downstream behaviour to see how the behaviour is influenced upstream and what happens to the behaviour downstream. ✅ Check for historical evidence of causality Dont ❌ Look at data to satisfy your bias ❌ Rush to conclude your interpretation ❌ Look at data in isolation - - - TLDR - Be curious. Not confirmed. #metrics #analytics #productmanagement #productmanager #productcraft #deepdiveswithdsk

Crowning a New Term: “Iceberg Metrics” 🧊 ✨ I’m calling it: Iceberg Metrics represent KPIs that only reveal the tip of what’s really happening below the surface. Metrics like abandoned carts seem simple but often mask much more—checkout friction, hidden costs, trust issues, and more. To truly understand and optimize, we need to dig deeper. Here’s how to dive into the “iceberg” of abandoned cart rates: 1. Establish Baseline Metrics: Start by gathering data on current abandoned cart rates, session times, and bounce rates using heat maps and session recordings to see where users drop off. 2. Segment the Audience: Analyze users by behavior (first-time vs. repeat visitors, mobile vs. desktop) and traffic source (organic, paid, email). 3. Experiment Hypotheses: Develop hypotheses for abandonment reasons—shipping costs, checkout friction, distractions, or lack of trust signals—and test them. 4. Run A/B Tests: Test variations like simplifying the checkout process, showing shipping costs earlier, adding trust badges, or retargeting abandoned cart emails. 5. Use Heat Maps & Session Recordings: Examine user behavior in real time. Look for confusion or hesitation, where users hover, and whether they engage with key information. 6. Contextualize Results: Analyze how changes impact overall user flow. Did simplifying checkout help, or did other metrics like bounce rate increase? 7. Ecosystem Approach: Examine how tweaks affect the full journey—from product discovery to checkout—balancing short-term improvements with long-term goals like lifetime value. 8. Iterate: Refine solutions based on experiment findings and continuously optimize the customer journey. This one’s mine, folks! #IcebergMetrics #OwnIt #DataDriven #EcommerceOptimization #NewMetricAlert Cheers, Your cross-legged CAC and CLV buddy 🤗

Most teams are just wasting their time watching session replays. Why? Because not all session replays are equally valuable, and many don’t uncover the real insights you need. After 15 years of experience, here’s how to find insights that can transform your product: — 𝗛𝗼𝘄 𝘁𝗼 𝗘𝘅𝘁𝗿𝗮𝗰𝘁 𝗥𝗲𝗮𝗹 𝗜𝗻𝘀𝗶𝗴𝗵𝘁𝘀 𝗳𝗿𝗼𝗺 𝗦𝗲𝘀𝘀𝗶𝗼𝗻 𝗥𝗲𝗽𝗹𝗮𝘆𝘀 𝗧𝗵𝗲 𝗗𝗶𝗹𝗲𝗺𝗺𝗮: Too many teams pick random sessions, watch them from start to finish, and hope for meaningful insights. It’s like searching for a needle in a haystack. The fix? Start with trigger moments — specific user behaviors that reveal critical insights. ➔ The last session before a user churns. ➔ The journey that ended in a support ticket. ➔ The user who refreshed the page multiple times in frustration. Select five sessions with these triggers using powerful tools like @LogRocket. Focusing on a few key sessions will reveal patterns without overwhelming you with data. — 𝗧𝗵𝗲 𝗧𝗵𝗿𝗲𝗲-𝗣𝗮𝘀𝘀 𝗧𝗲𝗰𝗵𝗻𝗶𝗾𝘂𝗲 Think of it like peeling back layers: each pass reveals more details. 𝗣𝗮𝘀𝘀 𝟭: Watch at double speed to capture the overall flow of the session. ➔ Identify key moments based on time spent and notable actions. ➔ Bookmark moments to explore in the next passes. 𝗣𝗮𝘀𝘀 𝟮: Slow down to normal speed, focusing on cursor movement and pauses. ➔ Observe cursor behavior for signs of hesitation or confusion. ➔ Watch for pauses or retracing steps as indicators of friction. 𝗣𝗮𝘀𝘀 𝟯: Zoom in on the bookmarked moments at half speed. ➔ Catch subtle signals of frustration, like extended hovering or near-miss clicks. ➔ These small moments often hold the key to understanding user pain points. — 𝗧𝗵𝗲 𝗤𝘂𝗮𝗻𝘁𝗶𝘁𝗮𝘁𝗶𝘃𝗲 + 𝗤𝘂𝗮𝗹𝗶𝘁𝗮𝘁𝗶𝘃𝗲 𝗙𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸 Metrics show the “what,” session replays help explain the “why.” 𝗦𝘁𝗲𝗽 𝟭: 𝗦𝘁𝗮𝗿𝘁 𝘄𝗶𝘁𝗵 𝗗𝗮𝘁𝗮 Gather essential metrics before diving into sessions. ➔ Focus on conversion rates, time on page, bounce rates, and support ticket volume. ➔ Look for spikes, unusual trends, or issues tied to specific devices. 𝗦𝘁𝗲𝗽 𝟮: 𝗖𝗿𝗲𝗮𝘁𝗲 𝗪𝗮𝘁𝗰𝗵 𝗟𝗶𝘀𝘁𝘀 𝗳𝗿𝗼𝗺 𝗗𝗮𝘁𝗮 Organize sessions based on success and failure metrics: ➔ 𝗦𝘂𝗰𝗰𝗲𝘀𝘀 𝗖𝗮𝘀𝗲𝘀: Top 10% of conversions, fastest completions, smoothest navigation. ➔ 𝗙𝗮𝗶𝗹𝘂𝗿𝗲 𝗖𝗮𝘀𝗲𝘀: Bottom 10% of conversions, abandonment points, error encounters. — 𝗕𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗮 𝗖𝗼𝗻𝘀𝗶𝘀𝘁𝗲𝗻𝘁 𝗦𝗲𝘀𝘀𝗶𝗼𝗻 𝗥𝗲𝗽𝗹𝗮𝘆 𝗣𝗿𝗮𝗰𝘁𝗶𝗰𝗲 Make session replays a regular part of your team’s workflow and follow these principles: ➔ Focus on one critical flow at first, then expand. ➔ Keep it routine. Fifteen minutes of focused sessions beats hours of unfocused watching. ➔ Keep rotating the responsibiliy and document everything. — Want to go deeper and get more out of your session replays without wasting time? Check the link in the comments!

