<?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 Roy on Medium]]></title>
        <description><![CDATA[Stories by Roy on Medium]]></description>
        <link>https://medium.com/@imsroy?source=rss-ba994a0adb4d------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*6V5EQjZo3HaImQM1j40xqw.jpeg</url>
            <title>Stories by Roy on Medium</title>
            <link>https://medium.com/@imsroy?source=rss-ba994a0adb4d------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 15:26:02 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@imsroy/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[Learning How to Learn]]></title>
            <link>https://imsroy.medium.com/learning-how-to-learn-1aff8111233c?source=rss-ba994a0adb4d------2</link>
            <guid isPermaLink="false">https://medium.com/p/1aff8111233c</guid>
            <category><![CDATA[hr-talent-management]]></category>
            <category><![CDATA[future-of-work]]></category>
            <category><![CDATA[talent-management]]></category>
            <dc:creator><![CDATA[Roy]]></dc:creator>
            <pubDate>Tue, 10 Nov 2026 03:01:19 GMT</pubDate>
            <atom:updated>2026-11-10T03:01:19.631Z</atom:updated>
            <content:encoded><![CDATA[<p><em>The future of talent may depend less on what we know today and more on how quickly we can learn what we don’t know tomorrow.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GObZyJrC8ddMIQ9Kd_LyiA.png" /></figure><p>There was a time when a professional career followed a fairly predictable path. We acquired knowledge, found our way into an organization, learned the craft from those who had walked the road before us, and gradually developed the expertise that would carry us through much of our working lives.</p><p>Experience was accumulated, almost like compound interest. Each assignment, each mistake, each difficult conversation added something to our professional judgment. The world was never entirely predictable, of course. Continuous learning was always encouraged, and the pace of change allowed us the underlying belief that expertise accumulated over time would remain valuable. AI is beginning to unsettle that belief.</p><p>When machines can write code, analyze financial statements, interpret medical records, and perform complex research in seconds, perhaps we are asking the wrong question about the future of talent. Rather than obsessing over what skills the next generation should acquire, shouldn’t we be thinking about <strong>how to teach them to learn, unlearn, and relearn?</strong></p><p>And there is a second, less obvious question lurking behind the first. In our enthusiasm to make experienced professionals more productive, are we quietly removing the very opportunities through which the next generation becomes experienced?</p><h3><strong>From Personnel Management to Talent Development</strong></h3><p>It is interesting to look back at why Human Resources came into existence in the first place. The origins of modern HR can be traced to the industrial revolution, when businesses needed a systematic way to manage an expanding workforce. Early personnel departments focused on hiring, wages, working conditions, employee welfare, and eventually labor relations.</p><p>As businesses became more sophisticated, the function evolved beyond managing people to developing their capabilities. Training, career progression, leadership development, and succession planning became central responsibilities. The underlying economic logic was straightforward. Organizations needed people with the right skills to operate and grow their businesses, and investing in those skills created value for both employees and employers.</p><p>For generations, this arrangement worked reasonably well. Organizations could anticipate the capabilities they would need, universities could prepare students for those expectations, and HR could help bridge the remaining gaps.</p><p>There was an implicit understanding that the future would resemble the present closely enough for this model to work. AI is making that understanding increasingly fragile.</p><h3><strong>The “Learning Ladder” we may be removing</strong></h3><p>There is another, perhaps less obvious, consequence of AI adoption that concerns me. As experienced professionals increasingly use AI to improve productivity, they are inadvertently eliminating the very learning opportunities that helped them become experienced in the first place.</p><p>Think about the traditional career of a consultant, software engineer, or financial analyst. Junior professionals started with research, documentation, analysis, and repetitive tasks. Through these assignments, they learned how businesses operate, how decisions are made, and how to apply judgment. Experienced professionals mentored them, reviewed their work, and gradually transferred greater responsibilities. This was the learning ladder that developed the next generation of professionals.</p><p>Today, many of those foundational tasks can be performed by AI, often faster and at a lower cost. While businesses celebrate productivity improvements, we may be creating an unintended <strong>talent development vacuum</strong>. If senior professionals can accomplish more with AI and organizations need fewer junior resources for routine work, where will the next generation gain the experience required to become senior professionals? What will happen when the senior professionals retire?</p><p>This is not an argument against AI adoption, rather an argument for consciously redesigning how we develop talent. And it fundamentally changes the mandate of HR, from developing competencies for known roles to building organizational capabilities for continuous learning.</p><h3><strong>Learning how to learn</strong></h3><p>I believe the most important capability we need to develop is <strong>learning how to learn</strong>. A few years ago, knowing a particular programming language, enterprise platform, or analytical methodology could provide a meaningful career advantage. Today, AI can help professionals acquire and apply many of those technical capabilities much faster. That does not make foundational knowledge irrelevant. In fact, understanding the fundamentals becomes even more important when evaluating AI-generated work. But the differentiator is shifting.