<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Marianne Berkovich on Medium]]></title>
        <description><![CDATA[Stories by Marianne Berkovich on Medium]]></description>
        <link>https://medium.com/@mberkovich?source=rss-35410c649a3e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*JzODXdceV6derjSF5Apk5g.jpeg</url>
            <title>Stories by Marianne Berkovich on Medium</title>
            <link>https://medium.com/@mberkovich?source=rss-35410c649a3e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 07:00:15 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@mberkovich/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Designing AI for the Clinical Relationship (Part 2)]]></title>
            <link>https://medium.com/@mberkovich/designing-ai-for-the-clinical-relationship-part-2-ba8e89fa2aaf?source=rss-35410c649a3e------2</link>
            <guid isPermaLink="false">https://medium.com/p/ba8e89fa2aaf</guid>
            <category><![CDATA[digital-health]]></category>
            <category><![CDATA[user-centered-design]]></category>
            <category><![CDATA[product-development]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Marianne Berkovich]]></dc:creator>
            <pubDate>Tue, 06 Jan 2026 21:38:01 GMT</pubDate>
            <atom:updated>2026-01-20T18:42:28.118Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WuqTlMEHFbXDYcL7VtTZRA.png" /></figure><h3><strong>Designing AI for the Clinical Relationship</strong></h3><p><em>(This is Part 2 of a series. See </em><a href="https://medium.com/@mberkovich/does-your-ai-need-a-hippocratic-oath-four-emerging-models-of-clinical-care-56be7c907543"><em>Part 1: Does Your AI Need a Hippocratic Oath? Four Emerging Models of Clinical Care</em></a><em>)</em></p><p>Part 1 is about naming the kind of clinical relationship your AI system creates, Part 2 is about building it responsibly through actionable design principles and evaluation frameworks for each model.</p><h3><strong>Why AI Changes the Clinical Relationship</strong></h3><p>AI isn’t just “smarter” software. It behaves in ways that invite anthropomorphism, which fundamentally changes how people relate to it. Several key traits distinguish AI from traditional software:</p><ul><li><strong>Natural language interaction:</strong> AI systems communicate in fluent, conversational language. Unlike earlier software that forced people to adapt to input formats, AI adapts to the patient. The risk is empathy inflation, where patients mistake conversational fluency for genuine moral reasoning or clinical judgment.</li><li><strong>Probabilistic:</strong> Traditional clinical software is deterministic: input X reliably produces output Y. Probabilistic AI systems can be confidently wrong.</li><li><strong>Black Box Explainability:</strong> Because AI synthesizes many variables to identify patterns, its outputs are often opaque. AI cannot intuitively self-reflect, making verification harder, which can erode trust.</li></ul><p>Patients do not experience AI as a tool, but as a participant. The table below summarizes each model’s relationship analog, and owner of medical judgment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Gu_9VcELj3Q344QeZ_jbIA.png" /></figure><p>Before addressing model-specific considerations, we’ll go through what is common across models:</p><p><strong>What never changes</strong></p><ul><li>Foundational requirements</li><li>Preserving the healing relationship</li></ul><p><strong>What you always do</strong></p><ul><li>Design the interactions</li><li>Evaluate the interactions</li></ul><p>Then each model section describes what changes for design and evaluation considerations.</p><h3><strong>Foundational Requirements for All Models</strong></h3><p>Before addressing model-specific considerations, there are universal, table stakes expectations that apply to any AI system deployed in healthcare.</p><ul><li><strong>Deliver a clinical benefit. </strong>The AI system should measurably improve diagnosis, treatment, or care delivery in ways that matter to patients.</li><li><strong>Follow responsible AI principles. </strong>This includes fairness, safety, privacy, inclusiveness, accountability, and transparency, while not encoding or amplifying existing inequities.</li><li><strong>Adhere to user-centered design best practices.</strong> Since human cognitive limitations have not changed, AI systems should help users accomplish their goals with minimal friction.</li></ul><h4>Preserving the Healing Relationship</h4><p>The healing relationship depends on trust: patients must believe the system is acting in their best interests. Because AI communicates with apparent authority and human-like fluency, it holds disproportionate influence, especially in moments of vulnerability. Designing responsibly requires acknowledging this power and constraining it intentionally.</p><ul><li><strong>Make identity unambiguous. </strong>Patients should never be uncertain whether they are engaging with a human or a machine. AI systems that interact with patients must clearly and continuously identify themselves as artificial, using persistent visual or interaction cues rather than a one-time disclosure.</li><li><strong>Define scope, accountability, and information provenance.</strong> Patients need an accurate mental model of the system’s boundaries. This includes understanding the AI’s capabilities, the limits of its competence, and the sources of information it relies on. Patients should know who is accountable for decisions influenced by the AI. Clear articulation of scope and accountability helps neutralize the veneer of objectivity that AI outputs often project.</li><li><strong>Provide meaningful recourse and choice</strong>. Trust is reinforced when patients retain agency. Patients should be able to decline AI involvement without penalty or degradation of care, and that choice should be respected throughout the care journey. AI systems must also provide explicit pathways to raise concerns, request clarification, or seek human review.</li><li><strong>Support trust over time, not just in isolated interactions</strong>. Trust unfolds longitudinally. Systems should support continuity by appropriately remembering patient history, preferences, and context, helping patients feel recognized as individuals rather than one-off cases. Long-term trust is built not through isolated moments of competence, but through consistent, bounded, and accountable behavior over time.</li></ul><h4>Designing Every Interaction with Care</h4><p>Because AI systems often feel human, every interaction shapes how patients perceive care, competence, and accountability.</p><p><em>Design for How Patients Share Information</em></p><p>AI systems should make it easier for patients to describe what they are experiencing.</p><ul><li><strong>Create a positive “bedside manner.” </strong>The quality of social interaction significantly impacts diagnosis, treatment, and patient outcomes. The AI system should signal that patients are believed and taken seriously.</li><li><strong>Elicit information conversationally, not interrogatively.</strong> The AI system should use scaffolding to guide patients toward relevant details without making them feel cross-examined or blamed.</li><li><strong>Account for context and culture.</strong> Language, pacing, and assumptions should adapt to cultural norms and individual comfort levels to avoid misinterpretation or exclusion.</li></ul><p><em>Design for How Information Is Returned</em></p><p>How information is delivered often matters as much as what is delivered.</p><ul><li><strong>Adapt explanations to patient preferences.</strong> Some patients want clinical detail; others need plain-language reassurance. AI systems should learn and respect these preferences over time.</li><li><strong>Design Explicitly for High-Risk Moments. </strong>When findings are ambiguous or potentially alarming, the AI system must demonstrate sensitivity. This may include offering emotional support, connecting to resources or escalating to a human doctor.</li><li><strong>Calibrate trust, not maximize it.</strong> The AI system should actively guard against overconfidence by making uncertainty visible, surfacing limitations, and encouraging human consultation when appropriate.</li></ul><h3><strong><em>Operationalizing Accountability After Launch</em></strong></h3><p>Because AI systems are probabilistic, they require ongoing supervision after launch. Baseline evaluation areas include:</p><ul><li>Overall correctness and guardrail adherence</li><li>Hallucination detection such as fabricated studies or protocols</li><li>Tone and humaneness, using frameworks such as <a href="https://github.com/buildinghumanetech/humanebench">Human Friendliness Evaluation</a></li></ul><p>Use evals to find behavioral signals that the AI system is delivering on the promises above as serving as an empathetic doctor. Fine-tune to ensure the relationship doesn’t degrade over time.</p><h3><strong>Dr. AI: Patient tool</strong></h3><h4><strong>Design Considerations for this Model</strong></h4><p>Because this model relies on general-purpose AI applied to healthcare, the design considerations listed in the <a href="https://www.microsoft.com/en-us/ai/principles-and-approach">Microsoft Responsible AI </a>are most relevant. But the stakes are higher because there is no human doctor in the loop, as the AI system plays the role of a knowledgeable but medically uncredentialed friend. General tools should remind patients about their limitations. For example, Gemini shows users this message:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7BaXf7rhJLHUbYga90cmJQ.png" /></figure><p>In the future, makers of these generalized systems could consider helping patients pull together and organize all their health information that they shared during chats. They could isolate it from other data about the user and ensure a higher level of security, while at the same time delivering the benefit of the patient feeling like they are “seeing the same doctor” where they do not need to repeat information and creating longitudinal care that is key to patient-doctor relationships.</p><h4><strong>Monitoring &amp; Evals: What to Measure</strong></h4><p>These foundation models should continue to monitor the guardrails around medical information, including:</p><p><strong>Boundary Integrity:</strong> Monitor whether the system stays within its “knowledgeable friend” role rather than drifting toward diagnostic authority with definitive diagnostic statements, specific treatment prescriptions, or advice that contradicts standard care protocols.