July 10, 2018

My journey running a design sprint

Recently, everyone in the design industry has been talking about design sprints. So, naturally, the team at Optimal Workshop wanted to see what all the fuss was about. I picked up a copy of The Sprint Book and suggested to the team that we try out the technique.

In order to keep momentum, we identified a current problem and decided to run the sprint only two weeks later. The short notice was a bit of a challenge, but in the end we made it work. Here’s a run down of how things went, what worked, what didn’t, and lessons learned.

A sprint is an intensive focused period of time to get a product or feature designed and tested with the goal of knowing whether or not the team should keep investing in the development of the idea. The idea needs to be either validated or not validated by the end of the sprint. In turn, this saves time and resource further down the track by being able to pivot early if the idea doesn’t float.

If you’re following The Sprint Book you might have a structured 5 day plan that looks likes this:

  • Day 1 - Understand: Discover the business opportunity, the audience, the competition, the value proposition and define metrics of success.
  • Day 2 - Diverge: Explore, develop and iterate creative ways of solving the problem, regardless of feasibility.
  • Day 3 - Converge: Identify ideas that fit the next product cycle and explore them in further detail through storyboarding.
  • Day 4 - Prototype: Design and prepare prototype(s) that can be tested with people.
  • Day 5 - Test: User testing with the product's primary target audience.
Design sprint cycle
 With a Design Sprint, a product doesn't need to go full cycle to learn about the opportunities and gather feedback.

When you’re running a design sprint, it’s important that you have the right people in the room. It’s all about focus and working fast; you need the right people around in order to do this and not have any blocks down the path. Team, stakeholder and expert buy-in is key — this is not a task just for a design team!After getting buy in and picking out the people who should be involved (developers, designers, product owner, customer success rep, marketing rep, user researcher), these were my next steps:

Pre-sprint

  1. Read the book
  2. Panic
  3. Send out invites
  4. Write the agenda
  5. Book a meeting room
  6. Organize food and coffee
  7. Get supplies (Post-its, paper, Sharpies, laptops, chargers, cameras)

Some fresh smoothies for the sprinters made by our juice technician
 Some fresh smoothies for the sprinters made by our juice technician

The sprint

Due to scheduling issues we had to split the sprint over the end of the week and weekend. Sprint guidelines suggest you hold it over Monday to Friday — this is a nice block of time but we had to do Thursday to Thursday, with the weekend off in between, which in turn worked really well. We are all self confessed introverts and, to be honest, the thought of spending five solid days workshopping was daunting. At about two days in, we were exhausted and went away for the weekend and came back on Monday feeling sociable and recharged again and ready to examine the work we’d done in the first two days with fresh eyes.

Design sprint activities

During our sprint we completed a range of different activities but here’s a list of some that worked well for us. You can find out more information about how to run most of these over at The Sprint Book website or checkout some great resources over at Design Sprint Kit.

Lightning talks

We kicked off our sprint by having each person give a quick 5-minute talk on one of these topics in the list below. This gave us all an overview of the whole project and since we each had to present, we in turn became the expert in that area and engaged with the topic (rather than just listening to one person deliver all the information).

Our lightning talk topics included:

  • Product history - where have we come from so the whole group has an understanding of who we are and why we’ve made the things we’ve made.
  • Vision and business goals - (from the product owner or CEO) a look ahead not just of the tools we provide but where we want the business to go in the future.
  • User feedback - what have users been saying so far about the idea we’ve chosen for our sprint. This information is collected by our User Research and Customer Success teams.
  • Technical review - an overview of our tech and anything we should be aware of (or a look at possible available tech). This is a good chance to get an engineering lead in to share technical opportunities.
  • Comparative research - what else is out there, how have other teams or products addressed this problem space?

Empathy exercise

I asked the sprinters to participate in an exercise so that we could gain empathy for those who are using our tools. The task was to pretend we were one of our customers who had to present a dendrogram to some of our team members who are not involved in product development or user research. In this frame of mind, we had to talk through how we might start to draw conclusions from the data presented to the stakeholders. We all gained more empathy for what it’s like to be a researcher trying to use the graphs in our tools to gain insights.