</p><p>Consider two young professionals entering the workforce. One has mastered a particular technology and knows how to execute a defined set of tasks. The other understands the underlying business problem, knows how to explore unfamiliar subjects, can ask thoughtful questions, and is comfortable experimenting with new tools. The first may be productive immediately. The second may be better equipped to remain productive as technology and business models evolve.</p><p>That is the essence of learning how to learn. It involves intellectual curiosity, critical thinking, adaptability, problem-solving, and the ability to connect knowledge across disciplines. In an AI-enabled workplace, these are no longer merely desirable soft skills. They are becoming fundamental professional competencies.</p><h3><strong>Bridging the gap between academia and industry</strong></h3><p>This is also one of the reasons I have increasingly valued my association with academic institutions. Through my engagements with universities, I have had opportunities to bring real-world business and technology consulting perspectives into the classroom.</p><p>Whether discussing digital transformation, emerging technologies, or the fundamentals of consulting, my objective has been to connect academic concepts with the ambiguity of actual business problems. In the classroom, we can explain what digital transformation means. But in the industry, a consultant must determine why a lending institution struggles with loan origination efficiency, why a healthcare organization faces revenue cycle challenges, or why an insurer cannot modernize its claims operations despite investing in technology.</p><p>These are not always problems with textbook answers. And that is precisely why I believe industry participation in education is so important. Students need opportunities to work through ambiguous situations, challenge assumptions, collaborate across disciplines, and develop solutions that balance technological possibilities with business realities.</p><p>We often use the term <em>job-ready graduates</em>. I prefer to think of them as <strong>hire-ready talent</strong>. Individuals who may not know every tool or process on their first day but possess the foundations, curiosity, and learning agility to become valuable contributors quickly.</p><h3><strong>A shared responsibility for the future workforce</strong></h3><p>Preparing this workforce cannot be the responsibility of universities alone. Businesses frequently complain about the gap between academic education and industry expectations. Yet many organizations are becoming less willing to invest in developing entry-level talent, particularly when AI can perform much of the foundational work. There is a contradiction here. If every business expects someone else to develop experienced professionals, where will the next generation acquire that experience?</p><p>I believe addressing this emerging talent vacuum requires a deliberate <strong>tripartite initiative involving businesses, academic institutions, and governments</strong>. <strong>Businesses</strong> must become active participants in talent creation, redesigning apprenticeships, mentoring, and early-career assignments for an AI-enabled workplace. HR should measure not just training hours or certifications, but how effectively employees develop judgment and new capabilities.</p><p><strong>Academic institutions</strong> must evolve from primarily delivering knowledge to developing capabilities, integrating interdisciplinary learning, experiential projects, AI literacy, and stronger industry collaboration.</p><p><strong>Governments</strong> must establish policies for lifelong learning through workforce development programs, incentives for industry-academia partnerships, and accessible reskilling opportunities.</p><p>None of these stakeholders can solve the problem independently.</p><h3><strong>The new talent development mandate</strong></h3><p>I find it somewhat ironic that as we move toward increasingly autonomous AI systems, the most important investment businesses may need to make is in distinctly human capabilities — judgment, curiosity, adaptability, creativity, and the ability to learn continuously.</p><p>AI creates an opportunity to rethink how we develop talent, moving from preparing people for predefined jobs to preparing them for careers that will evolve repeatedly. The next evolution of HR is not simply managing human capital, but <strong>building human adaptability as an organizational capability</strong>. For me, that is also what makes engagement with universities meaningful. It offers an opportunity to contribute, in a small but practical way, to developing the next generation of professionals.</p><p>The future of talent development should not be about predicting every skill the workforce will need three or five years from now. We will almost certainly get some of those predictions wrong. It should be about ensuring that, while AI transforms how work gets done, we do not inadvertently dismantle the learning ladder that creates future talent. In the age of AI, the most valuable skill will be our ability to keep learning.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1aff8111233c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Earned Autonomy: The Missing Link Between AI and ROI]]></title>
            <link>https://imsroy.medium.com/earned-autonomy-the-missing-link-between-ai-and-roi-c0776abfd896?source=rss-ba994a0adb4d------2</link>
            <guid isPermaLink="false">https://medium.com/p/c0776abfd896</guid>
            <category><![CDATA[earned-autonomy]]></category>
            <category><![CDATA[ai-governance]]></category>
            <category><![CDATA[enterprise-ai-roi]]></category>
            <category><![CDATA[ai-adoption]]></category>
            <dc:creator><![CDATA[Roy]]></dc:creator>
            <pubDate>Sat, 31 Oct 2026 23:35:07 GMT</pubDate>
            <atom:updated>2026-10-31T23:35:07.753Z</atom:updated>