</p><p><strong>Refusal Metric:</strong> Measure whether AI knows when not to answer. Track events such as refusal rate for questions requiring licensure, escalation prompting when symptoms suggest emergency care, false confidence situations when the AI should say “I don’t know” but doesn’t</p><p><strong>Information Quality:</strong> Track hallucination rates for medical facts, especially fabricated studies, incorrect drug interactions, or outdated treatment protocols. Consider implementing citation requirements for clinical claims.</p><h3><strong>The Invisible Assistant: Clinician tools to enhance patient care</strong></h3><h4><strong>Design Considerations for this Model</strong></h4><p>Since the patient isn’t using the tool directly, they might not realize it’s there. The core challenge is how do you ethically inform patients about AI they never directly see? Also, if the AI system is involved in diagnosing, prescribing, or treating, then a patient has a right to know who (or what) is guiding their care.</p><p><strong><em>Consent by design</em></strong></p><p>As <a href="https://hai.stanford.edu/news/how-is-ai-changing-your-doctor-visit">Dr. Nigam Shah from Stanford</a> points out, the design of the AI system must enable the <em>doctor</em> to have the consent conversation. The user experience should make it easy for them to answer patient questions such as: “Will a human review the AI-generated notes for accuracy?” and “Can I control whether my de-identified data is used for training AI models?”</p><p><strong><em>Seamless opt out</em></strong></p><p>Design for the patient who says “No.” If a patient declines an AI intervention, the AI system should have a capability for the doctor toggle the system off without disrupting the flow of the visit.</p><p><strong><em>Transparency and appeals</em></strong></p><p>Following the <a href="https://lightcollective.org/wp-content/uploads/2024/06/Collective-AI-Rights-For-Patients-v_2.0.pdf">Collective AI Patient rights</a> framework, build capabilities for patients to:</p><ul><li>Access AI-generated outputs about their care (notes, risk scores, recommendations)</li><li>Understand if and how AI influenced clinical decisions</li><li>Request human review or second opinions on AI-based recommendations</li></ul><p>Ensure that there is a user experience for the patient to access this information. Consider when and how in the user journey it will be available to them.</p><p>SIDE NOTE: In early January 2026, the FDA issued guidance on distinguishing when clinical decision support software is regulated as a medical device or not. Jacob Reider offers a <a href="https://docnotes.net/2026/01/07/clinical-decision-support/">concise summary</a>. As this regulation expands what can operate without FDA oversight, it becomes even more important for builders to responsibly keep their AI systems supporting rather than shaping decisions.</p><h4><strong>Monitoring &amp; Evals: What to Measure</strong></h4><p>In this model, success is measured by the quality of the care ultimately delivered to the patient. Provide information to healthcare organization clients data for their governance rubrics. For example, some of the relevant entries from the rubric that Enterprise Chief Digital Transformation Officer, <a href="https://www.linkedin.com/in/crystalbroj/">Crystal Broj </a>and her team at the Medical University of South Carolina use, include:</p><ul><li>Effectiveness &amp; Performance. Metrics/KPIs Tracked: (accuracy, sensitivity, specificity, throughput, adoption rates, etc.)</li><li>Bias, Fairness &amp; Ethics</li><li>Diversity of Training Data: Did the vendor attest to using representative and diverse datasets (age, gender, race, socioeconomic status, geography)?</li><li>Bias Mitigation Strategies: What processes are in place to monitor and mitigate algorithmic bias?</li><li>Equitable Outcomes: Have results been reviewed across demographic groups for fairness?</li><li>Audit Trail: Is there a way to see how decisions/recommendations were generated?</li><li>Transparency to Users: Are doctors/patients aware AI is being used?</li></ul><h3><strong>The Supervised Extender (Clinician-Supervised AI)</strong></h3><h4><strong>Design Considerations for this Model</strong></h4><p>When designing for this model, the focus is on creating a clear sandbox for the AI system: defining where it has autonomy, where it must stop, and how it escalates beyond its boundaries. In other words, implement similar guardrails that provide human support staff with autonomy and agility, and seamless integration when escalating to the doctor.</p><p><strong><em>Define the boundary</em></strong></p><p>Patients need to have a mental model of the AI system’s role as support staff and the doctor’s role as a supervisor. Speak to the policies you have in place to safeguard patients and the role that they have for human doctors to stay in the loop. For example, <a href="https://ubiehealth.com/">Ubie</a> has caveats about the information that it provides, delineating that it is not taking the place of a doctor: “The Ubie AI Symptom Checker is a service intended for the provision of information and does not constitute medical practice by physicians or healthcare professionals.” Alternatively, articulate what parts of care the AI does cover. <a href="https://ada.com/">Ada</a> carves out its role in the healthcare journey as “pre-diagnosing” that is, it “demystifies healthcare navigation, helping people understand what options exist to improve their health based on their conditions and uncover programs, benefits, or treatments that could save them money, improve outcomes, and guide them to the right next step.”</p><p><strong><em>Plan for the Handoff</em></strong></p><p>Consider the details of the handoff journey when the AI system reaches the limit of its scope of practice: What is the user experience of when the AI escalates to a human, and after the escalation with a human doctor in the mix? How does the patient know when they are speaking to the AI system versus the human? What is the user experience of when it’s the patient who wants to escalate?</p><p>Pay attention to the relationship and communication that the patient has had with the AI system, so patients understand what the human doctor already knows and sees. For example, does the doctor read every single chat log or only the summary?</p><p><strong><em>Offer an Opt-out</em></strong></p><p>As highlighted in the <a href="https://lightcollective.org/patient-ai-rights/"><em>AI Rights for Patients</em>,</a> and as we saw in the Extender model, patients must have a clear mechanism to bypass the AI system and request a human, even if it means a longer wait time. Or even opt-out of interacting with the AI system if that is their preference, anytime in their journey.</p><h4><strong>Monitoring &amp; Evals: What to Measure</strong></h4><p>Because the supervising doctor retains liability, evaluation must focus on whether the AI reliably operates within its delegated role, escalates appropriately, and preserves trust during transitions.</p><p><strong>Boundary Integrity:</strong> Does the AI system remain within its defined clinical boundaries? Monitor for adherence to defined clinical guardrails.</p><p><strong>Relationship Quality:</strong> How well does the AI system maintain trust during routine interactions. Rather than measuring “empathy” separately, evaluate whether the AI’s communication avoids overconfidence or emotional manipulation, and supports, rather than substitutes for, the patient-doctor relationship.</p><p><strong>Escalation Quality:</strong> Recognize critical situations and escalate them immediately. Ensure the AI system explains to the patient why escalation is happening. Verify that the system is not too sensitive with unnecessary handoffs, nor too loose with missed escalations. Track both the false positive escalations that erode trust and the more dangerous false negative failures.</p><p><strong>Handoff Continuity:</strong> When escalation occurs, does the system preserve coherence across the AI-human boundary? Assess both sides of the transition:</p><ul><li>For clinicians: Is the context sufficient, accurate, and concise?</li><li>For patients: Is it clear when responsibility shifts from AI to human?</li><li>For the relationship: Does the handoff feel seamless or jarring?</li></ul><h3><strong>AI is the third and equal participant</strong></h3><h4><strong>Design Considerations for this Model</strong></h4><p>In this model, the AI system is no longer subordinate but a partner in patient-centered care. The system AI enhances collaboration and shared decision making by surfacing data, translating jargon, catching blind spots, and advocating for the patient. Explore how to balance what the AI does best at key moments, and what the doctor does best, and how this dynamically and fluidly shifts during interactions with a patient.</p><p><strong>Co-Design is Mandatory</strong>: As PFCCpartners’s <a href="https://www.pfccpartners.com/actframework">ACT Framework</a> emphasizes, patients and doctors must co-design the AI system’s social presence. What role should the AI play? Translator? Advocate? Skeptic? Some combination? Define the interaction behaviors: Is the AI an active listener that surfaces data only when prompted, or a proactive partner that interrupts when it detects a potential contraindication or a missed social determinant of health?</p><p><strong>Define decision-making mechanics</strong></p><p>Consider how your AI system will support, alongside the doctor, each aspect of the participatory decision-making (PDM) model</p><ul><li>surface unvoiced patient concerns like medication costs or fear of procedures</li><li>understand the patient’s experience and expectations</li><li>build the partnership with empathy and trustworthiness</li><li>convey the evidence and uncertainties in a way that makes sense to the patient</li><li>present the recommendations and the rationale behind them</li><li>check the patient’s understanding and agreement</li></ul><p><strong>Preserve the Psychological Signals</strong></p><p>The Oxford Internet Institute warns that AI tends to prioritize “measurable quantities.” In a Triad, the design must ensure the conversation doesn’t collapse into a data review. The AI system should know when to step back and let human connection take center stage.</p><p><strong>Manage power dynamics transparently</strong></p><p>Since authority shifts in real-time, patients and doctors must understand the ground rules:</p><ul><li>When does the AI have definitive information versus provisional suggestions?</li><li>If the doctor and AI disagree, who makes the final call?</li></ul><h4><strong>Monitoring &amp; Evals: What to Measure</strong></h4><p>In a partnership, measure the team’s performance and cohesion, not just individual contributions.