Your users leave a trail of behavioral breadcrumbs with every transaction, and your recommendation engine might be stepping right over them. A new study by Upwork analyzed 9M marketplace users across 62M interactions and found that combining text-based profile analysis with behavioral data improved matching accuracy by 8-12% compared to text-only approaches. The system learns simultaneously from what users write about themselves and how they actually behave on the platform. Who they hire, what they buy, which connections succeed. This architecture works anywhere you're connecting two sides of a market. - Airbnb matching guests to hosts. - Amazon connecting buyers to sellers. - Uber pairing riders with drivers. - Dating apps. - B2B sales platforms. The pattern is the same. You have profiles (text people write about themselves), and you have behavior (the trail of interactions in your database). Most recommendation systems use one or the other. Combining both produces substantially better matches. If you run a two-sided marketplace, your transaction and interaction logs are an underutilized asset. The patterns of who your users connect with contain a real signal about who you should connect them with next.

Ever wondered how Netflix knows you binge-watched an entire season in one sitting, or how Google Analytics separates your morning browsing from your evening research - that's sessionization at work. Sessionization is one of those fundamental data engineering problems that sounds simple but reveals layers of complexity when you dig in. 𝙒𝙝𝙖𝙩 𝙞𝙨 𝙎𝙚𝙨𝙨𝙞𝙤𝙣𝙞𝙯𝙖𝙩𝙞𝙤𝙣? At its core, sessionization is about grouping sequential user events into meaningful "sessions" based on activity patterns. The standard rule: if a user goes idle for more than 30 minutes, we consider that session ended. Simple concept, but with a tricky implementation. 𝙒𝙝𝙮 𝙄𝙩 𝙈𝙖𝙩𝙩𝙚𝙧𝙨 Product teams need sessions to understand: → How long users actually engage with your platform  → What features are used together in a single visit  → Where users drop off in their journey  → Peak usage patterns throughout the day Without proper sessionization, you're just looking at disconnected events. With it, you see user behavior as a story. 𝙏𝙝𝙚 𝙎𝙌𝙇 𝘾𝙝𝙖𝙡𝙡𝙚𝙣𝙜𝙚 One of the problems in today's 50-Day Data Challenge asked: given a table of user events with timestamps, assign session numbers where any gap > 30 minutes starts a new session. Let's solve this using SQL (this is pretty straightforward in Flink using the Session Window class: https://lnkd.in/eZYACdnG): 1. Look Backward: Use LAG() to grab each event's previous timestamp for the same user. You can't identify gaps without knowing what came before. 2. Identify Boundaries: Flag rows where either there's no previous event (first-time user) OR the time gap exceeds 30 minutes. These flags mark session boundaries. 3. Propagate State Forward: Use a running SUM() of those boundary flags. Each time you hit a boundary (flag = 1), the cumulative sum increments, creating a new session ID that carries forward to all subsequent events. This transforms: [0, 1, 0, 0, 1, 0] into session IDs: [1, 2, 2, 2, 3, 3] 4. Aggregate: Now that every event knows its session, GROUP BY to calculate session start, end, duration, and event count. 𝙏𝙝𝙚 𝙍𝙚𝙖𝙡 𝙇𝙚𝙖𝙧𝙣𝙞𝙣𝙜 This pattern is a blueprint for solving "groups with gaps" problems: → Detecting streaks (consecutive days of activity)  → Identifying downtime periods in system logs  → Finding engagement patterns in subscription products  → Measuring time-to-resolution in support tickets The technique remains the same: compare to previous → flag boundaries → propagate state → aggregate groups. Week 2 of Zach's 50-Day Data Challenge: SQL 🔥 Moved from data modeling last week to SQL this week. Today was all about sessionization - one of those problems that looks straightforward until you actually build it. Day 8/50 ✅ #DataEngineering #50DayDataChallenge #SQL #Analytics #Sessionization

