User Experience Metrics for Success

⏳ Designing Better Loading and Progress Indicators UX. Practical UX guidelines to reduce the impact of waiting and choose the right loading indicator based on anticipated wait time ↓ ✅ Perception of wait time is more important than its duration. 🤔 Users overestimate passive waiting (standing still) by 36%. ✅ Active waiting (walking, interacting) feels much shorter. ✅ 20% rule: users only notice speed changes of at least 20%. 🤔 Small optimizations (e.g. shaving 0.2s off 5s) go unnoticed. ✅ 2 questions: "How much longer?" and "Is it working?" 🚫 Don’t use any loading indicators for waiting times < 1s. ✅ Short wait times (1–3s): use skeleton screens or spinners. ✅ Medium wait times (3–10s): use progress bars or indicators. ✅ Long wait time (10+s): show progress and allow interaction. 🤔 Uncertainty makes waiting feel significantly longer. ✅ Explain to users what’s happening in the background. ✅ Optimistic UI: ask for next steps while procees is running. ✅ The more valuable the reward, the longer tolerance to wait. ✅ Aim for improving perceived speed with reduced passive wait. Often we can’t speed up interactions for technical reasons. But we can reduce the perceived waiting time, which is often way more important than the actual duration. When a UI visualizes progress, users accept longer waits because they have right expectations and can track progress ((Buell & Norton, 2011). People are impatient if they don’t know how long to wait. Waiting without any explanation (spinning circle) feels longer than one where the product says why it’s busy. Also, waiting to START a task feels longer than waiting for a task to FINISH, so early start helps reduce frustrations as well. Users also tend to be highly sensitive to “queue jumping”. If a process they started later finishes earlier than a previous one, it creates significant frustration and abandonment. In the end, it’s all about setting right expectations, explaining what happens frequently and keeping people busy when waiting. It might not necessarily help make the application faster, but it will make it feel faster — and it could be enough to keep users on the page for just a little bit longer, and drive them to success from there. – ✤ Useful resources: Perceived Performance (Series), by Denys Mishunov https://lnkd.in/dvVkt3r3 Loading and Progress Indicators UX, by Taras Bakusevych https://lnkd.in/e5KFPiiq ↓

Everyone’s excited to launch AI agents. Almost no one knows how to measure if they’re actually working. Over the last year, we’ve seen brands launch everything from GenAI assistants to support bots to creative copilots but the post-launch metrics often look like this: • Number of chats • Average latency • Session duration • Daily active users Useful? Yes. But sufficient? Not even close. At ALTRD, we’ve worked on AI agents for enterprises and if there’s one lesson it’s this: Speed and usage mean nothing if the agent isn’t solving the actual problem. The real performance indicators are far more nuanced. Here’s what we’ve learned to track instead: 🔹 Task Completion Rate — Can the AI go beyond answering a question and actually complete a workflow? 🔹 User Trust — Do people come back? Do they feel confident relying on the agent again? 🔹 Conversation Depth — Is the agent handling complex, multi-turn exchanges with consistency? 🔹 Context Retention — Can it remember prior interactions and respond accordingly? 🔹 Cost per Successful Interaction — Not just cost per query, but cost per outcome. Massive difference. One of our clients initially celebrated their bot’s 1 million+ sessions - until we uncovered that less than 8% of users actually got what they came for. That 8% wasn’t a usage issue. It was a design and evaluation issue. They had optimized for traffic. Not trust. Not success. Not satisfaction. So we rebuilt the evaluation framework - adding feedback loops, success markers, and goal-completion metrics. The results? CSAT up by 34% Drop-off down by 40% Same infra cost, 3x more value delivered The takeaway: Don’t just measure what’s easy. Measure what matters. AI agents aren’t just tools - they’re touchpoints. They represent your brand, shape user experience, and influence business outcomes. P.S. What’s one underrated metric you’ve used to evaluate AI performance? Curious to learn what others are tracking.