</p><p><strong>Boundary Integrity: </strong>Ensure that patients can distinguish between evidence-based recommendations and AI-generated suggestions. Evaluate whether the AI system contributes insights the human doctor missed? (e.g., rare drug interaction, recent relevant research). Track the “delta” between doctor alone versus doctor + AI outcomes.</p><p><strong>Relationship &amp; Communication Integrity: </strong>Assess if patients feel heard by the <em>team</em>, not just individual members. Do patients perceive the AI system and doctor as aligned partners or competing authorities? Were “unmeasurable” factors (quality of life, patient preferences, social context) given appropriate weight in decisions?</p><h3><strong>Conclusion: Designing for the Relationship</strong></h3><p>AI is re-shaping relationships, not just workflows so product leaders need to design AI systems suitable for their role within the patient-doctor relationship. Whether you’re building a symptom checker or a sophisticated clinical partner, success depends on three commitments:</p><p><strong>1. Name your model explicitly.</strong> Be clear internally and externally about which relationship structure you’re creating.</p><p><strong>2. Establish firm guardrails. </strong>Define the AI’s scope, escalation triggers, and accountability mechanisms.</p><p><strong>3. Measure rigorously and continuously.</strong> Monitor not just for technical performance, but for relationship integrity.</p><p>Regardless of which model you choose, the healing relationship remains the north star. Patients come to healthcare in moments of vulnerability, seeking not just treatment but understanding, dignity, and hope. The most successful AI systems will be those that honor that vulnerability, clarify the machine’s boundaries, and ultimately strengthen the human bonds at the center of medicine.</p><h3>References &amp; Further Reading</h3><h4>The Evolving Patient-Provider Relationship</h4><ul><li><a href="https://www.google.com/search?q=https://journalofethics.ama-assn.org/article/are-there-blueprints-building-strong-patient-physician-relationship/2021-01">Are There Blueprints for Building a Strong Patient-Physician Relationship?</a> — <em>AMA Journal of Ethics</em></li><li><a href="https://www.google.com/search?q=https://www.annfammed.org/content/7/4/353">The Patient–Doctor Relationship: A Synthesis of the Qualitative Literature on Patients’ Perspectives</a> — <em>Annals of Family Medicine</em></li><li><a href="https://pubmed.ncbi.nlm.nih.gov/16573302/">Toward a conceptual reexamination of the patient-physician relationship in the healthcare institution for the new millennium </a>— <em>Journal of the National Medical Association</em></li><li><a href="https://newsnetwork.mayoclinic.org/discussion/9-qualities-of-a-great-doctor/">9 Qualities of a Great Doctor</a> — <em>Mayo Clinic News Network</em></li><li><a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC11997521/"><em>Online Health Information–Seeking in the Era of Large Language Models: Cross-Sectional Web-Based Survey Study</em></a> <em>— Journal of Medical Internet Research</em></li></ul><h4>Ethics &amp; Responsible AI Principles</h4><ul><li><a href="https://www.microsoft.com/en-us/ai/responsible-ai">Microsoft Responsible AI Principles</a> — <em>Microsoft</em></li><li><a href="https://www.google.com/search?q=https://lightcollective.org/ai-rights/">AI Rights for Patients</a> — <em>The Light Collective</em></li><li><a href="https://www.humanetech.com/">Building Humane Tech</a></li><li><a href="https://www.humanetech.com/">Center for Humane Technology</a></li><li><a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7354474/">Ethical and Regulatory Challenges of AI Technologies in Healthcare: A Narrative Review</a> — <em>Expert Review of Medical Devices</em></li><li><a href="https://www.pfccpartners.com/actframework">ACT Framework </a><em>— PFCCpartners</em></li><li><a href="https://www.nature.com/articles/s42256-025-01113-8?mkt_tok=NTcwLVJDSC03NTgAAAGe0Sm8BoypU9pMZ3yl_3Tb18Xf1K3tmc0Wqi9Eln3brV5UdpNmMALHTgB_XuymllzciZetLD32b1CNyumOYTY38LfSeQcPe83lXhjPYOs">Language models cannot reliably distinguish belief from knowledge and fact</a><em> — Nature</em></li></ul><h4>Impact of AI on Clinical Dynamics</h4><ul><li><a href="https://www.google.com/search?q=https://www.oii.ox.ac.uk/news-events/news/report-on-the-impact-of-artificial-intelligence-on-the-doctor-patient-relationship/">Report on the Impact of Artificial Intelligence on the Doctor-Patient Relationship</a> — <em>Oxford Internet Institute</em></li><li><a href="https://www.google.com/search?q=https://www.jmir.org/2023/1/e43444/">Existing Barriers Faced by and Future Design Recommendations for Direct-to-Consumer Health Care AI Apps: Scoping Review</a> — <em>Journal of Medical Internet Research</em></li></ul><h4>Media &amp; Practitioner Perspectives</h4><ul><li><a href="https://www.google.com/search?q=https://tradeoffs.org/2024/02/22/ai-diagnosis/">Should I Trust AI to Diagnose Me?</a> — <em>Tradeoffs Podcast</em></li><li><a href="https://www.google.com/search?q=https://pear.vc/podcasts/pear-healthcare-playbook-travis-zack-of-openevidence/">Pear Healthcare Playbook Podcast: Travis Zack of OpenEvidence</a> — <em>Pear VC</em></li><li><a href="https://www.google.com/search?q=https://www.newyorker.com/science/annals-of-medicine/seven-years-later-what-ai-has-and-hasnt-changed-about-medicine">Seven Years Later: What AI Has — and Hasn’t — Changed About Medicine</a> — <em>Lloyd Minor on LinkedIn</em></li><li><a href="https://hai.stanford.edu/news/how-is-ai-changing-your-doctor-visit">How is AI Changing Your Doctor Visit?</a> <em>— Stanford Human Centered Artificial Intelligence</em></li><li><a href="https://www.nytimes.com/2025/11/16/well/ai-chatbot-doctors-health-care-advice.html">Empathetic, Available, Cheap: When A.I. Offers What Doctors Don’t</a> <em>— The New York Times</em></li><li><a href="https://ai.gopubby.com/ai-is-coming-for-healthcare-and-it-terrifies-me-30dc2583e87d">AI Is Coming for Healthcare, and It Terrifies Me</a> <em>— Iryna Nozdrin on Medium</em></li><li><a href="https://www.8vc.com/resources/ai-healthcare-vision">A Vision for Healthcare AI in America </a>— <em>8VC</em></li><li><a href="https://docnotes.net/2026/01/07/clinical-decision-support/">Clinical Decision Support</a> — <em>Jacob Reider blog</em></li><li><a href="https://www.amazon.com/Deep-Medicine-Artificial-Intelligence-Healthcare/dp/1541644638">Deep Medicine</a> — <em>Eric Topol’s book</em></li></ul><p><strong>Evals Resources</strong></p><ul><li><a href="https://www.lennysnewsletter.com/p/why-ai-evals-are-the-hottest-new-skill">Why AI evals are the hottest new skill for product builders | Hamel Husain &amp; Shreya Shankar (creators of the #1 eval course)</a> — <em>Lenny’s Podcast</em></li><li><a href="https://www.lennysnewsletter.com/p/building-eval-systems-that-improve">Building eval systems that improve your AI product</a> — <em>Lenny’s Newsletter</em></li><li><a href="https://github.com/buildinghumanetech/humanebench">Human Friendliness Evaluation</a> — <em>Building Humane Tech</em></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ba8e89fa2aaf" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Does Your AI Need a Hippocratic Oath? Four Emerging Models of Clinical Care]]></title>
            <link>https://medium.com/@mberkovich/does-your-ai-need-a-hippocratic-oath-four-emerging-models-of-clinical-care-56be7c907543?source=rss-35410c649a3e------2</link>
            <guid isPermaLink="false">https://medium.com/p/56be7c907543</guid>
            <category><![CDATA[user-centered-design]]></category>
            <category><![CDATA[digital-health]]></category>
            <category><![CDATA[product-development]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Marianne Berkovich]]></dc:creator>
            <pubDate>Tue, 06 Jan 2026 21:36:02 GMT</pubDate>
            <atom:updated>2026-01-15T16:33:53.035Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Four models of the patient-doctor-AI relationship" src="https://cdn-images-1.medium.com/max/1024/1*WuqTlMEHFbXDYcL7VtTZRA.png" /></figure><p>AI and chatbots have moved into nearly every corner of our lives, and lately they’ve been entering the exam room. We have used software in clinical settings for decades, but AI is different. Traditional software was a tool — passive, explicit, and controlled. But AI <em>feels</em> human. It adopts a voice, infers patterns, and speaks authoritatively and empathetically. This anthropomorphism takes aim at healthcare’s north star: the healing relationship itself.</p><p>This framework helps product leaders name the relationship structures they are implicitly creating during clinical encounters. Four distinct models are emerging that aren’t product categories, but relationship structures:</p><ul><li><strong>Dr. AI:</strong> The patient uses AI independently</li><li><strong>The Invisible Assistant:</strong> The doctor uses AI in the background</li><li><strong>The Supervised Extender:</strong> The AI interacts with the patient under doctor supervision</li><li><strong>The Triad:</strong> The doctor, patient, and AI collaborate openly</li></ul><p>The objective is to highlight patients’ expectations of the patient-doctor-AI relationship that will continue to be relevant in the future even as the real world example companies change and as regulations evolve.</p><p>This article has two parts. Part 1 describes the four models in the framework and examples of companies that are using that model currently. Part 2 turns to making the framework actionable: what builders of a new and powerful technology should be mindful of, to create a positive and responsible patient experience.</p><h3><strong>Dr. AI: Patient tool</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/626/1*j3-ZshC2eqGBVSSporoRXg.png" /></figure><p>Let’s start with the model that follows the path of least resistance for patients: consulting “Dr. AI.” Like Dr. Google before it, Large Language Models (LLMs) have become the de facto starting point for medical advice. People are already heavily reliant on this model, using LLM chatbots for treatment (46%) and self-diagnosis (23%) according to a late 2024 Northeastern University survey.</p><h4><strong>The Relationship Dynamic</strong></h4><p>The primary relationship is between the <strong>patient and the AI </strong>in this model. AI acts as a knowledgeable but not medically-credentialed friend since <a href="https://www.nature.com/articles/s42256-025-01113-8?mkt_tok=NTcwLVJDSC03NTgAAAGe0Sm8BoypU9pMZ3yl_3Tb18Xf1K3tmc0Wqi9Eln3brV5UdpNmMALHTgB_XuymllzciZetLD32b1CNyumOYTY38LfSeQcPe83lXhjPYOs">LLMs can’t tell the difference between fact and belief</a>. AI systems offers convenient, free, and judgement-free care with a veneer of credibility. Patients must exert their own medical judgement to discern the quality and relevance of the output. <em>Caveat emptor </em>is the hallmark of this model because there is no human safety net, though it is a “good enough” starting point.