            <content:encoded><![CDATA[<p><em>The AI Catch-22 facing business executives</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*E2iKu9LiIL0dIU_oDHY6iQ.png" /></figure><p>In my earlier blog, <a href="https://medium.com/@imsourav_52632/transfer-of-decision-rights-are-humans-on-the-hook-b320d4b2bfce"><em>Transfer of Decision Rights</em></a>, I explored a fundamental question emerging with agentic AI: As we transfer decision-making authority from humans to AI agents, who becomes accountable for those decisions?</p><p>But there is another side to this conversation, one that is creating an interesting dilemma for business executives tasked with delivering measurable ROI from their AI investments.</p><p>Imagine a conversation between a CEO and a business executive.</p><p>The CEO asks, “We have invested millions in AI over the last two years. When do we start seeing meaningful returns?”</p><p>The executive responds, “We have several copilots in production, employees are becoming more productive, and we are seeing efficiency gains. But to realize the true potential, we need to move toward autonomous agents that can execute entire business processes.”</p><p>“Great. What’s stopping us?”</p><p>“Well, before we allow AI agents to independently access our core systems and make decisions, we need to establish the right governance, security and accountability mechanisms.”</p><p>And there lies the <strong>AI Catch-22</strong>. The opportunity to generate transformational ROI requires greater autonomy, while the responsibility to protect the enterprise requires confidence in how that autonomy is exercised.</p><h3>The ROI is in the Autonomy</h3><p>Think about a mortgage origination process, something I have spent a significant part of my professional career working with.</p><p>Today, an AI copilot can help a loan officer summarize borrower information, identify missing documents, analyze income statements and recommend the next action. These capabilities improve productivity, reduce manual effort and allow employees to focus on higher-value activities.</p><p>But imagine taking this a step further. An AI agent independently identifies missing documents, requests them from the borrower, validates the information received, reconciles discrepancies, updates the loan origination system and advances the application to the next stage, escalating only those situations that require human judgment.</p><p>The difference is significant. In the first scenario, AI helps an employee work faster. In the second, AI can fundamentally change how the process operates, potentially reducing cycle times, improving consistency and lowering the cost of execution.</p><p><strong>The first improves individual productivity. The second creates an opportunity to transform process economics.</strong></p><p>And that is where the real promise of agentic AI begins to emerge. However, realizing that promise requires something fundamentally different from traditional automation: the ability to delegate defined decision rights to AI agents.</p><h3>The Enterprise Dilemma: Autonomy and Control</h3><p>Now imagine the same executive discussing the initiative with the Chief Risk Officer.</p><p>The conversation naturally shifts toward a different set of questions. What decisions should the agent be authorized to make? How do we know what information it considered? Can we trace its actions and reconstruct the reasoning behind a decision? When should a human intervene, and who remains accountable for the outcome?</p><p>These are not obstacles to AI adoption. They are essential questions for building an enterprise that can confidently operate with autonomous systems, particularly in regulated industries such as lending, insurance and healthcare.</p><p>One approach is to require human approval at every meaningful decision point. While this provides an additional layer of oversight, it can also limit the operational benefits of autonomy if the same manual approvals continue to govern every transaction.</p><p>At the other extreme, allowing agents to operate without clearly defined boundaries creates challenges around security, accountability and operational resilience.</p><p>The opportunity, therefore, is not to choose between autonomy and governance, but to design an operating model where <strong>governance enables autonomy to increase progressively and responsibly.</strong></p><h3>Autonomy has to be earned</h3><p>I find an interesting parallel in how organizations delegate authority to people.</p><p>When a new employee joins an organization, we don’t immediately provide unrestricted access to systems, budgets and decision-making authority. Instead, we define the person’s role, establish responsibilities, provide appropriate access and set approval limits. As the employee demonstrates competence, sound judgment and reliability, their authority gradually expands.</p><p>Consider a branch manager who can approve loans within defined limits but must escalate larger or more complex decisions. The organization has established a framework that balances empowerment with oversight, allowing decisions to be made efficiently while maintaining accountability.</p><p><strong>Why should we think about AI agents any differently?</strong></p><p>The same Role-Based Access Control (RBAC) principles that govern employee access to enterprise systems should extend to AI agents. Each agent should have a defined identity, role-specific permissions and clearly established decision boundaries, following the principle of least privilege, with access and authority expanding only as trust is earned.</p><p>McKinsey’s research on AI governance points toward a similar measured approach, emphasizing tiered approvals, clear ownership, continuous monitoring, observability and escalation mechanisms.</p><p>Rather than treating autonomy as a binary choice, organizations can progressively expand an agent’s decision rights based on demonstrated performance and the risks associated with those decisions.</p><p>Start with a narrow set of tasks and clearly defined authority. Observe the agent’s actions, measure outcomes, understand exceptions and establish thresholds for human intervention. As the agent consistently demonstrates reliability, expand its scope and decision-making authority while maintaining appropriate oversight.