How Might We

In the beginning, it’s important to be open to all ideas. One way we did this was to phrase questions in the format: “How might we…” At this stage (day two) we weren’t trying to come up with solutions — we were trying to work out what problems there were to solve. ‘We’ is a reminder that this is a team effort, and ‘might’ reminds us that it’s just one suggestion that may or may not work (and that’s OK). These questions then get voted on and moved into a workshop for generating ideas (see Crazy 8s).Read a more detailed instructions on how to run a ‘How might we’ session on the Design Sprint Kit website.

Crazy 8s

This activity is a super quick-fire idea generation technique. The gist of it is that each person gets a piece of paper that has been folded 8 times and has 8 minutes to come up with eight ideas (really rough sketches). When time is up, it’s all pens down and the rest of the team gets to review each other's ideas.In our sprint, we gave each person Post-it notes, paper, and set the timer for 8 minutes. At the end of the activity, we put all the sketches on a wall (this is where the art gallery exercise comes in).

Mila our data scientist sketching intensely during Crazy 8s
 Mila our data scientist sketching intensely during Crazy 8s

A close up of some sketches from the team
 A close up of some sketches from the team

Art gallery/Silent critique

The art gallery is the place where all the sketches go. We give everyone dot stickers so they can vote and pull out key ideas from each sketch. This is done silently, as the ideas should be understood without needing explanation from the person who made them. At the end of it you’ve got a kind of heat map, and you can see the ideas that stand out the most. After this first round of voting, the authors of the sketches get to talk through their ideas, then another round of voting begins.

Mila putting some sticky dots on some sketches
 Mila putting some sticky dots on some sketches

Bowie, our head of security/office dog, even took part in the sprint...kind of.
 Bowie, our head of security, even took part in the sprint...kind of

Usability testing and validation

The key part of a design sprint is validation. For one of our sprints we had two parts of our concept that needed validating. To test one part we conducted simple user tests with other members of Optimal Workshop (the feature was an internal tool). For the second part we needed to validate whether we had the data to continue with this project, so we had our data scientist run some numbers and predictions for us.

6-dan-design-sprintOur remote worker Rebecca dialed in to watch one of our user tests live
 Our remote worker Rebecca dialed in to watch one of our user tests live
"I'm pretty bloody happy" — Actual feedback.
 Actual feedback

Challenges and outcomes

One of our key team members, Rebecca, was working remotely during the sprint. To make things easier for her, we set up 2 cameras: one pointed to the whiteboard, the other was focused on the rest of the sprint team sitting at the table. Next to that, we set up a monitor so we could see Rebecca.

Engaging in workshop activities is a lot harder when working remotely. Rebecca would get around this by completing the activities and take photos to send to us.

8-rebecca-design-sprint
 For more information, read this great Medium post about running design sprints remotely

Lessons

  • Lightning talks are a great way to have each person contribute up front and feel invested in the process.
  • Sprints are energy intensive. Make sure you’re in a good place with plenty of fresh air with comfortable chairs and a break out space. We like to split the five days up so that we get a weekend break.
  • Give people plenty of notice to clear their schedules. Asking busy people to take five days from their schedule might not go down too well. Make sure they know why you’d like them there and what they should expect from the week. Send them an outline of the agenda. Ideally, have a chat in person and get them excited to be part of it.
  • Invite the right people. It’s important that you get the right kind of people from different parts of the company involved in your sprint. The role they play in day-to-day work doesn’t matter too much for this. We’re all mainly using pens and paper and the more types of brains in the room the better. Looking back, what we really needed on our team was a customer support team member. They have the experience and knowledge about our customers that we don’t have.
  • Choose the right sprint problem. The project we chose for our first sprint wasn’t really suited for a design sprint. We went in with a well defined problem and a suggested solution from the team instead of having a project that needed fresh ideas. This made the activities like ‘How Might We’ seem very redundant. The challenge we decided to tackle ended up being more of a data prototype (spreadsheets!). We used the week to validate assumptions around how we can better use data and how we can write a script to automate some internal processes. We got the prototype working and tested but due to the nature of the project we will have to run this experiment in the background for a few months before any building happens.

Overall, this design sprint was a great team bonding experience and we felt pleased with what we achieved in such a short amount of time. Naturally, here at Optimal Workshop, we're experimenters at heart and we will keep exploring new ways to work across teams and find a good middle ground.