</p><h4><strong>Examples in the Wild</strong></h4><p>General purpose LLMs such as ChatGPT, Gemini and Co-pilot act as broad knowledge engines for self diagnosis and treatment, combined with patience and empathy. “ChatGPT has all day for me — it never rushes me out of the chat,” said a woman in the New York Times article, whose title captures the main benefits: <a href="https://www.nytimes.com/2025/11/16/well/ai-chatbot-doctors-health-care-advice.html">Empathetic, Available, Cheap: When A.I. Offers What Doctors Don’t</a>.</p><h4>The Product Challenge</h4><p>Dr. AI feels empowering precisely because it is independent, continuous, and always available. But that same independence removes the safety net. The challenge for builders is insulation without infantilization: How do you reduce harm, surface uncertainty, and encourage appropriate escalation without undermining the patient’s agency?</p><h3><strong>The Invisible Assistant: Clinician tool to enhance patient care</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/859/1*vAShi4h8DDevB0f1jCJ5_A.png" /></figure><p>The AI is primarily in the background in this model; the patient may never interact with the AI system directly, but they feel its effects. The model aspires to “restore human connection at the heart of healthcare” as <a href="https://www.nabla.com/whitepapers/ai-whitepaper">Nabla’s whitepaper </a>says, by removing the administrative burden and allowing the doctor’s focus to be entirely on the patient.</p><h4><strong>The Relationship Dynamic</strong></h4><p>The relationship remains the traditional <strong>doctor-patient dyad</strong>, but the AI boosts the capabilities. The AI system joins an invisible infrastructure — lab techs, pathologists, patient safety officers, etc that patients may not see, but are critical for care delivery. However, when AI systems working similarly behind the scenes begin diagnosing, prescribing, or treating, a patient has a right to know what is guiding their care.</p><h4><strong>Examples in the Wild</strong></h4><p>This model is seeing rapid enterprise adoption in two main areas:</p><ul><li><strong>Ambient Scribes</strong> (<a href="https://www.abridge.com/">Abridge</a>, <a href="https://www.suki.ai/">Suki</a>, <a href="https://www.augmedix.com/">Augmedix</a>, <a href="https://www.nabla.com/">Nabla</a>, among many others) listen to patient visits and automatically generate clinical notes, billing codes, and task lists.</li><li><strong>Clinical Decision Support</strong> (<a href="https://www.openevidence.com/">OpenEvidence</a><a href="https://www.doximity.com/gpt">,</a> <a href="https://www.wolterskluwer.com/en/solutions/uptodate/ai-clinical-decision-support">UptoDate</a>) offers real-time evidence lookup to support diagnostic reasoning, medication selection, and treatment guidelines.</li></ul><h4>The Product Challenge</h4><p>This model succeeds by disappearing. But when AI meaningfully shapes diagnosis, treatment, or documentation, invisibility becomes an ethical and legal liability. The challenge is transparency: How do you disclose AI’s role in care without reintroducing cognitive or emotional burden into the clinical encounter?</p><h3><strong>The Supervised Extender: Clinician-Supervised AI</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/868/1*o68qZn7zIFkWXLKfvswotw.png" /></figure><p>Once the AI system begins speaking directly to the patient on behalf of a doctor, we enter a fundamentally different relational dynamic.</p><p>Unlike the Dr. AI model, this AI system acts with oversight of a provider. The AI system handles day-to-day interactions (checking in during physical therapy, tracking mood, answering FAQs, following standardized protocols), but when the patient’s needs exceed the AI’s scope, the AI system acknowledges its limits and triggers an escalation to the human supervisor.</p><h4><strong>The Relationship Dynamic</strong></h4><p>The primary relationship is still between the <strong>patient and the AI,</strong> but the defining distinction is accountability. In the Dr. AI model, the patient is accountable for using the advice. In the Supervised Extender model, the doctor is legally/ethically accountable for the AI system’s output.</p><p>For the patient, this relationship feels similar to interacting with clinical support staff at the doctor’s office. Support staff collect and organize data, execute established tasks under protocols, triage, and administer what a doctor has ordered. They perform emotional and relational labor as they are seen by patients as allies or peers who are safer, less judgemental, more approachable, and less rushed than a doctor. AI system capabilities excel at both taking on tasks that involve data organization and simulating a deferential and supportive ally, distinct from the doctor.</p><h4><strong>Examples in the Wild</strong></h4><p>This model dominates digital health sectors where scale can be achieved by carving out tasks:</p><ul><li><strong>Symptom Checkers:</strong> (<a href="https://ada.com/">Ada</a>, <a href="https://ubiehealth.com/">Ubie</a>, <a href="https://about.mydoctorfriend.ai/">My Doctor Friend</a>) These are purpose-built, often regulated tools.</li><li><strong>Physical Therapy:</strong> (<a href="https://swordhealth.com/ai">Sword Health</a>, <a href="https://www.hingehealth.com/">Hinge Health</a>) These platforms use AI to monitor movement and provide real-time feedback.</li><li><strong>Mental Health:</strong> (<a href="https://www.wysa.com/">Wysa</a>) Chatbots that provide 24/7 mental health support, such as Cognitive Behavioral Therapy (CBT).</li><li><strong>Natural Language Pre-visit Intake:</strong> (<a href="https://healthyoda.ai/">Health Yoda</a>, <a href="https://www.insighthealth.ai/pre-visit-intake">Insight Health</a>, <a href="https://www.sully.ai/">Sully.ai</a>) These chatbots capture a patient’s symptoms ahead of their meeting with the doctor.</li></ul><p>A notable boundary case is the “AI Doctor” like <a href="https://www.doctronic.ai/">Doctronic</a> and <a href="http://cody.md/">Cody.md</a>. These platforms attempt to replace the visit entirely. While seemingly similar to the Dr. AI model, these solutions are supervised by doctors, so are included under this model.</p><h4>The Product Challenge</h4><p>What does a humane handoff feel like at the moment the AI steps aside? This model hinges on escalation when the AI reaches the edge of its scope. The challenge is designing escalation as a relationship moment, not a technical one: How does the AI build a relationship with the patient while also communicating limits, urgency, and continuity so the patient feels supported?</p><h3><strong>The Triad: AI as equal partner</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/902/1*ThS_k8dBpDsNWy_SSkTz-g.png" /></figure><p>Finally, we arrive at the most complex and transformative model: The Triad. In this model, the linear doctor-patient relationship expands into a triangle. Both the doctor and the patient are interacting with the AI system. The AI system is an active, equal participant in the room, contributing unique capabilities. For example, it can act as a specialist with infinite memory that excels at pattern recognition, real-time synthesis of vast datasets, and real time medical jargon translator, as well as someone who has unlimited time and patience to spend with a patient.</p><p>It is truly a team effort in which human and AI clinicians are vital. Dr. Lloyd Minor, Dean of Stanford University School of Medicine &amp; Vice President for Medical Affairs recently <a href="https://www.linkedin.com/pulse/seven-years-later-what-ai-has-hasnt-changed-medicine-lloyd-minor-z53ic/">said</a>, “We still need clinicians who can work across disciplines, interpret data in context, and guide patients through uncertainty, all while ensuring that the outputs of AI align with the enduring values of medicine. Teamwork, judgment, and compassion remain as essential as ever.”</p><h4><strong>The Relationship Dynamic</strong></h4><p>Power is shared among the two humans and the AI system. Whereas the Extender model is hierarchical in order to improve efficiency and delegation, the Triad model is about augmentation and collaboration in a flat hierarchy. The AI system can play one or more roles in the relationship that capitalize on the things that AI is good at, for example:</p><ul><li><strong>Patient advocate: </strong>helping the patient articulate their story, their needs, etc. before, during and after the appointment.</li><li><strong>Real-time Assistant: </strong>synthesizing disparate information and surfacing relevant research. As Dr. Dhruv Khullar on the <a href="https://tradeoffs.org/2025/10/30/should-i-trust-ai-to-diagnose-me/">Tradeoffs podcast</a> describes it: helping with “wayfinding along the diagnostic journey,” alerting clinicians to recent studies or proposing next tests.</li><li><strong>Collaborator: </strong>an AI system could co-create a care plan in which the AI brings together all the patient data and drafts a plan, and the doctor and patient imbue it with the patient’s values and preferences.</li><li><strong>Coach</strong> <strong>and</strong> <strong>Skeptic: </strong>analyzing the interaction to ask, “Did we miss anything?” and coaching the doctor after the appointment.</li></ul><p>The AI system feels like a peer in the triad that combines the best of human clinical staff and technology: quietly essential, operationally flawless, and clinically humble. This model explores the possible ways that this third participant with certain superpowers is added to the mix of a fallible and quirky relationship dynamic that is formed between two particular humans (patient and doctor).</p><h4><strong>Examples in the Wild</strong></h4><p>This model is still emerging as there are legal and regulatory hurdles in the current environment; right now it’s not legal for AI to prescribe, treat, diagnose, and refer without an appropriate medical license in the US, and a patchwork of state regulations are being enacted. However, we see an early signal of what’s to come in Therapy Lab’s Sage. On the <a href="https://www.therapylab.com/therapy-with-ai-support">Therapy Lab’s site</a>, Sage is described as “an AI therapy companion that optimizes your time in therapy, like a smart companion who joins you in a session and offers personalized input between sessions. Unlike other AI tools, Sage is specifically designed for cognitive behavioral therapy. It’s not a replacement for your therapist — it’s part of your team, working with your clinician and available 24/7.”