</p><p>Of course, AI agents differ from employees in one important respect. Their behavior can be nondeterministic, meaning that a history of correct decisions does not guarantee the next decision will be right. This makes continuous observability, validation and the ability to intervene essential components of earned autonomy.</p><p><strong>Trust in AI should not be a one-time approval. It should be continuously established through evidence.</strong></p><h3>Governance as an enabler of ROI</h3><p>This brings me to what I believe is an important shift in how business executives should approach AI governance.</p><p>Traditionally, governance has been viewed primarily as a mechanism to manage risk, enforce policies and establish controls. In an agentic enterprise, governance has an opportunity to play a much more strategic role by enabling organizations to confidently delegate greater authority to AI.</p><p>The question should evolve from simply asking, “What should we prevent this agent from doing?” to also asking, <strong>“What would it take for us to confidently allow this agent to do more?”</strong></p><p>That shift can fundamentally change the conversation between business, technology, risk and compliance teams. Instead of debating whether an agent should be autonomous, they can jointly establish the conditions under which autonomy can safely increase.</p><p>Importantly, those conditions should be linked to business outcomes, not just technical performance. An agent that consistently executes tasks correctly but creates additional exceptions, customer friction or downstream rework may not be generating the expected economic value.</p><p>Measuring success therefore requires looking at the complete process: cycle time, cost per transaction, straight-through processing rates, exception volumes, customer experience and the effectiveness of human oversight.</p><p>This is where governance and ROI begin to reinforce each other. Better observability creates greater confidence, greater confidence enables broader autonomy, and broader autonomy creates opportunities for more meaningful business transformation.</p><h3>The next measure of AI maturity</h3><p>In my earlier article, I argued that transferring decision rights to AI agents creates a new accountability challenge. The next question is how organizations should manage that transfer in a way that creates both business value and institutional trust.</p><p>I believe the enterprises that generate meaningful returns from agentic AI will not necessarily be those deploying the largest number of agents or automating the greatest number of individual tasks. They will be the ones that establish a disciplined approach to progressively transferring decision rights while maintaining visibility, control and accountability.</p><p><strong>Perhaps the real measure of enterprise AI maturity is not how many agents we have deployed, but how much autonomy we have learned to responsibly delegate.</strong></p><p>In the agentic enterprise, autonomy should neither be unrestricted nor unnecessarily constrained. I<strong>t should be earned.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c0776abfd896" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Are Businesses Having “Insta Moments”?]]></title>
            <link>https://imsroy.medium.com/are-businesses-having-insta-moments-29a51dece636?source=rss-ba994a0adb4d------2</link>
            <guid isPermaLink="false">https://medium.com/p/29a51dece636</guid>
            <category><![CDATA[instagram]]></category>
            <category><![CDATA[aifomo]]></category>
            <category><![CDATA[fovo]]></category>
            <dc:creator><![CDATA[Roy]]></dc:creator>
            <pubDate>Thu, 22 Oct 2026 15:29:47 GMT</pubDate>
            <atom:updated>2026-10-22T15:29:47.685Z</atom:updated>
            <content:encoded><![CDATA[<p><em>When the fear of missing out starts driving AI investments.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WUDQxIVlr5qUe5mKf5Wicw.png" /></figure><p>There is a question increasingly finding its way into executive leadership meetings: <strong>“What are we doing with AI?”</strong></p><p>Competitors are announcing AI transformations, technology providers are showcasing autonomous agents, and industry conferences are filled with impressive success stories. Every announcement creates a sense of urgency. Executives want to demonstrate progress, remain competitive, and ensure their organizations are part of the next wave of innovation.</p><p>The sentiment feels remarkably familiar. Have you ever scrolled through Instagram and suddenly felt that you were missing something in life?</p><p>Someone is vacationing in an exotic destination, another is dining at a restaurant you have never heard of, and someone else seems to have discovered the perfect fitness routine. Before you know it, you are researching your next vacation, booking that restaurant, or contemplating a lifestyle change that wasn’t even on your mind five minutes earlier.</p><p>We are influenced by what others are experiencing, often before considering whether it fits our own interests, priorities, or circumstances.</p><p>Welcome to the world of FOMO — the Fear of Missing Out.</p><p>Now, replace Instagram with industry conferences, analyst reports, technology demonstrations, and LinkedIn posts. Replace exotic vacations with AI agents, autonomous workflows, and enterprise copilots.</p><p>Suddenly, the behavior looks remarkably similar. <strong>Are businesses having their own Instagram moments with AI?</strong></p><h3>When everyone else seems to be ahead</h3><p>Imagine a typical executive leadership meeting. Someone mentions that a competitor has deployed AI agents in customer service. Another executive refers to an impressive AI demonstration at a recent conference. The CEO/Board asks a seemingly reasonable question: “What are we doing with AI?”</p><p>Soon, an AI initiative is launched, budgets are allocated, technology partners are engaged, and teams begin identifying use cases.