𝗔𝗿𝗲 𝘆𝗼𝘂 𝗮𝗻𝗮𝗹𝘆𝘇𝗶𝗻𝗴 𝗱𝗮𝘁𝗮… 𝗼𝗿 𝗷𝘂𝘀𝘁 𝘀𝘂𝗺𝗺𝗮𝗿𝗶𝘇𝗶𝗻𝗴 𝗶𝘁? Dashboards stop at “𝘄𝗵𝗮𝘁 𝗵𝗮𝗽𝗽𝗲𝗻𝗲𝗱.” Real analytics goes further - it explains 𝘄𝗵𝘆, 𝘄𝗵𝗲𝗿𝗲, 𝗮𝗻𝗱 𝘄𝗵𝗮𝘁 𝘁𝗼 𝗱𝗼 𝗻𝗲𝘅𝘁. This carousel breaks down the 𝟭𝟬 𝗰𝗼𝗿𝗲 𝘄𝗮𝘆𝘀 𝗱𝗮𝘁𝗮 𝗶𝘀 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝗮𝗻𝗮𝗹𝘆𝘇𝗲𝗱 𝗶𝗻 𝗿𝗲𝗮𝗹 𝗯𝘂𝘀𝗶𝗻𝗲𝘀𝘀𝗲𝘀, not just taught in courses. It shows how different questions demand different approaches: • When to use descriptive vs trend analysis • How comparisons drive real decisions • Why segmentation and cohorts explain user behavior • Where funnels reveal revenue leaks • How root cause analysis moves teams from guessing to fixing • Why anomalies matter before problems explode • How predictive analysis helps teams plan ahead The difference between a report-maker and a real analyst is simple: One summarizes numbers. The other chooses the right analytical lens. If you’re learning data analytics, preparing for interviews, or working with stakeholders - this is the mental model that actually matters. Save it. Use it. And next time, analyze - don’t just summarize.

(3/3) The most useful research I run isn't a study. It's a question bolted onto someone else's A/B test... Your product team runs experiments constantly. Almost none of that behavioral data gets paired with a single word about why, so every test answers "what happened" and nearly none of them answer "what were people actually thinking." That gap is the cheapest, most overlooked research most teams have sitting right in front of them. The move is simple. When a variant goes live, you target a short intercept survey at the users inside that variant. Behavioral data from the experiment, attitudinal data from the survey, same users, same moment. When the test wins, you know what changed in their heads. When it loses, you still walk away with the why. The part I value most: it preserves the learning even when an experiment gets killed early. A test shut off on day three normally teaches you nothing. Pair it with a survey and even a dead experiment leaves you smarter than you were. This is exactly what we built experiment-paired surveys for at Sprig. You point a study at a specific variant, by URL or by a list of user IDs, and capture the attitude alongside the behavior without standing up a full research cycle. And the survey isn't a static form. It can ask an AI-driven follow-up based on what someone just told you, so when a user says the new layout "felt cluttered," the next question digs into which part, in their own words. You get the reason, not just the checkbox. It's the third column of the score card, made practical. A score tells you the room got hotter. Behavior tells you where the heat is. Attitude tells you why someone lit the match. You want all three, and you can have them on the experiments you're already running. That's the series: stop letting a single number have the last word. Read the signals underneath it. Curious which metric your team over-trusts the most?