Further reading

Share this article
Author
Optimal
Workshop
Topics

Related articles

View all blog articles
Learn more
1 min read

When Personalization Gets Personal: Balancing AI with Human-Centered Design

AI-driven personalization is redefining digital experiences, allowing companies to tailor content, recommendations, and interfaces to individual users at an unprecedented scale. From e-commerce product suggestions to content feeds, streaming recommendations, and even customized user interfaces, personalization has become a cornerstone of modern digital strategy. The appeal is clear: research shows that effective personalization can increase engagement by 72%, boost conversion rates by up to 30%, and drive revenue growth of 10-15%.

However, the reality often falls short of these impressive statistics. Personalization can easily backfire, frustrating users instead of engaging them, creating experiences that feel invasive rather than helpful, and sometimes actively driving users away from the very content or products they might genuinely enjoy. Many organizations invest heavily in AI technology while underinvesting in understanding how these personalized experiences actually impact their users.

The Widening Gap Between Capability and Quality

The technical capability to personalize digital experiences has advanced rapidly, but the quality of these experiences hasn't always kept pace. According to a 2023 survey by Baymard Institute, 68% of users reported encountering personalization that felt "off-putting" or "frustrating" in the previous month, while only 34% could recall a personalized experience that genuinely improved their interaction with a digital product.

This disconnect stems from a fundamental misalignment: while AI excels at pattern recognition and prediction based on historical data, it often lacks the contextual understanding and nuance that make personalization truly valuable. The result? Technically sophisticated personalization regularly misses the mark on actual user needs and preferences.

The Pitfalls of AI-Driven Personalization

Many companies struggle with personalization due to several common pitfalls that undermine even the most sophisticated AI implementations:

Over-Personalization: When Helpful Becomes Restrictive

AI that assumes too much can make users feel restricted or trapped in a "filter bubble" of limited options. This phenomenon, often called "over-personalization," occurs when algorithms become too confident in their understanding of user preferences.

Signs of over-personalization include:

  • Content feeds that become increasingly homogeneous over time
  • Disappearing options that might interest users but don't match their history
  • User frustration at being unable to discover new content or products
  • Decreased engagement as experiences become predictable and stale

A study by researchers at University of Minnesota found that highly personalized news feeds led to a 23% reduction in content diversity over time, even when users actively sought varied content. This "filter bubble" effect not only limits discovery but can leave users feeling manipulated or constrained.

Incorrect Assumptions: When Data Tells the Wrong Story

AI recommendations based on incomplete or misinterpreted data can lead to irrelevant, inappropriate, or even offensive suggestions. These incorrect assumptions often stem from:

  • Limited data points that don't capture the full context of user behavior
  • Misinterpreting casual interest as strong preference
  • Failing to distinguish between the user's behavior and actions taken on behalf of others
  • Not recognizing temporary or situational needs versus ongoing preferences

These misinterpretations can range from merely annoying (continuously recommending products similar to a one-time purchase) to deeply problematic (showing weight loss ads to users with eating disorders based on their browsing history).

A particularly striking example occurred when a major retailer's algorithm began sending pregnancy-related offers to a teenage girl before her family knew she was pregnant. While technically accurate in its prediction, this incident highlights how even "correct" personalization can fail to consider the broader human context and implications.

Lack of Transparency: The Black Box Problem

Users increasingly want to understand why they're being shown specific content or recommendations. When personalization happens behind a "black box" without explanation, it can create:

  • Distrust in the system and the brand behind it
  • Confusion about how to influence or improve recommendations
  • Feelings of being manipulated rather than assisted
  • Concerns about what personal data is being used and how

Research from the Pew Research Center shows that 74% of users consider it important to know why they are seeing certain recommendations, yet only 22% of personalization systems provide clear explanations for their suggestions.

Inconsistent Experiences Across Channels

Many organizations struggle to maintain consistent personalization across different touchpoints, creating disjointed experiences:

  • Product recommendations that vary wildly between web and mobile
  • Personalization that doesn't account for previous customer service interactions
  • Different personalization strategies across email, website, and app experiences
  • Recommendations that don't adapt to the user's current context or device

This inconsistency can make personalization feel random or arbitrary rather than thoughtfully tailored to the user's needs.

Neglecting Privacy Concerns and Control

As personalization becomes more sophisticated, user concerns about privacy intensify. Key issues include:

  • Collecting more data than necessary for effective personalization
  • Lack of user control over what information influences their experience
  • Unclear opt-out mechanisms for personalization features
  • Personalization that reveals sensitive information to others

A recent study found that 79% of users want control over what personal data influences their recommendations, but only 31% felt they had adequate control in their most-used digital products.