</p><p>The enthusiasm is understandable. AI offers extraordinary possibilities, and organizations have every reason to explore them. But somewhere in that excitement, a more fundamental question deserves attention.</p><p><strong>What business problem are we actually trying to solve?</strong></p><p>This is where I see an interesting parallel between individual behavior on social media and corporate decision-making.</p><p>On Instagram, we see the outcomes people choose to display, rather than the compromises, expenses, or disappointments behind them. Similarly, in the business world, we hear about successful AI implementations, productivity improvements, and impressive automation stories. The failed pilots, integration complexities, escalating costs, and organizational resistance receive considerably less attention.</p><p>We are comparing our internal operational reality with someone else’s carefully curated success story. And that comparison can become an expensive foundation for investment decisions.</p><h3>The expensive side of corporate FOMO</h3><p>Consider a mortgage lender introducing an AI-powered borrower engagement agent to improve loan conversion. The agent responds instantly, follows up with applicants, and guides borrowers through the application journey. Engagement improves, application completion rates increase, and the implementation appears successful.</p><p>But suppose the lender’s primary operational challenge lies further downstream. Underwriting takes too long, documentation exceptions require repeated intervention, and closing teams struggle with capacity.</p><p>The new agent may generate more completed applications while adding pressure to an already constrained fulfillment process. The technology has improved one metric without necessarily improving the economics of the overall lending operation.</p><p>Or consider an insurance carrier investing in AI-powered claims settlement recommendations. The model assesses claims and recommends settlement amounts quickly. The technology performs well, and decision turnaround improves.</p><p>However, the carrier’s largest source of claims leakage may be inconsistent repair estimates, poor vendor management, or ineffective fraud investigation. Faster settlement recommendations may have limited impact on the underlying loss ratio if those issues remain unresolved.</p><p>Similarly, a healthcare provider might deploy an AI assistant to improve patient scheduling and appointment availability. The assistant efficiently manages appointments and reduces administrative effort. Yet if the real operational challenge is high no-show rates, limited provider capacity, or delays in prior authorization, scheduling automation alone may produce modest financial benefits.</p><p>These are illustrative examples, but they reveal an important distinction.</p><p><strong>An AI solution can work exceptionally well without solving the business problem that matters most.</strong></p><p>The same AI capability that creates significant value for one organization may deliver limited returns for another because their operating models, process maturity, data readiness, and economics are different.</p><p>Just because a vacation destination looks beautiful on Instagram doesn’t mean it is the right vacation for you. And just because an AI use case delivers impressive results for a competitor doesn’t mean it deserves a place on your technology roadmap.</p><h3>From FOMO to FOVO — Fear of Value Overlooked</h3><p>There is an important distinction I would make here.</p><p>I believe organizations should move from FOMO to what I would call <strong>FOVO — the Fear of Value Overlooked.</strong></p><p>FOMO encourages businesses to ask, “What AI technologies are out there or our competitors implementing?”</p><p>FOVO encourages a more valuable question: “Where are we leaving measurable business value on the table, and how can AI help us capture it?” That seemingly small change in perspective can fundamentally alter the investment conversation.</p><p>For FOVO to become a practical management discipline rather than another acronym, executives need to approach AI investments through four questions.</p><p><strong>First, where is the value leakage?</strong></p><p>Start with the economics of the business. Where are margins eroding? Which processes consume disproportionate effort? Where are customers abandoning transactions? What prevents the organization from scaling?</p><p><strong>Second, what is the most effective intervention?</strong></p><p>AI should be evaluated alongside process simplification, traditional automation, better integration, and improved data quality. Sometimes an autonomous agent will offer the greatest opportunity. At other times, redesigning the underlying workflow may create more value at a lower cost.</p><p><strong>Third, how will we measure realized value?</strong></p><p>The number of agents deployed, prompts executed, or employees using copilots provides useful information about adoption, but it tells us relatively little about financial performance.</p><p>Executives should establish measurable outcomes such as reduced operating costs, increased capacity, improved margins, faster cycle times, or incremental revenue. More importantly, they need to distinguish theoretical productivity gains from benefits actually captured in the business.</p><p><strong>Finally, what evidence justifies scaling?</strong></p><p>Begin with a defined business hypothesis, establish a baseline, test the solution, and measure incremental improvement against the total cost of implementation and operation.</p><p>Scale when the results support it. Refine the approach when the evidence suggests a better opportunity. This is how FOVO becomes actionable: a disciplined approach to identifying, prioritizing, measuring, and realizing value.</p><h3>The courage to not follow the crowd</h3><p>One of the most important lessons in business leadership comes from the 2008 financial crisis.