</p><p>Future implementations might range from AI trained in a specific doctor’s style to standalone companions with their own personalities, but the core principle of shared partnership remains constant.</p><h4>The Product Challenge</h4><p>How do three parties co-design trust when power is fluid? In the triad, authority is no longer linear. The AI may correct the clinician. The patient may defer to the system. Expertise shifts in real time. The challenge is intentional power design: How do you ensure collaboration without confusion, augmentation without overreach, and shared decision-making without eroding human judgment? Power and authority are shared, and roles must be explicitly designed.</p><h3><strong>Designing for the Relationship</strong></h3><p>In <a href="https://medium.com/@mberkovich/designing-ai-for-the-clinical-relationship-ba8e89fa2aaf">Part 2</a>, we will convert this framework into a toolkit. For each model, we offer design principles about the nature of the relationship when building, and then suggestions for what to be mindful of when continuously evaluating the AI system after it launches.</p><p><a href="https://medium.com/@mberkovich/designing-ai-for-the-clinical-relationship-part-2-ba8e89fa2aaf">CONTINUE TO PART 2: Designing AI for the Clinical Relationship</a></p><h4>Disclosure &amp; Acknowledgments</h4><p>I used AI to research examples of companies using each model, generate the illustrations of the models, review drafts and improve the overall structure of the article, as well as to wordsmith particular sections.</p><p>Thank you to Micah Alpern, Dan Russell, Preston Tollinger, and Matt Williams, for your very human and humane input and feedback!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=56be7c907543" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Day in the Life of a First User Researcher]]></title>
            <link>https://medium.com/@mberkovich/day-in-the-life-of-a-first-user-researcher-5d48c82304ca?source=rss-35410c649a3e------2</link>
            <guid isPermaLink="false">https://medium.com/p/5d48c82304ca</guid>
            <category><![CDATA[ethnography]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[user-research]]></category>
            <category><![CDATA[ux]]></category>
            <category><![CDATA[usability-testing]]></category>
            <dc:creator><![CDATA[Marianne Berkovich]]></dc:creator>
            <pubDate>Wed, 19 Jun 2019 20:17:48 GMT</pubDate>
            <atom:updated>2019-06-19T20:21:25.392Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*eOKEH4gtsc-VlAiy" /><figcaption>Photo by <a href="https://unsplash.com/@aaronburden?utm_source=medium&amp;utm_medium=referral">Aaron Burden</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p><em>“Mighty oaks from little acorns grow.”</em></p><p>And mighty teams start with one person. At Glooko, I am that one person in user experience research. (<a href="https://www.glooko.com/">Glooko</a> is diabetes management platform which includes a mobile app for people with diabetes and an online dashboard for health care providers.) My goal is to eventually grow the user research practice from one little acorn to a mighty oak team.</p><p>To make this future a reality, I’m cultivating (pun intended) a strong root system that is both <em>deep</em> — involving a variety of user research methods throughout the software development lifecycle phases — and <em>wide</em> — working with many teams across the company.</p><p>On a particularly fertile (ha ha) day a few weeks ago — let’s call it Root Day — I found myself tending the three vital roots which are intertwined (pun-y again) with the questions that drive the three main phases of the software development lifecycle:</p><ul><li>What should we build?</li><li>How should we build it?</li><li>Did we build it right?</li></ul><p>Note that each start-up’s needs are different, so these are some examples of the root cultivation activities that I’ve been doing that fit Glooko’s unique and current needs.</p><h3><strong>“What should we build?” Root</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*gl0ZSbYsN9Thdu7U" /></figure><p>This root is about getting involved in the early exploratory stages of product development. The question “What should we build?” points to understanding users’ challenges, needs, and context. It’s the phase where the team looks for opportunities of new features or products to build.</p><p><strong>Prepping for Type 1 Ethnography workshop</strong></p><p>In the past few months we’ve been focusing on getting a deeper understanding of what it’s like to manage diabetes on a daily basis for people with Type 1 diabetes. I had led a series of in-home ethnographic style interviews with people who have Type 1 diabetes; I had folks from Marketing, Product Managers, Engineering, Data Science, and UX join me as notetakers and videographers on the visits. As the first step in bringing the lives of our participants to everyone at Glooko, I scheduled a one-day workshop to tell their stories. I assigned a Glooko employee to each participant; they were asked to watch the video, review the notes, photos, etc from each session and write a summary using a template I provided. (See my story <a href="https://medium.com/@mberkovich/three-ways-to-build-empathy-when-working-in-health-tech-b2b0e9ec03a6">“Three Ways to Build Empathy when working in Health Tech” </a>for more detail)</p><p>On Root Day, I was taking care of the logistics for the workshop: creating a summary template, making video clips, and printing pictures of the participants and of key artifacts from their visits.</p><p>Later, at the workshop, we watched a video clip of each participant so everyone could get a sense of who they are, and the Glooko employee told the participant’s story. Everyone else in the room jotted down observations on sticky notes. By the end of the day, we were surrounded by our insights and the beginnings of significant patterns about our participants and their needs, which will lead to new feature ideas that we should build. All in all, about 30 people (about a ⅕ of the company!) took part in the visits themselves, wrote summaries, or took part in the workshop — including our CEO.</p><p><strong>Healthcare Provider Interview</strong></p><p>On the clinician side of the company, I was working with a Product Manager and a Designer to explore healthcare providers’ needs for a potential new feature. We had already conducted a series of interviews to understand the broad needs, and now wanted to get more specific using a series of wireframes to illustrate the experience. I worked with them to:</p><ul><li>Articulate the research questions</li><li>Draft the discussion guide</li><li>Role-play with the Designer so she’d feel prepared to conduct the interviews</li></ul><p>On Root Day, since we had arranged this project so they could conduct the interviews themselves, I was playing a back-up role in one of these interviews — just asking additional questions rather than conducting the full interview. With my mentoring and support, the Product Manager and Designer were able to get the feedback they needed to understand what we should build for healthcare professionals.</p><h3><strong>“How should we build it?” Root</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*HeO5LuiRYgn5Jl42" /></figure><p>The “How should we build it?” root picks up after the team identifies some opportunities areas, and begins to build the feature. This root is about helping the team evaluate a variety of concepts, diverging on the right solution, and honing that solution until the feature launches.</p><p><strong>Usability testing a mobile app feature</strong></p><p>We were working on a new feature in the mobile app, and two of our Interaction Designers had collaborated on the user experience and were ready to get feedback from users. I worked with them to:</p><ul><li>Articulate the research questions and craft a research plan using the template I had developed</li><li>Pick the recruiting criteria to find the right participants and which screener library questions to use</li><li>Think through the tasks we needed in order to answer those questions and how to chunk the prototype according to the tasks to effectively evaluate them on<a href="https://www.usertesting.com/"> UserTesting.com</a></li><li>Set up the tasks in UserTesting.com</li></ul><p>On Root Day, after individually watching the videos of the user session of the couple of variations of the flow we were testing, we came together to compare notes, to analyze, and to synthesize our findings, and figure out the next steps of how we should build this feature. Additionally, as a result of taking the designers through setting up the tools, I wrote up a handy Marianne’s Guide to UserTesting.com to avoid common pitfalls I had discovered and that could be a useful piece of organizational knowledge.</p><p><strong>Usability Testing an ad campaign</strong></p><p>In preparation for an upcoming big ad campaign, our Marketing Designer had designed ads with two different visual styles. She wanted help with assessing whether users would understand the copy, and which style would be a better fit for the audience that we wanted to attract. I worked with her to:</p><ul><li>Articulate the research questions and craft a research plan using the template I had developed</li><li>Weigh the pros and cons of the tools we could use to evaluate these questions. We decided to do a test on<a href="https://fivesecondtest.com/"> Five Second Test</a> to test initial impressions, and another on<a href="https://www.usertesting.com/"> UserTesting.com</a> for understandability.</li><li>Write the set of tasks for each test</li><li>Pick the recruiting criteria to find the right participants</li></ul><p>On Root Day, we reviewed the tests in Five Second Test and on UserTesting.com environments to make sure they avoided common pitfalls, and to fix any final issues like typos or unclear instructions. Later, the Designer was able to do the analysis on her own, and to identify the elements of the design that were working and not working so she knew which ads she should build.</p><h3><strong>“Did we build it right?” Root</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*pIVOXeJv6Hn8DXz1" /></figure><p>The final root is about validating the decisions we made by listening to users’ reactions, and looking for areas of improvement that could drive the next cycle of building.