How Product Managers Can Leverage UX Insight for Better AI Personalization

To create a personalized experience that feels natural and helpful rather than creepy or restrictive, UX teams need to validate AI-driven decisions through systematic research with real users. Rather than treating personalization as a purely technical challenge, successful organizations recognize it as a human-centered design problem that requires continuous testing and refinement.

Understanding User Mental Models Through Card Sorting & Tree Testing

Card sorting and tree testing help structure content in a way that aligns with users' expectations and mental models, creating a foundation for personalization that feels intuitive rather than imposed:

  • Open and Closed Card Sorting – Helps understand how different user segments naturally categorize content, products, or features, providing a baseline for personalization strategies
  • Tree Testing – Validates whether personalized navigation structures work for different user types and contexts
  • Hybrid Approaches – Combining card sorting with interviews to understand not just how users categorize items, but why they do so

Case Study: A financial services company used card sorting with different customer segments to discover distinct mental models for organizing financial products. Rather than creating a one-size-fits-all personalization system, they developed segment-specific personalization frameworks that aligned with these different mental models, resulting in a 28% increase in product discovery and application rates.

Validating Interaction Patterns Through First-Click Testing

First-click testing ensures users interact with personalized experiences as intended across different contexts and scenarios:

  • Testing how users respond to personalized elements vs. standard content
  • Evaluating whether personalization cues (like "Recommended for you") influence click behavior
  • Comparing how different user segments respond to the same personalization approaches
  • Identifying potential confusion points in personalized interfaces

Research by the Nielsen Norman Group found that getting the first click right increases the overall task success rate by 87%. For personalized experiences, this is even more critical, as users may abandon a site entirely if early personalized recommendations seem irrelevant or confusing.

Gathering Qualitative Insights Through User Interviews & Usability Testing

Direct observation and conversation with users provides critical context for personalization strategies:

  • Moderated Usability Testing – Reveals how users react to personalized elements in real-time
  • Think-Aloud Protocols – Help understand users' expectations and reactions to personalization
  • Longitudinal Studies – Track how perceptions of personalization change over time and repeated use
  • Contextual Inquiry – Observes how personalization fits into users' broader goals and environments

These qualitative approaches help answer critical questions like:

  • When does personalization feel helpful versus intrusive?
  • What level of explanation do users want for recommendations?
  • How do different user segments react to similar personalization strategies?
  • What control do users expect over their personalized experience?

Measuring Sentiment Through Surveys & User Feedback

Systematic feedback collection helps gauge users' comfort levels with AI-driven recommendations:

  • Targeted Microsurveys – Quick pulse checks after personalized interactions
  • Preference Centers – Direct input mechanisms for refining personalization
  • Satisfaction Tracking – Monitoring how personalization affects overall satisfaction metrics
  • Feature-Specific Feedback – Gathering input on specific personalization features

A streaming service discovered through targeted surveys that users were significantly more satisfied with content recommendations when they could see a clear explanation of why items were suggested (e.g., "Because you watched X"). Implementing these explanations increased content exploration by 34% and reduced account cancellations by 8%.

A/B Testing Personalization Approaches

Experimental validation ensures personalization actually improves key metrics:

  • Testing different levels of personalization intensity
  • Comparing explicit versus implicit personalization methods
  • Evaluating various approaches to explaining recommendations
  • Measuring the impact of personalization on both short and long-term engagement

Importantly, A/B testing should look beyond immediate conversion metrics to consider longer-term impacts on user satisfaction, trust, and retention.

Building a User-Centered Personalization Strategy That Works

To implement personalization that truly enhances user experience, organizations should follow these research-backed principles:

1. Start with User Needs, Not Technical Capabilities

The most effective personalization addresses genuine user needs rather than showcasing algorithmic sophistication:

  • Identify specific pain points that personalization could solve
  • Understand which aspects of your product would benefit most from personalization
  • Determine where users already expect or desire personalized experiences
  • Recognize which elements should remain consistent for all users

2. Implement Transparent Personalization

Users increasingly expect to understand and control how their experiences are personalized:

  • Clearly communicate what aspects of the experience are personalized
  • Explain the primary factors influencing recommendations
  • Provide simple mechanisms for users to adjust or reset their personalization
  • Consider making personalization opt-in for sensitive domains