</p><p>In the years leading up to the crisis, the US housing market was booming. Mortgage originations were accelerating, subprime lending was expanding, and financial institutions were competing aggressively for market share. The prevailing belief was that housing prices would continue to rise, and the opportunities seemed too attractive to ignore.</p><p>While many financial institutions aggressively expanded their <strong>subprime mortgage lending</strong>, Jamie Dimon, CEO of JPMorgan Chase took a more cautious approach. He recognized the growing risks in the subprime mortgage market and substantially reduced the bank’s exposure, even as competitors continued pursuing aggressive growth.</p><p>JPMorgan was certainly not immune to the crisis and experienced significant mortgage-related losses. However, its comparatively stronger financial position allowed it to navigate the turmoil and acquire Bear Stearns and Washington Mutual while several other financial institutions struggled to survive.</p><p>I worked closely with the above organizations during that phase. The lesson extends well beyond banking.</p><p><strong>Sometimes, the most valuable leadership decision is choosing not to participate in something that everyone else considers an opportunity.</strong></p><p>I see an interesting parallel with the current enthusiasm surrounding AI.</p><p>There is considerable pressure on executives to demonstrate that their organizations are embracing AI. Boards want progress, investors want productivity, employees want modern tools, and technology providers are constantly presenting new possibilities. Under those circumstances, launching another AI initiative can sometimes feel easier than explaining why a particular investment may offer limited business value.</p><p>Yet the organizations that ultimately create sustainable value from AI may be the ones that distinguish genuine business opportunities from technological excitement. They will understand the economics of a process before automating it, establish measurable outcomes before committing significant investments, and recognize that process redesign or better data can sometimes deliver greater value than a sophisticated AI solution.</p><p>The lesson from 2008 is not that organizations should avoid emerging opportunities. It is that the popularity of an investment does not establish its value, and the fear of being left behind should never replace independent judgment.</p><p>That is as relevant to AI investments today as it was to sub prime lending nearly two decades ago.</p><h3>A different kind of AI leadership</h3><p>AI represents an extraordinary opportunity to rethink how organizations operate, serve customers, and compete. The pace of innovation makes experimentation and organizational learning increasingly important.</p><p>But there is a meaningful difference between moving with urgency and moving with purpose. A strong AI strategy should emerge from business priorities rather than a collection of promising demonstrations.</p><p>It requires executives to understand their operating economics, challenge assumptions, and recognize that different organizations will derive value from different applications of the same technology. Sometimes the highest-value decision will be to deploy an autonomous agent. At other times, it will be to simplify a process, improve data quality, or address an operational bottleneck before introducing AI.</p><p><strong>The ability to make that distinction is becoming an important leadership competency.</strong></p><h3>Beyond the Instagram moment</h3><p>Instagram has made us more aware of what others are experiencing, but that doesn’t necessarily make those experiences right for us. AI presents a similar challenge for business leaders. The opportunity is enormous, and the urgency is real, but <strong>the true measure of AI success is how effectively we translate its potential into business value.</strong> Ultimately, the winners will be those who move beyond FOMO, embrace FOVO, and let business priorities drive their AI journey.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=29a51dece636" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Book That Defines Me]]></title>
            <link>https://imsroy.medium.com/the-book-that-defines-me-68fff32806bf?source=rss-ba994a0adb4d------2</link>
            <guid isPermaLink="false">https://medium.com/p/68fff32806bf</guid>
            <dc:creator><![CDATA[Roy]]></dc:creator>
            <pubDate>Wed, 14 Oct 2026 03:02:58 GMT</pubDate>
            <atom:updated>2026-10-14T03:02:58.837Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A reflection on why being a generalist in a world obsessed with specialization has shaped the way I think, learn, and solve business problems.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rRTpelyCROWzLWHMIzqlGA.png" /></figure><p>There are books that teach you something new. There are books that challenge what you believe. And occasionally, there is a book that helps you understand something about yourself that you had never quite been able to articulate.</p><p>For me, that book is <strong><em>Range: Why Generalists Triumph in a Specialized World by David Epstein</em></strong>.</p><p>I came across Range after spending more than two decades in business and technology consulting, working across industries, technologies, and business functions. By then, I had built a career that was difficult to describe through a single specialization. I had worked in lending, financial services, insurance, healthcare, enterprise technology, and business operations. Each experience was different, yet somehow they all seemed connected.</p><p>For a long time, I wondered whether I should have gone deeper into one particular area rather than constantly expanding into new ones.</p><p>Reading Range gave me a different perspective.</p><h3>The 10,000-hour question</h3><p>Years earlier, I had read Outliers by Malcolm Gladwell. One of its most memorable ideas was the 10,000-hour rule, inspired by research on deliberate practice. The popular interpretation was simple, spend roughly 10,000 hours practicing something, and you can become an expert.</p><p>It was a compelling argument. And in many ways, it made intuitive sense. If you want to become an exceptional pianist, chess player, or athlete, years of focused practice are enormously valuable. Expertise requires discipline, repetition, and commitment.