🔥 If your Data Catalog isn’t measured, it’s probably failing. Most data catalogs don’t fail because of technology. They fail because success is never clearly defined. So let’s be blunt. Here’s how you actually know whether your data catalog works. ❌ Vanity metric to forget: “Number of datasets cataloged” ✔️ Metrics that matter: 🔴 1. Do people come back? (Adoption) One login ≠ success. Are users still active after onboarding? Are they searching… or asking Slack instead? If usage drops, your catalog is just expensive documentation. 🔴 2. Is the metadata good enough to trust? Auto-ingested metadata ≠ usable metadata. Do datasets have owners? Are descriptions written for humans? No context = no trust = no usage. 🔴 3. Does it actually save time? If analysts still spend hours “data hunting”, the catalog failed. Can users find the right dataset in minutes? Are the same questions still asked every week? If nothing changes, value is zero. 🔴 4. Who is accountable for the data? “Shared responsibility” usually means “no responsibility”. Is every critical dataset owned? Do stewards respond? Governance starts with naming names. 🔴 5. Can users tell which data is safe to use? Without trust signals, catalogs create confusion — not clarity. Certified datasets Data quality visibility Clear warnings for risky data No signals = no confidence = shadow data. 🔴 6. Is the platform reducing manual effort — or creating more? If stewardship feels like extra work, it won’t scale. How much is automated? Is steward workload increasing or decreasing? If governance doesn’t scale, it dies. 🔴 7. Does the business feel the impact? This is the uncomfortable question. Faster decisions? More reuse? Fewer duplicated datasets? If leadership can’t feel the difference, they won’t fund it. ⚠️ Hard truth: A data catalog is not a compliance tool. It’s not a metadata repository. It’s not a checkbox. It’s a product, and products live or die by adoption, trust, and impact. 💬 Be honest: Which of these KPIs are you actually tracking today?

In UX, we talk a lot about what users think, but we rarely study how their attitudes actually change over time. Most research still relies on one-time surveys like SUS, NPS, or post-test ratings. These snapshots are useful, but they tell us almost nothing about how trust grows, how frustration accumulates, or how confidence rises and collapses after a single confusing update. Attitudes are not steady states. They are trajectories shaped by experience. There are scientific ways to track those trajectories. Continuous-Time SEM lets researchers measure how satisfaction or trust evolves in real time, even if we collect feedback at irregular moments. A streaming app can trigger a question after each session and see exactly when enjoyment starts to drop, so recommendations can intervene before disengagement sets in. Latent Transition Analysis helps us understand how people move between hidden states such as novice, intermediate, competent, or stuck. Instead of guessing who needs help in onboarding, we can calculate the probability a user will progress or remain frustrated and then redesign tutorials to move them forward. Bayesian Hierarchical Models solve a common UX problem. What if we do not have huge samples like consumer apps do? With twenty or thirty enterprise users, traditional statistics break down, but Bayesian methods still model growth and decline in attitudes. They can reveal that confidence improves for new employees but decreases for experts after a redesign, a pattern that would otherwise remain invisible. Joint Modeling goes further by connecting attitude trends with real outcomes such as churn. It can show that a drop in usability or motivation predicts cancellation two weeks before users actually leave, turning measurement into prevention. One of the most powerful and practical tools is Hidden Markov Modeling. Instead of relying on surveys, it infers emotional states from behavior like hesitation, rage clicks, repeated backtracking, or abandoned tasks. It detects frustration even when people are silent, revealing emotional shifts that traditional surveys fail to capture. If you want to go deeper into these methods and see more concrete examples, I put together a full breakdown on the blog. You can read it here: https://lnkd.in/eY_Nwme2

How are your Trust & Safety teams measuring success today? Are you tracking the number of times the New York Times reports on your platform’s mishaps? Or when the CEO flags unpleasant content on their feed? While the prematurity of these “metrics” may seem exaggerated, #trustandsafety decisions are often made this way, even in the most data-driven organizations. Measuring the impact of T&S efforts are critical not only for compliance and transparency reports, but also for understanding the gaps and opportunities for improvement. Without objective #metrics, it is hard to quantify the effect the T&S teams are having on the user experience, or the unintentional side effects of overenforcement. Metrics also function as a north star for aligning the team. I’ve created a framework to categorize some of the key metrics: ⚙ Operational: Efficiency of the systems in place, track the day-to-day and overall health of operations, crucial for managing and optimizing the workflow and processes of the T&S team. 🎯 First-degree: Direct outcomes or immediate impacts of interventions, help understand if policies and actions are working well and how they are affecting the safety of users. 🔭 Second-degree: Long-term, indirect impacts and implications of interventions and policies on the ecosystem, provide a broader perspective on the strategic effectiveness of T&S efforts on overall user experience and business health. Traditionally, the focus of T&S teams has been on the operational and first-degree metrics, which are necessary for compliance and falls directly under the role of a T&S team. They also form the building blocks to measure second-degree metrics. However, in order to receive more investment and recognition from the company, T&S teams must track second-degree metrics. While protecting users from harmful content is seen as the “right thing to do,” it can lead to a short-term decrease in engagement (and thus, ad revenue). Having to spend money on #contentmoderation, which may lead to a loss in immediate revenue, can make companies wary of investing in T&S. However, we have seen that investment in T&S can lead to long-term financial gains. My former Metamates Glenn and Matt analyzed experiments conducted at Meta where the company found that the users who did not receive algorithmic protections from harmful content over the span of two years began to disengage and even quit the platform, while users who received the protections engaged more over time (article linked in the comments). Matthew, former head of T&S at Spectrum Labs, also said that their clients experienced an increase of customer LTV by 30% by reducing harmful content and promoting healthy behaviors. Veterans like Jeff Dunn have found that the better framing for T&S is as a “driver of organic growth rather than a cost center.” T&S teams can have so much more success by aligning their impact with the priorities of the company. The first step is measurement.