3. Design for Serendipity and Discovery

Effective personalization balances predictability with discovery:

  • Deliberately introduce variety into recommendations
  • Include "exploration" categories alongside highly targeted suggestions
  • Monitor and prevent increasing homogeneity in personalized feeds over time
  • Allow users to easily branch out beyond their established patterns

4. Apply Progressive Personalization

Rather than immediately implementing highly tailored experiences, consider a gradual approach:

  • Begin with light personalization based on explicit user choices
  • Gradually introduce more sophisticated personalization as users engage
  • Calibrate personalization depth based on relationship strength and context
  • Adjust personalization based on user feedback and behavior

5. Establish Continuous Feedback Loops

Personalization should never be "set and forget":

  • Implement regular evaluation cycles for personalization effectiveness
  • Create easy feedback mechanisms for users to rate recommendations
  • Monitor for signs of over-personalization or filter bubbles
  • Regularly test personalization assumptions with diverse user groups

The Future of Personalization: Human-Centered AI

As AI capabilities continue to advance, the companies that will succeed with personalization won't necessarily be those with the most sophisticated algorithms, but those who best integrate human understanding into their approach. The future of personalization lies in creating systems that:

  • Learn from qualitative human feedback, not just behavioral data
  • Respect the nuance and complexity of human preferences
  • Maintain transparency in how personalization works
  • Empower users with appropriate control
  • Balance algorithm-driven efficiency with human-centered design principles

AI should learn from real people, not just data. UX research ensures that personalization enhances, rather than alienates, users by bringing human insight to algorithmic decisions.

By combining the pattern-recognition power of AI with the contextual understanding provided by UX research, organizations can create personalized experiences that feel less like surveillance and more like genuine understanding: experiences that don't just predict what users might click, but truly respond to what they need and value.

Learn more
1 min read

Designing User Experiences for Agentic AI: The Next Frontier

Beyond Generative AI: A New Paradigm Emerges

The AI landscape is undergoing a profound transformation. While generative AI has captured public imagination with its ability to create content, a new paradigm is quietly revolutionizing how we think about human-computer interaction: Agentic AI.

Unlike traditional software that waits for explicit commands or generative AI focused primarily on content creation, Agentic AI represents a fundamental shift toward truly autonomous systems. These advanced AI agents can independently make decisions, take actions, and solve complex problems with minimal human oversight. Rather than simply responding to prompts, they proactively work toward goals, demonstrating initiative and adaptability that more closely resembles human collaboration than traditional software interaction.

This evolution is already transforming industries across the board:

  • In customer service, AI agents handle complex inquiries end-to-end
  • In software development, they autonomously debug code and suggest improvements
  • In healthcare, they monitor patient data and flag concerning patterns
  • In finance, they analyze market trends and execute optimized strategies
  • In manufacturing and logistics, they orchestrate complex operations with minimal human intervention

As these autonomous systems become more prevalent, designing exceptional user experiences for them becomes not just important, but essential. The challenge? Traditional UX approaches built around graphical user interfaces and direct manipulation fall short when designing for AI that thinks and acts independently.

The New Interaction Model: From Commands to Collaboration

Interacting with Agentic AI represents a fundamental departure from conventional software experiences. The predictable, structured nature of traditional GUIs—with their buttons, menus, and visual feedback—gives way to something more fluid, conversational, and at times, unpredictable.

The ideal Agentic AI experience feels less like operating a tool and more like collaborating with a capable teammate. This shift demands that UX designers look beyond the visual aspects of interfaces to consider entirely new interaction models that emphasize:

  • Natural language as the primary interface
  • The AI's ability to take initiative appropriately
  • Establishing the right balance of autonomy and human control
  • Building and maintaining trust through transparency
  • Adapting to individual user preferences over time

The core challenge lies in bridging the gap between users accustomed to direct manipulation of software and the more abstract interactions inherent in systems that can think and act independently. How do we design experiences that harness the power of autonomy while maintaining the user's sense of control and understanding?

Understanding Users in the Age of Autonomous AI

The foundation of effective Agentic AI design begins with deep user understanding. Expectations for these autonomous agents are shaped by prior experiences with traditional AI assistants but require significant recalibration given their increased autonomy and capability.

Essential UX Research Methods for Agentic AI

Several research methodologies prove particularly valuable when designing for autonomous agents:

User Interviews provide rich qualitative insights into perceptions, trust factors, and control preferences. These conversations reveal the nuanced ways users think about AI autonomy—often accepting it readily for low-stakes tasks like calendar management while requiring more oversight for consequential decisions like financial planning.