</p><p>But I always wondered how well that principle applied to the business world. What happens when the problem you are trying to solve is something you have never encountered before? When the rules keep changing, the technology is evolving, and the industry itself is being disrupted?</p><p>You could spend 10,000 hours mastering a particular technology, only to discover that the technology is no longer relevant. That is where Epstein’s argument resonated with me.</p><p>In Range, he distinguishes between what psychologists call kind learning environments, where patterns are relatively predictable and feedback is clear, and wicked learning environments, where the rules are ambiguous, feedback is delayed, and past experience may not reliably predict future outcomes.</p><p>Business consulting, in my experience, operates much closer to the second category.</p><h3>The value of knowing a little about many things</h3><p>Looking back at my career, I realize that some of my most valuable contributions came not from knowing more about a particular industry (though I spent a long time in mortgage) than everyone else in the room, but from recognizing something familiar in an unfamiliar problem.</p><p>A mortgage origination process, for example, may appear very different from healthcare revenue cycle management. One involves borrowers, credit decisions, underwriting, and loan funding. The other involves patients, insurance eligibility, claims, and reimbursements.</p><p>Yet step back from the industry terminology, and the similarities become interesting. Both involve complex workflows, document-intensive processes, multiple decision points, regulatory requirements, exceptions, and handoffs between people and systems.</p><p>A solution that improves workflow orchestration in mortgage lending might offer valuable lessons for healthcare claims processing. An approach to fraud detection in insurance might inspire a different way to identify risk in consumer lending.</p><p>The technology may be different. The business language certainly is. But the underlying patterns are often remarkably similar. And that is where I believe a generalist can bring something different to the table.</p><p>Not by replacing deep expertise, but by connecting expertise that would otherwise remain isolated.</p><h3>The Consultant’s advantage</h3><p>Over the years, I have had the opportunity to work with specialists who possess extraordinary depth in their respective fields. I have spent two decades in consumer lending. That knowledge is indispensable, particularly when solving complex technical or domain-specific problems.</p><p>But I have also observed something interesting. The deeper we become in a particular discipline, the more likely we are to approach new problems through the lens of what we already know.</p><p>A technology specialist may see a technology problem. An operations executive may see a process problem. A data scientist may see an analytics opportunity.</p><p>None of them is necessarily wrong. But the real business problem may sit somewhere between those perspectives.</p><p>I have increasingly come to believe that an executive consultant’s value lies in connecting these dots. Understanding enough about business strategy, operations, technology, people, and economics to recognize the larger picture, and then bringing the right specialists together to solve it.</p><p>It is less about having every answer and more about knowing which questions to ask, which patterns to recognize, and where to look for ideas.</p><p>In that sense, breadth of experience becomes a form of expertise in itself.</p><h3>Why this matters even more in the age of AI</h3><p>Today, as AI begins to reshape how organizations operate, I find Epstein’s argument even more relevant. Technical knowledge is becoming increasingly accessible. AI can help us write code, analyze documents, develop models, and retrieve specialized information in seconds.</p><p>But knowing how to apply that knowledge to a business problem remains a different challenge. The real opportunity is not simply automating what an organization already does. It is questioning whether the process should exist in its current form, whether a model from another industry could work better, or whether the problem needs to be reframed altogether.</p><p>That requires curiosity, context, judgment, and the ability to think across disciplines. In other words, it requires range.</p><h3>Perhaps I was never meant to specialize</h3><p>I don’t believe Epstein’s argument invalidates Gladwell’s. In fact, I see them as complementary.</p><p>Deep expertise helps us execute exceptionally well within a discipline. Broad experience helps us recognize when an entirely different discipline might hold the answer.</p><p>The challenge is knowing when to go deep and when to look wide.</p><p>After nearly three decades in consulting, I have stopped viewing my experience across industries and technologies as a collection of unrelated chapters. I now see it as a connected body of knowledge, a T-shaped knowledge graph, where each experience adds another perspective to the next problem.</p><p>Perhaps that is why Range feels so personal. It did not convince me to become a generalist. It helped me recognize that I had been building that capability throughout my career, often without realizing it.</p><p>And as an executive consultant, I have come to appreciate that the most valuable idea in the room may not come from someone who has solved the same problem a hundred times. It may come from someone who has seen a similar problem in an entirely different world.</p><p>That, more than anything, is why Range is the book that defines me.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=68fff32806bf" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Transfer of Decision Rights: Are Humans on the Hook?]]></title>
            <link>https://imsroy.medium.com/transfer-of-decision-rights-are-humans-on-the-hook-b320d4b2bfce?source=rss-ba994a0adb4d------2</link>
            <guid isPermaLink="false">https://medium.com/p/b320d4b2bfce</guid>