</p><p><strong>Discussion with Marketing about moving to Quarterly Satisfaction survey</strong></p><p>Glooko has been conducting an NPS survey for the past several years, but given some of the<a href="https://articles.uie.com/net-promoter-score-considered-harmful-and-what-ux-professionals-can-do-about-it/"> shortcomings of NPS</a>, I was advocating our moving to a quarterly survey (and ultimately an in-app feedback mechanism).</p><p>On Root Day, we had a discussion with our Marketing teams in Sweden and the US to understand how they were using the NPS and whether the new proposed solution would meet their needs. By the end of the meeting, we had a clear sense of what the new survey would need to accomplish and I volunteered to write up a proposal to share with Leadership. Since then we’ve adopted the quarterly survey approach and are exploring even more immediate yet not intrusive ways to get feedback from our users.</p><p>By using a variety of user research methods throughout the software development lifecycle, being a coaching resource for every team to engage in user research, and striving to make Glooko as a company even more user-focused, I’m cultivating the ground for where a mighty User Research team can begin to sprout.</p><p>Are you a little acorn researcher at your organization? How are you tending your user research roots?</p><p>Are you a Product leader at your organization considering if it’s time to plant the first seed with a dedicated researcher? How does this jive with your expectations of what that acorn researcher could do for your organization?</p><p>I’d love to hear from you.</p><p>Thank you to Allison Gentes, Amber Hampton Greene, Amy Ng, Ashley Tudor, George Fairbanks, Kim Do, and Robin Jeffries for their feedback on this article!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5d48c82304ca" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Three Ways to Build Empathy when working in Health Tech]]></title>
            <link>https://medium.com/@mberkovich/three-ways-to-build-empathy-when-working-in-health-tech-b2b0e9ec03a6?source=rss-35410c649a3e------2</link>
            <guid isPermaLink="false">https://medium.com/p/b2b0e9ec03a6</guid>
            <category><![CDATA[healthcare]]></category>
            <category><![CDATA[health-technology]]></category>
            <category><![CDATA[user-experience]]></category>
            <category><![CDATA[user-research]]></category>
            <category><![CDATA[diabetes]]></category>
            <dc:creator><![CDATA[Marianne Berkovich]]></dc:creator>
            <pubDate>Mon, 14 May 2018 17:03:53 GMT</pubDate>
            <atom:updated>2018-05-14T17:04:59.525Z</atom:updated>
            <content:encoded><![CDATA[<p>Building something for yourself is easy — you know exactly what you’d like. Building something for others is harder since you can’t assume what works for you will work for others. This is doubly true when the problem that you are trying to solve for others is way outside your own experience, such as when building technology solutions for health issues.</p><p>I suggested three ways for the whole Glooko team to build empathy for our users:</p><ul><li>Experience it yourself</li><li>See for yourself</li><li>Listen to others</li></ul><p>Last summer I joined Glooko as Head of User Research and Customer Insights. <a href="http://www.glooko.com/">Glooko</a> is an online diabetes management platform used by people with diabetes, doctors, and diabetes educators to collect data from devices like blood glucose meters and apps like Fitbit. Glooko brings together the data in order to help see patterns and make better decisions.</p><p>Some of the folks who work at Glooko have diabetes, and some don’t. Being among those who don’t, I decided to try our tools on myself to get some first-hand insight into what it’s like to measure blood glucose (BG) every day, and track food, exercise and “medication.” I hoped that it would give me a very small sense of what daily life is like for people with diabetes, as well as let me get acquainted with the Glooko app. Working with our Certified Diabetes Educator, we planned out how I would follow the journey of someone who was just diagnosed with Type 2 diabetes. She went over what a clinician discusses with someone during their initial diagnosis and taught me how to measure my BG using a blood glucose meter. It took me four tries on get a reading on my first solo attempt, and here’s the evidence:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PRNaGWhYsLMPv_zMwQFAGA.png" /><figcaption>Four tries to get my first blood glucose reading</figcaption></figure><p>For two weeks I tracked my BG twice a day, food, and exercise in the Glooko app. I did it at home, in restaurants, and even when hiking with my dog on dusty trails that made my hands dirty when I needed to do a BG reading. After the two weeks, I shared a summary of my experience in a presentation to the whole company. Afterwards several employees asked me how they might try this for themselves too. I realized that there was a need among many of our employees to learn and to gain empathy for our users — people with diabetes.</p><h3><strong>Experience it yourself: “Trying Glooko” Program</strong></h3><p>As a direct result of my “two weeks with Glooko” experiment, we created the “Trying Glooko” Program. This program helps curious employees get a hands-on view of what it’s like for our users to measure their blood glucose, get onboarded with our system, and view their data in the Glooko mobile app. The program provides them with a getting started guide, a blood glucose measuring kit, and a list of mentors who can show them how to measure their blood glucose.</p><p>Since the program kicked off a few months ago, we’ve had about 10% of the company choose to participate and it’s introduced as part of onboarding for new employees. It’s helped us with a personal awareness of how much effort is involved in tracking, and particularly to see how much effort our users must expend before they can start to see patterns in their data.</p><h3><strong>See for yourself: Go into the field</strong></h3><p>In my role I conduct qualitative user researcher to understand the needs of our users, how they use our app now, and to look for opportunities of how we can make diabetes a little easier to manage through technology. One of the ways we do this is by going out into the field and visiting people in the context where they manage their diabetes — in their homes. Last fall we did a round of these ethnographic-style interviews to better understand the needs of people with Type 2 diabetes. I invited people from the product teams to come and observe, and serve as videographers and notetakers during the session. Product managers, engineers, UX designers, a technical writer, a clinical researcher, and a VP joined me on the visits.</p><p>One of the product managers, a 20-something healthy guy, couldn’t understand why people with diabetes didn’t do more to manage their condition, and asked to come along. During the analysis workshop, he shared that the visit radically changed his perception of what someone with Type 2 diabetes is like. He had been on a visit that turned out to be particularly emotionally difficult — our participant was homebound and suffered from many other health issues besides diabetes. By sitting in her home and listening to her story, he developed a deeper understanding of the struggle of managing a chronic condition, especially when one has to juggle managing other health concerns at the same time.</p><h3><strong>Listen to others: Life with Diabetes Storytelling</strong></h3><p>Since diabetes touches so many lives (<a href="https://www.cdc.gov/diabetes/pdfs/data/statistics/national-diabetes-statistics-report.pdf">30 million people in the US</a>), there are a lot of personal stories that vividly portray what it’s like to live with diabetes. In our monthly “Life with Diabetes” Storytelling series, we invite 2 or 3 employees to share some stories. We encourage both those who have diabetes to tell us about their lives, and those who don’t — but perhaps have someone in their life who does — or even if they’ve read or watched one of the dozens blogs, videos or articles by people with diabetes, to share something.</p><p>We’ve heard stories about getting diagnosed with Type 1 as a child or as a college student. We’ve heard stories about what it’s like to be a mom supporting their child as they are diagnosed and learn to take care of themselves. We’ve heard stories of scary close calls that required a trip to hospital, and triumphs of being able to go back to all the active sports — just this time with an insulin pump. We’ve heard about how frustrating it can be for an adult child to watch an older parent managing Type 2 diabetes, and how to be a supportive friend to someone with Type 1. We’ve heard stories that made us laugh, that brought a tear to our eyes, and having a shared space to tell each other about our lives has brought us a little closer to one another.</p><p>This program is an opportunity for our colleagues to reflect on their own experiences, and to bring their whole self to work by sharing their challenges and triumphs; having a supportive audience to listen has been a boon for our speakers. Not everyone who has diabetes at Glooko chooses to speak, but everyone is invited to bring their own unique perspective to give us a glimpse into daily life with diabetes. Each person who shares a story is presented with a Rufus the bear (a cousin of <a href="http://cc.jdrf.org/bag-of-hope/">JDRF’s Rufus, the Bear with Diabetes</a>) — who usually ends up sitting on people’s desk as a badge of honor.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dqCJnTYrilvZIdrz2NyqKg.png" /><figcaption>Life With Diabetes Storytellers are thanked by Rufus the Bear</figcaption></figure><h3><strong>Get creative</strong></h3><p>These are just three examples of ways to help your team build empathy and feel what it’s like for your users. Get creative in what aspects of your users’ life would be worthwhile for your team to walk a mile in your users’ shoes. For example, a user researcher who works at another diabetes company, 3D printed facsimiles of insulin pumps — with the same weight and form factor as typical insulin pumps, so that her team could experience what it’s like to have something strapped to your body all day long.</p><p>Go out and do something! You will create better products and your users will thank you for it.</p><p><em>Marianne Berkovich is the Head of User Research + Customer Insights at Glooko, and enjoys hiking with her dog on dusty and not so dusty trails</em>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b2b0e9ec03a6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Don’t Build Stuff No One Uses: Lessons from 3 Startups]]></title>
            <link>https://medium.com/@mberkovich/dont-build-stuff-no-one-uses-lessons-from-3-startups-468e3a2a0dd3?source=rss-35410c649a3e------2</link>