We do not experience the world in neat, discrete categories, yet much of UX research still measures behavior as if we do. Real experiences exist in the gray zone where satisfaction, trust, confusion, effort, and motivation overlap rather than fall into clean categories. When we compress this psychological complexity into Likert scales or binary outcomes, we lose the intensity and uncertainty that often signal early friction and churn. Most classic UX metrics summarize what users select, not what they actually feel. A single satisfaction score can hide hesitation, mixed emotions, and declining confidence, even though these blended states drive real behavioral change. By forcing fluid cognition into rigid buckets, we frame experience as static when in reality it is continuously evolving. Fuzzy logic approaches UX measurement differently by modeling experience as degrees of membership instead of fixed categories. Using membership functions, telemetry and survey inputs become graded psychological states in which multiple conditions coexist at once. Cognitive load, trust, frustration, and engagement are not treated as on–off switches but as overlapping mental states, allowing UX researchers to detect subtle tensions long before they appear as abandonment or negative feedback. Traditional regression assumes linear relationships and independence between variables, while ANOVA struggles to integrate many experiential dimensions into a single coherent signal. Fuzzy inference systems naturally combine correlated inputs into holistic experience indices, and through defuzzification these blended psychological states become continuous, actionable metrics such as friction levels or churn risk scores that support proportionate design responses instead of blunt thresholds. You might think Likert scales already work like fuzzy logic because they use graded numbers, but they are fundamentally different. Likert forces users to choose a single category, compressing mixed emotions into one number. When we later average scores or run regressions, we treat those values as if they represent continuous psychological intensity, even though the underlying uncertainty has already been removed at the moment of response. Fuzzy logic does the opposite. It preserves uncertainty instead of eliminating it, allowing users to belong partially to multiple psychological states at the same time. A person can be modeled as 70% satisfied, 20% neutral, and 10% confused simultaneously, rather than being forced into selecting whichever single box feels closest. Fuzzy logic does not replace traditional statistics, but it fills the gap where human psychology is layered, nonlinear, and ambiguous. Likert tells us which box users pick, classical statistics compare group averages, but fuzzy logic models how experience actually unfolds inside the mind, enabling UX research to move from static description toward psychologically grounded prediction and adaptive design.

Sometimes the smartest thing you can do for your users is make them wait. That sounds counterintuitive, right? In a world obsessed with speed, “instant” has become the ultimate UX religion. But psychology, design research, and even social media experiments point to the opposite: friction, when intentional, creates trust, thoughtfulness, and quality. 1. The Labor Illusion: Why We Value “Effort” When a chatbot responds instantly, users often dismiss it as scripted or mechanical. But add a short, 1–3 second pause (paired with a “typing…” indicator), and satisfaction scores rise. Why? Because the delay signals effort. Users feel like the system is “thinking” for them. This is the labor illusion: we value work more when we see (or think) effort is being invested. Too fast feels robotic; too slow feels broken. The sweet spot? Just long enough to feel intentional. 2. Even Social Media Learned This Lesson In 2020, Twitter tested a prompt: “Want to read this article before retweeting?” The results? - 40% more opens on articles. - 33% more people read before retweeting. One tiny pause. Massive behavior shift. It didn’t break the product. It improved it. 3. Where Friction Becomes a Feature Not all delays are good, but here’s where they shine: - Chatbots: Typing indicators and micro-pauses that humanize. - Surveys & Forms: Mental effort that filters noise and raises quality. - High-Stakes Actions: Confirmations before deleting, sending money, or posting. - Community Health: Pauses that nudge people to reflect before reacting. The takeaway: Don’t just obsess over removing friction. Ask: Where should I add it? Because sometimes, the best user experience isn’t about moving faster. It’s about giving people a moment to stop, think, and trust what happens next.