Usability Testing with Agentic AI prototypes reveals how users react to AI initiative in real-time. Observing these interactions highlights moments where users feel empowered versus instances where they experience discomfort or confusion when the AI acts independently.

Longitudinal Studies track how user perceptions and interaction patterns evolve as the AI learns and adapts to individual preferences. Since Agentic AI improves through use, understanding this relationship over time provides critical design insights.

Ethnographic Research offers contextual understanding of how autonomous agents integrate into users' daily workflows and environments. This immersive approach reveals unmet needs and potential areas of friction that might not emerge in controlled testing environments.

Key Questions to Uncover

Effective research for Agentic AI should focus on several fundamental dimensions:

Perceived Autonomy: How much independence do users expect and desire from AI agents across different contexts? When does autonomy feel helpful versus intrusive?

Trust Factors: What elements contribute to users trusting an AI's decisions and actions? How quickly is trust lost when mistakes occur, and what mechanisms help rebuild it?

Control Mechanisms: What types of controls (pause, override, adjust parameters) do users expect to have over autonomous systems? How can these be implemented without undermining the benefits of autonomy?

Transparency Needs: What level of insight into the AI's reasoning do users require? How can this information be presented effectively without overwhelming them with technical complexity?

The answers to these questions vary significantly across user segments, task types, and domains—making comprehensive research essential for designing effective Agentic AI experiences.

Core UX Principles for Agentic AI Design

Designing for autonomous agents requires a unique set of principles that address their distinct characteristics and challenges:

Clear Communication

Effective Agentic AI interfaces facilitate natural, transparent communication between user and agent. The AI should clearly convey:

  • Its capabilities and limitations upfront
  • When it's taking action versus gathering information
  • Why it's making specific recommendations or decisions
  • What information it's using to inform its actions

Just as with human collaboration, clear communication forms the foundation of successful human-AI partnerships.

Robust Feedback Mechanisms

Agentic AI should provide meaningful feedback about its operations and make it easy for users to provide input on its performance. This bidirectional exchange enables:

  • Continuous learning and refinement of the agent's behavior
  • Adaptation to individual user preferences
  • Improved accuracy and usefulness over time

The most effective agents make feedback feel conversational rather than mechanical, encouraging users to shape the AI's behavior through natural interaction.

Thoughtful Error Handling

How an autonomous agent handles mistakes significantly impacts user trust and satisfaction. Effective error handling includes:

  • Proactively identifying potential errors before they occur
  • Clearly communicating when and why errors happen
  • Providing straightforward paths for recovery or human intervention
  • Learning from mistakes to prevent recurrence

The ability to gracefully manage errors and learn from them is often what separates exceptional Agentic AI experiences from frustrating ones.

Appropriate User Control

Users need intuitive mechanisms to guide and control autonomous agents, including:

  • Setting goals and parameters for the AI to work within
  • The ability to pause or stop actions in progress
  • Options to override decisions when necessary
  • Preferences that persist across sessions

The level of control should adapt to both user expertise and task criticality, offering more granular options for advanced users or high-stakes decisions.

Balanced Transparency

Effective Agentic AI provides appropriate visibility into its reasoning and decision-making processes without overwhelming users. This involves:

  • Making the AI's "thinking" visible and understandable
  • Explaining data sources and how they influence decisions
  • Offering progressive disclosure—basic explanations for casual users, deeper insights for those who want them

Transparency builds trust by demystifying what might otherwise feel like a "black box" of AI decision-making.

Proactive Assistance

Perhaps the most distinctive aspect of Agentic AI is its ability to anticipate needs and take initiative, offering:

  • Relevant suggestions based on user context
  • Automation of routine tasks without explicit commands
  • Timely information that helps users make better decisions

When implemented thoughtfully, this proactive assistance transforms the AI from a passive tool into a true collaborative partner.

Building User Confidence Through Transparency and Explainability

For users to embrace autonomous agents, they need to understand and trust how these systems operate. This requires both transparency (being open about how the system works) and explainability (providing clear reasons for specific decisions).