Descriptive. Diagnostic. Predictive. Prescriptive. I heard these words so many times when I started learning data analytics. Everyone explained them in theory. I nodded along. But honestly? I didn’t feel them. They just felt like complicated labels people threw around. But then I started doing small projects and suddenly, they made sense. Not in slides or definitions. But in real, messy, confusing datasets. 📌 Descriptive: "What happened?" I opened my dataset and asked: ▪️"How many people visited the site last month?" ▪️ "What was the average time spent?" ▪️ "How many people bought something?" Just basic numbers. No story yet. Just what happened. And weirdly enough that’s where everything begins. A simple count. A line chart. A trend. 📌Diagnostic: "Why did that happen?" ▪️Okay, cool. Sales dropped in May. But why? ▪️ Was it fewer users? ▪️ Was it only Android users who dropped? ▪️ Was it something we changed on the product page? ▪️ This part was tough. I had to slice, filter, go back and forth. ▪️ It felt like solving a puzzle. Frustrating, but satisfying. 📌Predictive: "What might happen next?" ▪️This one scared me. ▪️Because now I had to look ahead. ▪️Not with 100% accuracy. ▪️But with logic. With patterns. With the past. ✏️ "If this drop continues, will July be even worse?" ✏️"If users are clicking more on videos, can we expect higher engagement next month?" 📌Prescriptive: "What should we do now?" This is the part where everyone looks at you and goes: So... what’s your recommendation? This is where your analysis turns into action. That’s when it all clicked for me. These 4 aren’t just fancy types. They’re how your brain actually works as a data analyst. Start with: ▪️What happened? ▪️Why did it happen? ▪️What might happen next? ▪️What should we do? ♻️ Repost : If you found this helpful, to reach others who might need it. ✳️ Follow Mariya Joseph for more daily content!

𝐋𝐞𝐭’𝐬 𝐬𝐨𝐥𝐯𝐞 𝐚 𝐁𝐮𝐬𝐢𝐧𝐞𝐬𝐬 𝐂𝐚𝐬𝐞 𝐏𝐫𝐨𝐛𝐥𝐞𝐦 𝐭𝐨𝐠𝐞𝐭𝐡𝐞𝐫, If you're preparing for Data or Product Analyst roles — this is exactly the type of case round you should practice. It’s not about jumping into queries — it’s about structured thinking. 𝐒𝐜𝐞𝐧𝐚𝐫𝐢𝐨: You're a Data Analyst at a food delivery company like Zomato or Swiggy. In the past 15 days, there’s been a 5% drop in active customers. You’re asked: “What could be the reason behind this churn, and how would you investigate it?” 𝐒𝐭𝐞𝐩 𝟏: Clarify the Problem Before solving, ask: Does “churn” mean no orders? Or no activity at all? Is it across all users or specific cohorts (new users, Prime, etc.)? Any specific regions more impacted? These questions help you define the problem — not just guess a solution. 𝐒𝐭𝐞𝐩 𝟐: Structure Your Investigation Break down your thinking into: 🔹 Internal Factors (platform-level issues) App crashes, login issues → Check crash logs, screen exits Delivery delays → Compare SLA metrics over time Key restaurant unavailability → Partner downtime, stockouts Reduction in discounts → Drop in coupon usage or redemptions Checkout issues → Cart-to-payment funnel drop-offs 🔹 External Factors (outside control) Weather/strikes/curfews → Regional impact data Seasonality → Historical trends from previous years Competitor activity → Market-level discounts or ad campaigns 𝐒𝐭𝐞𝐩 𝟑: Go Deep on the Root Cause Let’s say the team confirms: “Yes, we reduced discount campaigns.” Now prove it with data: Analyze sessions reaching the “Apply Coupon” page Compare order completion rate before & after discount application Study cart abandonment after coupon screen Look at this metric over last 15 days vs previous months This validates the impact of discounts on churn — using real funnel data. 𝐓𝐡𝐞 𝐫𝐞𝐚𝐥 𝐭𝐚𝐤𝐞𝐚𝐰𝐚𝐲? Case rounds like this aren’t about correct answers. They’re about how you think, how you structure messy problems, and how well you connect business context with data. This is the exact type of round I’ve seen in companies like Zomato, Blinkit, Flipkart, Meesho, etc. So if you’re preparing — don’t stop at SQL or dashboards. Practice thinking like a business analyst. If you want more real case problems like this — drop a “Case” in comments. I’ll share a few more from my interview experience.