            <guid isPermaLink="false">https://medium.com/p/468e3a2a0dd3</guid>
            <category><![CDATA[entrepreneur]]></category>
            <category><![CDATA[entrepreneurship]]></category>
            <category><![CDATA[user-research]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[customer-development]]></category>
            <dc:creator><![CDATA[Marianne Berkovich]]></dc:creator>
            <pubDate>Fri, 19 May 2017 16:44:55 GMT</pubDate>
            <atom:updated>2017-05-22T15:51:34.759Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*sgiLHPDSM_jt8M70." /></figure><p>When you are building a new venture, it can feel like aiming an arrow at a target off in the distance. Product market fit is a key material of your arrow (in addition to things like timing, team, funding, channels, revenue models, etc). You make the best arrow you can, draw your bow and let the arrow fly, hoping to hit the bullseye of a successful product.</p><p>As a user research partner for early stage entrepreneurs, I’ve heard lots of people’s adventures with finding product-market fit. Here are three stories of real entrepreneurs hitting or missing that bullseye because they did — or didn’t — do customer discovery/user research and what happened. If you want to better aim your venture’s arrow towards a bullseye, keep reading to avoid Rob’s and Mike’s mistakes, and find out what Jason did right.</p><h4><strong>Miss: Just Building</strong></h4><p>Rob (not his real name) had assembled a crack team of engineers and data scientists to build products that help people make better decisions using Artificial Intelligence. They had won a contract to develop a website for a very large client. They spent a year developing it, polishing the algorithm to make sure that they were showing the best possible recommendations, and worked their butts off right up until the deadline to make the big launch date where their client announced the site to millions of their users.</p><p>The response from users was underwhelming. Few people were sticking around after their first visit. What went wrong? Although internally the team had raised the question among themselves once or twice about whether they were building the right thing, they didn’t voice them or test early versions with users. They had felt sure of their solution. And they missed the mark of building something that people used.</p><h4><strong>Miss then a Hit: Just Building, then Research</strong></h4><p>Mike (not his real name) was a Product Manager for a startup that made small businesses loans. Noticing that income is irregular for small businesses, the team had an idea for a new feature for their software: What if small businesses could repay their loans in increments as a percentage of their revenue each month rather than the same set amount? This would take the pressure off having to make a large payment at the end of each month when revenue was small or variable. It was brilliant!</p><p>The team built the feature and launched it. They saw a small trickle of adoption but most people weren’t using it. Mike was dumbfounded — why weren’t they using it? He sat down with the customer list and started making phone calls, asking customers about the feature and their experiences with it. Through these interviews, he learned that many of the customers hated it; it resembled the tactics that shady, predatory lenders used and they didn’t trust it. He also found that the customers who hadn’t experienced predatory lenders loved the new feature and it solved a real problem for them.</p><p>Mike took a step back to regroup. He realized he had uncovered another pain point that he could solve for his the majority of his users — dealing with shady lenders. He and his team built a feature to go after these guys and help his customers get out of these bad deals. His initial feature may have missed the mark, but it was a bullseye on the second try.</p><h4><strong>Hit a Bullseye: Research from the Start</strong></h4><p>Several years ago, Jason, an engineer, noticed that companies like his were devoting a lot of engineering resources to building internal tools that were essentially simple forms on top of a database. Over and over again.</p><p>He realized that these were just spreadsheets on steroids — contorted to do many things in the corporate world like data models, process tracking, etc. And he became obsessed with the idea of a tool that has the familiar UI of a spreadsheet, but backed by the power of a relational database. A database would let the tool do those crazy contortions well, rather than cobbled together.</p><p>He started doing interviews with people who use spreadsheets. He asked people to show him their spreadsheets and talked to them about their pain points of where spreadsheets missed the mark. He started to hear themes and he developed a clearer vision of what his new product could be. He made a quick prototype that took less than a week to build and was just 1500 lines of Javascript. He created a short YouTube video introducing the product, showing how to use it, and describing the future direction. This vision based on what Jason saw and heard in interviews struck a chord; he raised $500k in seeding funding to build Version 1.</p><p>He brought together a small team and they started building <a href="https://fieldbook.com/">Fieldbook</a>. As they were approaching their private beta, Jason left time in the schedule for two weeks of user testing to fix any small bugs that came up. But those early tests were painful; people didn’t get how to do basic functions. The problems were not small bugs. Jason realized that they would have to push their private beta date out. He knew that the product concept resonated but the actual experience of using the product was too difficult. So he took his beta waiting list, and turned it into a potential participant list for user studies. This was the perfect pool of target users — people who had a real need for the product, really wanted to use it, and hadn’t used it before. And he rewarded them for participating in a study with early access to Fieldbook.</p><p>He organized 5 user sessions per week, for week after week; they would test with users on Monday afternoon and Tuesday morning. Then he’d build a new release that was ready to be tested the following Monday morning. During the session, participants had an opportunity to use the product while Jason and his team observed. And he also spent part of the session interviewing them to better understand their context: “What got you interested in product?” and “How do you do it today?”</p><p>The Fieldbook team iterated and iterated. Five months later, they were confident they had a product that met the needs of their users. Their user growth and retention numbers reflected that. And investors noticed too. Fieldbook raised $2.2 million in seed funding in April 2016.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cqu8FwKdcj8jvKjY6zXRHA.png" /><figcaption>How Fieldbook looks now</figcaption></figure><h4><strong>Research moves the Target Closer</strong></h4><p>Rob and his team skipped user research entirely and missed the target entirely. Mike reached out to users only after an “Oh sh*t!” moment where few people were using the brilliant new feature, and he course-corrected to hit the target on a second try. Jason reached out to users throughout his process, moving the target closer, and making it easier to hit that bullseye.</p><p>As a founder, it’s critical to start talking to your users early and often to deeply understand them, their problems and needs, in order to build the right thing. Doing this helps move that far off target closer, making it that much easier to hit a bullseye. As you are developing your own product, consider whether you are moving the target closer through user research. If you aren’t sure how to do user research, let’s talk.</p><p><a href="http://marianneberkovich.com/"><em>Marianne Berkovich</em></a><em> has over 12 years experience as a qualitative user researcher at Google and Adobe, helping teams build products that users love. She has worked with and mentored dozens of smaller companies and early stage entrepreneurs to gain the know-how to do their own user research. Marianne teaches classes at the Nasdaq Entrepreneurial Center as well as tailored workshops for intact teams.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=468e3a2a0dd3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Honing your user research craft: Doing a critique through a “Don’t Leave Data on the Table” class]]></title>
            <link>https://medium.com/@mberkovich/honing-your-user-research-craft-doing-a-critique-through-a-dont-leave-data-on-the-table-class-23871b7bbde2?source=rss-35410c649a3e------2</link>
            <guid isPermaLink="false">https://medium.com/p/23871b7bbde2</guid>
            <category><![CDATA[ux]]></category>
            <category><![CDATA[user-research]]></category>
            <category><![CDATA[user-interviews]]></category>
            <category><![CDATA[user-experience]]></category>
            <dc:creator><![CDATA[Marianne Berkovich]]></dc:creator>
            <pubDate>Thu, 01 Dec 2016 20:31:21 GMT</pubDate>