Several techniques can enhance these critical qualities:

  • Feature visualization that shows what the AI is "seeing" or focusing on
  • Attribution methods that identify influential factors in decisions
  • Counterfactual explanations that illustrate "what if" scenarios
  • Natural language explanations that translate complex reasoning into simple terms

From a UX perspective, this means designing interfaces that:

  • Clearly indicate when users are interacting with AI versus human systems
  • Make complex decisions accessible through visualizations or natural language
  • Offer progressive disclosure—basic explanations by default with deeper insights available on demand
  • Implement audit trails documenting the AI's actions and reasoning

The goal is to provide the right information at the right time, helping users understand the AI's behavior without drowning them in technical details.

Embracing Iteration and Continuous Testing

The dynamic, learning nature of Agentic AI makes traditional "design once, deploy forever" approaches inadequate. Instead, successful development requires:

Iterative Design Processes

  • Starting with minimal viable agents and expanding capabilities based on user feedback
  • Incorporating user input at every development stage
  • Continuously refining the AI's behavior based on real-world interaction data

Comprehensive Testing Approaches

  • A/B testing different AI behaviors with actual users
  • Implementing feedback loops for ongoing improvement
  • Monitoring key performance indicators related to user satisfaction and task completion
  • Testing for edge cases, adversarial inputs, and ethical alignment

Cross-Functional Collaboration

  • Breaking down silos between UX designers, AI engineers, and domain experts
  • Ensuring technical capabilities align with user needs
  • Creating shared understanding of both technical constraints and user expectations

This ongoing cycle of design, testing, and refinement ensures Agentic AI continuously evolves to better serve user needs.

Learning from Real-World Success Stories

Several existing applications offer valuable lessons for designing effective autonomous systems:

Autonomous Vehicles demonstrate the importance of clearly communicating intentions, providing reassurance during operation, and offering intuitive override controls for passengers.

Smart Assistants like Alexa and Google Assistant highlight the value of natural language processing, personalization based on user preferences, and proactive assistance.

Robotic Systems in industrial settings showcase the need for glanceable information, simplified task selection, and workflows that ensure safety in shared human-robot environments.

Healthcare AI emphasizes providing relevant insights to professionals, automating routine tasks to reduce cognitive load, and enhancing patient care through personalized recommendations.

Customer Service AI prioritizes personalized interactions, 24/7 availability, and the ability to handle both simple requests and complex problem-solving.

These successful implementations share several common elements:

  • They prioritize transparency about capabilities and limitations
  • They provide appropriate user control while maximizing the benefits of autonomy
  • They establish clear expectations about what the AI can and cannot do

Shaping the Future of Human-Agent Interaction

Designing user experiences for Agentic AI represents a fundamental shift in how we think about human-computer interaction. The evolution from graphical user interfaces to autonomous agents requires UX professionals to:

  • Move beyond traditional design patterns focused on direct manipulation
  • Develop new frameworks for building trust in autonomous systems
  • Create interaction models that balance AI initiative with user control
  • Embrace continuous refinement as both technology and user expectations evolve

The future of UX in this space will likely explore more natural interaction modalities (voice, gesture, mixed reality), increasingly sophisticated personalization, and thoughtful approaches to ethical considerations around AI autonomy.

For UX professionals and AI developers alike, this new frontier offers the opportunity to fundamentally reimagine the relationship between humans and technology—moving from tools we use to partners we collaborate with. By focusing on deep user understanding, transparent design, and iterative improvement, we can create autonomous AI experiences that genuinely enhance human capability rather than simply automating tasks.

The journey has just begun, and how we design these experiences today will shape our relationship with intelligent technology for decades to come.

Learn more
1 min read

The Role of Usability Metrics in User-Centered Design

The term ‘usability’ captures sentiments of how usable, useful, enjoyable, and intuitive a website or app is perceived by users. By its very nature, usability is somewhat subjective. But what we’re really looking for when we talk about usability is how well a website can be used to achieve a specific task or goal. Using this definition we can analyze usability metrics (standard units of measurement) to understand how well user experience design is performing.

Usability metrics provide helpful insights before and after any digital product is launched. They help us form a deeper understanding of how we can design with the user front of mind. This user-centered design approach is considered the best-practice in building effective information architecture and user experiences that help websites, apps, and software meet and exceed users' needs.

In this article, we’ll highlight key usability metrics, how to measure and understand them, and how you can apply them to improve user experience.