AI changes how we measure UX. We’ve been thinking and iterating on how we track user experiences with AI. In our open Glare framework, we use a mix of attitudinal, behavioral, and performance metrics. AI tools open the door to customizing metrics based on how people use each experience. I’d love to hear who else is exploring this. To measure UX in AI tools, it helps to follow the user journey and match the right metrics to each step. Here's a simple way to break it down: 1. Before using the tool Start by understanding what users expect and how confident they feel. This gives you a sense of their goals and trust levels. 2. While prompting  Track how easily users explain what they want. Look at how much effort it takes and whether the first result is useful. 3. While refining the output Measure how smoothly users improve or adjust the results. Count retries, check how well they understand the output, and watch for moments when the tool really surprises or delights them. 4. After seeing the results Check if the result is actually helpful. Time-to-value and satisfaction ratings show whether the tool delivered on its promise. 5. After the session ends See what users do next. Do they leave, return, or keep using it? This helps you understand the lasting value of the experience. We need sharper ways to measure how people use AI. Clicks can’t tell the whole story. But getting this data is not easy. What matters is whether the experience builds trust, sparks creativity, and delivers something users feel good about. These are the signals that show us if the tool is working, not just technically, but emotionally and practically. How are you thinking about this? #productdesign #uxmetrics #productdiscovery #uxresearch

💡 How to design effective progress bar (+ Figma UI Kits) Progress bar visually communicates the progress of an ongoing task to the user. It gives users a better idea of how much of the task has been completed and how much is left so that they can accurately plan their time. ✅ When to show the progress bar: Typically, you need to show a progress bar for long-running tasks that take more than 10 seconds to complete, such as file upload or installation process. When the wait is long enough, users will notice it and appreciate feedback on the progress. For tasks that take less than 10 seconds, consider using other indicators like infinite loading spinners or skeleton screens. ✅ How to design an informative progress bar: ✔ The progress bar should appear as soon as the user initiates a process. ✔ Show the progress bar in the context of interaction (follow the principle of proximity). ✔ Ensure the progress bar moves in accordance with the actual task progress. Misleading progress bars can frustrate users. Adjust the progress bar speed to match the expected time for task completion, slowing down if necessary to match longer tasks. ✔ Use the right size and shape. The progress bar should be large enough to be easily noticed but not so large that it distracts from other content. ✔ Add supporting text to communicate the status (e.g., "45% completed" or "Uploading file (3 of 5)"). ✔ Communicate remaining time for lengthy processes (>10min). It will mitigate the fear that the system is not responding.  ✔ Ensure that the progress bar animation is smooth & steady and does not hinder the performance of the app. ✔ Provide clear feedback when the process is complete (e.g., show a green checkmark icon) ✅ Psychological trick: When designing animation for a progress bar, you can start quickly but end slowly. Begin the progress bar quickly, even if it's just a small jump. The immediate response reduces perceived wait times. Slowing down towards the end can create a sense of anticipation and reduce perceived inaccuracy if the process takes longer than expected. 📖 Design Systems with Figma UI Kits: Practical design recommendations & UI assets for the progress bar. ✔ IBM Carbon: https://lnkd.in/d8N5U2n5 ✔ Dell Design System https://lnkd.in/d8DxM23A ✔ Microsoft Fluent 2 https://lnkd.in/dDZyPRxd 🖼 by Cristina Guedes #UI #userinterface #userinterfacedesign #productdesign #ux #userexperience #uxdesign

For two years, I sat in QBRs where the CSAT scores were green. We celebrated. We reported up. We told the story of a program that was working. Then someone pulled the retention data. Customers were rating interactions 4 out of 5. And not coming back. I did not understand it at first. The scores said they were happy. The behavior said something else entirely. Here is what I got wrong. I was measuring how customers felt at the end of one conversation. I was calling that loyalty. Those are not the same thing. A customer can feel good about a return interaction and still decide the brand is not worth the effort next time. A customer can rate a support call a 4 and quietly move their spend somewhere else. CSAT captures a moment. Loyalty is a pattern. The thing that changed how I saw everything was a single Forrester data point. Trust-driven customers generate 2.6 times more revenue than satisfied ones. Not 10% more. 2.6 times more. I had been optimizing for satisfaction when the money was in trust. And trust is built on completely different signals. Whether the brand did what it promised. Whether the policy felt fair. Whether getting a problem fixed required a fight. None of that appeared in our post-interaction survey. I rebuilt the measurement framework after that. It was uncomfortable to present because the new numbers were not as green. But they were honest. If your CSAT scores are green and your retention is flat, what are your scores actually measuring? #csat #customercare #contactcenter