            <atom:updated>2016-12-06T23:11:42.844Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zz-AR_NFkk7xUsLnx3jmFA.jpeg" /></figure><p>If you are someone who conducts user interviews, you’ve probably developed a personal style for how you run a session. When you do something repeatedly, it becomes second nature, and you stop thinking critically about what you’re doing. Bad habits emerge, and best practices erode over time. Haven’t we all unintentionally raced through the intro spiel with a participant because WE’VE heard it a million times?</p><p>Unlike our design partners, user experience researchers do not have a regular practice of critiques — observing each other’s sessions and offering suggestions on how to improve. You probably have product team members watching your sessions, but they are focused on the participant, not you — and not on bringing a critical eye to your research technique. When they ask, “How’d it go?” they are interested in hearing what insights and findings you uncovered about the product. If someone also asks you that question about whether the research “worked,” consider yourself lucky; most of us rarely take time to reflect and refine our own techniques and approaches or discuss this with others. As user researchers, we need opportunities to get feedback on our techniques and approach in order to continually improve our craft.</p><p><strong>“DON’T LEAVE DATA ON THE TABLE” CLASS: A SAFE PLACE FOR CRITIQUE</strong></p><p>To avoid the user researcher rut and create a safe place for user researchers to give feedback to each other about their session moderation practices and technique, I recruited my fellow user researchers Martin Ortlieb and Elizabeth Baylor, to create a class called “Don’t Leave Data on the Table” for Google’s UX University 2015 (an annual gathering where Google’s user experience professionals learn from one another). The class was so popular and oversubscribed that we have repeated it several times thereafter.</p><p>This story describes the key principles of running this class.</p><ol><li>Create a safe space</li><li>Conduct the critique</li><li>Reflect and make an action plan</li></ol><p><strong>FIRST, CREATE A SAFE SPACE</strong></p><p>When we invited class members, we were transparent about what would happen in the class — that we would form small groups to watch a video of each class member moderating a user interview. We asked class members to send us links to their videos ahead of time so we could preview them and pick the “juiciest” one for the session, one that seemed to have the most interesting moments.</p><p>When class members arrived in the classroom, we made the environment playful and friendly — playing music and decorating the space with a special surprise.</p><p>We knew that people would be uneasy about watching themselves on video, so we did an ice breaker that also got people feeling good from the get go. The ice breaker did double duty by modeling giving positive feedback and building on the suggestions of others — both of which would be important for good group critique. We used an Amplification exercise that <a href="http://www.thoughtleadershiplab.com/AboutDenise">Denise Brosseau</a> introduced us to: in pairs or threes, one person says something they are good at — it could be something silly like “I am great at picking the right size Tupperware for leftovers” or ‘I’m great at changing the oil in my car myself” Then a partner amplifies this trait, saying something that might sound like, “Marianne is so great at changing the oil in her car that she can do it in under 5 minutes — take that Jiffy Lube. She is so fast at it that she has NASCAR pit crews asking her for tips.” You get the idea — the amplification doesn’t have to be based in fact, it only needs to take a small trait and amplify it to something enormous. And it feels good.</p><p>As we started the class, we reminded the class members that we are all in this together: “You are all good researchers, and we are all here to learn together — none of what we’ll be doing today is about our flaws, but about improving ourselves. We know that watching videos of yourself is hard, and we’ll be making ourselves vulnerable in order to learn.” We were lucky at Google that there was a strong culture of respect and trust already, so we did not need to do much to bring that spirit to the class. If you try this class on your own, we suggest doing it with a group of people with an existing sense of trust, whether this is coworkers or people in your own network.</p><p><strong>SECOND, CONDUCT THE CRITIQUE</strong></p><p>We formed teams of 3 or 4 class members per instructor. Each student was allotted 15 minutes. We started watching 16 minutes into each session so it would be after the introduction and technical set up, and in the heart of the content of the session. The student introduced the video and context of the study and then we hit play.</p><p>We borrowed a concept from the Kaizen philosophy of continuous improvement, where each worker on the production line is empowered to stop production when they spot a defect by pulling the <a href="https://en.wikipedia.org/wiki/Andon_(manufacturing)">Andon cord</a>. The Andon cord is a device used to notify management, maintenance, and other workers of a quality or process problem. We started watching the video together, and everyone was invited to pull our version of the Andon cord when they spotted something that was deserving of feedback. Our version added an air of whimsy to the class (and we’d like to keep it a surprise until you take our class!). You can get creative with your own lighthearted version that creates a visual or auditory cue.</p><p>We encouraged the other group members to go with their hunches and to not be shy about pulling the cord based on just an inkling of an observation — when a little something just didn’t sit right with them. When we stopped the video to make an observation, we invited the person whose video it was to “rent the idea” — not necessarily “buy it.” This meant listening to what people said, and evaluating later. It’s possible that the group picked up on something that was a fluke or that the class member was having a bad day, but “renting the idea” was an opportunity for the person to consider whether what was noticed was part of a larger pattern.</p><p>We invited the person whose video it was to critique themselves first since they have the most context about what was happening, and are the most familiar with their own actions and behaviors.</p><p>We drew inspiration about how to give feedback from Patrick Cox’s <a href="http://tympanus.net/codrops/2012/10/15/the-unwritten-rules-of-a-great-design-critique/">Unwritten Rules of a Design Critique</a>. In particular, we thought the following rules apply well to a user research critique:</p><ul><li>Use comparisons between class members sparingly — focus on the current session</li><li>Be specific</li><li>Ask why</li><li>Offer suggestions and alternatives</li><li>Be positive — stop the video to comment on things that work well too</li></ul><p><strong>THIRD, REFLECT AND MAKE AN ACTION PLAN</strong></p><p>We wanted to make sure that class members would walk away with some actionable insights. We gave them a note-taking sheet with the <a href="https://en.wikipedia.org/wiki/SKS_process">SKS process </a>(Stop, Keep doing, Start) framework:</p><ul><li>Things to stop doing</li><li>Things to continue doing</li><li>Things to start doing</li></ul><p>Each group member was responsible for taking notes for themselves. We believe that each person should be responsible for their own learning and is the best judge of what was most applicable to them and what they could best absorb right now. We also found that people learned as much from other people’s videos as they did from their own, as they saw themselves in the missteps and missed opportunities of others; many people took as many notes during other people’s sessions as they did their own.</p><p>The facilitators had watched the videos beforehand and were prepared with observations to share, in case the person or the group members didn’t bring these up. This helped ensure that each person came away with substantive insights.</p><p>At the end of class, we gave class members time to reflect on what they heard and decide which ideas they wanted to buy and how they were going to make those changes.</p><p><strong>PATTERNS WE OBSERVED</strong></p><p>After running several sessions of the class, we started to notice patterns that have almost become “folk wisdom” within the user research community, but can thwart creating the type of atmosphere where deep insights emerge. Let’s take a look at three of them.</p><p><em>1. “We’re not testing you”</em></p><p>The ostensible goal of saying this is to put the participant at ease and reassure them that any issues they encounter are with the software, not with themselves. However, by bringing up testing and then reinforcing that with subtle (and sometimes not so subtle) cues that this is a test, our words undermine our intentions. For example, later in the session, you may find yourself saying, “Can you see a way to do X here?” implying that there is a way to do X and the participant needs to find it to pass the exam, or showing excitement when the participant does the “right thing” to progress in a task. In our class, we’ve watched videos of participants tensing up and then guessing different actions until they found the right one that won the approval of researcher. This testing atmosphere is generally not what we want our participants to experience, so we need to pay attention to the messages we are sending both with our words and body language.</p><p>We suggest focusing on what the experience is rather than what it isn’t. Rather than introducing the idea of a test, frame the session as a conversation in which the researcher and the participant will be looking at or doing some things. Another solution is to remind the participant of why THEY were selected for the study — because of their unique experience which is exactly what we’re interested in.</p><p><em>2. “We’d like your feedback on…”</em></p><p>This one is related to the testing and evaluative atmosphere that you can unintentionally create. The word “feedback” implies that you are asking the user to make an evaluation. Sometimes this is exactly what you are looking for, but most of the time, you are much more interested in understanding and observing how participants interact with a product or service. You can ask for a summary assessment at the end of the session, but opening the session saying that we are looking for feedback, is begging the participant to put on their ‘Evaluator’ hat rather than their ‘User Using a Product’ hat.</p><p><em>3. “What are your thoughts on this?”</em></p><p>Open ended questions are great, but there is too much of a good thing. If your goal is to get a sense of what’s most important to someone, then by all means ask this very open ended question, but beware the trap of the lazy question. Beware of asking this question because you couldn’t come up with something better to ask; it puts the onus on the participant to try and guess what you are interested in hearing about. For example, are you interested in their understanding of what happened, how useful this feature is in their workflow, what does the feature not do that they need it to, or something else entirely.</p><p><strong>DON’T LEAVE DATA ON THE TABLE</strong></p><p>We’ll be teaching this class as workshop at the upcoming <a href="https://chi2017.acm.org/">CHI conference in May 2017</a> called “UX Interviewing: Personalized Coaching to Avoid Leaving Data on the Table.” We’d love to see you there.</p><p>If you can’t make it, we’d love to hear from you on how you hone your craft, and what you think of this approach. Leave a comment below or contact us at <a href="/service/mailto:hello@marianneberkovich.com">hello@marianneberkovich.com</a>.</p><p>___________________</p><p>Huge thank you to Elizabeth Elliott Baylor on her invaluable feedback on this story!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=23871b7bbde2" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>