Understanding Usability Metrics

Usability metrics aim to understand three core elements of usability, namely: effectiveness, efficiency, and satisfaction. A variety of research techniques offer designers an avenue for quantifying usability. Quantifying usability is key because we want to measure and understand it objectively, rather than making assumptions.

Types of Usability Metrics

There are a few key metrics that we can measure directly if we’re looking to quantify effectiveness, efficiency, and satisfaction. Here are four common examples:

  • Success rate: Also known as ‘completion rate’, success rate is the percentage of users who were able to successfully complete the tasks.
  • Time-based efficiency: Also known as ‘time on task’, time-based efficiency measures how much time a user needs to complete a certain task.
  • Number of errors: Sounds like what it is! It measures the average number of times where an error occurred per user when performing a given task.
  • Post-task satisfaction: Measures a user's general impression or satisfaction after completing (or not completing) a given task.

How to Collect Usability Metrics


Usability metrics are outputs from research techniques deployed when conducting usability testing. Usability testing in web design, for example, involves assessing how a user interacts with the website by observing (and listening to) users completing defined tasks, such as purchasing a product or signing up for newsletters.

Conducting usability testing and collecting usability metrics usually involves:

  • Defining a set of tasks that you want to test
  • Recruitment of test participants
  • Observing participants (remotely or in-person)
  • Recording detailed observations
  • Follow-up satisfaction survey or questionnaire

Tools such Reframer are helpful in conducting usability tests remotely, and they enable live collaboration of multiple team members. It is extremely handy when trying to record and organize those insightful observations! Using paper prototypes is an inexpensive way to test usability early in the design process.

The Importance of Usability Metrics in User-Centered Design

User-centered design challenges designers to put user needs first. This means in order to deploy user-centered design, you need to understand your user. This is where usability testing and metrics add value to website and app performance; they provide direct, objective insight into user behavior, needs, and frustrations. If your user isn’t getting what they want or expect, they’ll simply leave and look elsewhere.

Usability metrics identify which parts of your design aren’t hitting the mark. Recognizing where users might be having trouble completing certain actions, or where users are regularly making errors, are vital insights when implementing user-centered design. In short, user-centered design relies on data-driven user insight.

But why hark on about usability metrics and user-centered design? Because at the heart of most successful businesses is a well-solved user problem. Take Spotify, for example, which solved the problem of dodgy, pirated digital files being so unreliable. People liked access to free digital music, but they had to battle viruses and fake files to get it. With Spotify, for a small monthly fee, or the cost of listening to a few ads, users have the best of both worlds. The same principle applies to user experience - identify recurring problems, then solve them.

Best Practices for Using Usability Metrics

Usability metrics should be analyzed by design teams of every size. However, there are some things to bear in mind when using usability metrics to inform design decisions:

  • Defining success: Usability metrics are only valuable if they are being measured against clearly defined benchmarks. Many tasks and processes are unique to each business, so use appropriate comparisons and targets; usually in the form of an ‘optimized’ user (a user with high efficiency).
  • Real user metrics: Be sure to test with participants that represent your final user base. For example, there’s little point in testing your team, who will likely be intimately aware of your business structure, terminology, and internal workflows.
  • Test early: Usability testing and subsequent usability metrics provide the most impact early on in the design process. This usually means testing an early prototype or even a paper prototype. Early testing helps to avoid any significant, unforeseen challenges that could be difficult to rewind in your information architecture.
  • Regular testing: Usability metrics can change over time as user behavior and familiarity with digital products evolve. You should also test and analyze the usability of new feature releases on your website or app.

Remember, data analysis is only as good as the data itself. Give yourself the best chance of designing exceptional user experiences by collecting, researching, and analyzing meaningful and accurate usability metrics.

Conclusion

Usability metrics are a guiding light when it comes to user experience. As the old saying goes, “you can’t manage what you can’t measure”. By including usability metrics in your design process, you invite direct user feedback into your product. This is ideal because we want to leave any assumptions or guesswork about user experience at the door.

User-centered design inherently relies on constant user research. Usability metrics such as success rate, time-based efficiency, number of errors, and post-task satisfaction will highlight potential shortcomings in your design. Subsequently, they identify where improvements can be made, AND they lay down a benchmark to check whether any resulting updates addressed the issues.

Ready to start collecting and analyzing usability metrics? Check out our guide to planning and running effective usability tests to get a head start!

Seeing is believing

Explore our tools and see how Optimal makes gathering insights simple, powerful, and impactful.