            <category><![CDATA[observability]]></category>
            <category><![CDATA[ai-governance]]></category>
            <category><![CDATA[agentic-ai-governance]]></category>
            <category><![CDATA[right-decision]]></category>
            <dc:creator><![CDATA[Roy]]></dc:creator>
            <pubDate>Mon, 05 Oct 2026 05:24:18 GMT</pubDate>
            <atom:updated>2026-10-05T05:24:18.387Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*e7fuaSnvDlnV3_KJOPxCLA.png" /></figure><p>Imagine a lending company a few years from now.</p><p>A customer applies for a loan. An AI agent reviews the application, another pulls together the customer’s financial history, a specialized agent evaluates risk, and an orchestrator coordinates the entire process before deciding whether the loan should be approved. There is no loan officer and underwriter reviewing every application, because that would defeat much of the reason the company introduced autonomous agents in the first place.</p><p>Now imagine the customer is declined. A few days later, fair lending audit discovers that the decision was wrong. Who is responsible?</p><p>Is it the business executive who approved the deployment of the agent (s)? The team that designed it? The orchestrator agent that chose the wrong path? The owner of the context and memory layer that failed to provide one critical piece of information? Or perhaps the LLM provider, because the model produced one of those perfectly plausible but nondeterministic responses that nobody could reproduce later?</p><p>This is where the conversation about agentic AI becomes much more interesting than simply asking how many processes we can automate.</p><p><strong>We are not just transferring work from humans to machines. We are transferring decision rights. </strong>And our operating models have not caught up.</p><h3>We have seen this movie before</h3><p>Organizations have been delegating decision rights for decades.</p><p>A branch manager may approve a transaction up to a certain amount. A sales executive can discount within an agreed range. A claims adjuster has authority within defined limits, while larger or unusual claims are escalated. Behind these decisions are policies, approval thresholds, audit trails and, importantly, clearly identified people who remain accountable.</p><p>Technology traditionally supported that structure. Even sophisticated workflow systems largely executed rules that humans had already defined. The system might calculate, route, validate or recommend, but somewhere in the process there was usually a person or a business function that clearly owned the decision.</p><p>Agents change that relationship!</p><p>An agent can now interpret information, retrieve context, reason across alternatives, invoke tools, consult other agents and take action. As these capabilities improve, organizations will naturally want to remove more human checkpoints, because much of the ROI from agentic AI comes from allowing the process to operate autonomously.</p><p>That creates an uncomfortable trade-off. The more autonomy we give agents, the more valuable they potentially become. But the more autonomy we give them, the harder it becomes to answer who actually owns their decisions.</p><h3>Human-in-the-Loop may not be the end state</h3><p>Our immediate response has been to put humans in the loop.</p><p>That makes sense while organizations learn how these systems behave, but it is difficult to imagine this being the permanent operating model. If an employee needs to approve every meaningful action taken by an agent, we have created a faster recommendation engine rather than a truly autonomous process.</p><p>Perhaps the more useful concept is <strong>human-on-the-hook rather than human-in-the-loop.</strong></p><p>The human does not necessarily need to approve every decision, but someone must remain accountable for the decisions made within a defined boundary.</p><p>A customer-service agent might autonomously issue a $25 credit. A claims agent may settle certain straightforward claims below a threshold. A lending agent might make some decisions independently but escalate others based on value, confidence, exceptions or regulatory requirements.</p><p>In other words, we may need to manage AI agents much the same way we manage delegated authority within organizations today. Give them decision rights, but define the boundaries.</p><h3>Accountability cannot disappear into the technology stack</h3><p>This is where AI governance needs to evolve.</p><p>It will not be enough to say that the business owns the process, the AI team owns the agent, the data team owns the context, the platform team owns orchestration and the technology provider owns the underlying model.</p><p>That may accurately describe the architecture, but it does not describe accountability. When something goes wrong, distributing responsibility across five technology layers effectively means nobody owns the decision.</p><p>Every autonomous process therefore needs a clear answer to a few basic questions: <strong>What decisions can the agent make? Within what boundaries? What information must be available before it acts? When must it escalate? And, ultimately, which human leader owns the outcome?</strong></p><p>Observability can tell us what happened. Traceability can help explain how it happened. Governance can define what was allowed to happen.</p><p>But none of those answers the final question:</p><p><strong>Who owns what happened?</strong></p><p>As agents begin replacing humans inside business processes, the biggest organizational change may not be the disappearance of tasks. It may be the redistribution of authority.</p><p>For decades, companies have carefully designed who gets to make which decisions. We now need to do the same for machines. Because an organization can transfer decision rights to an agent.</p><p><strong>It cannot transfer accountability with them.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b320d4b2bfce" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The library I didn’t know I was building]]></title>
            <link>https://imsroy.medium.com/the-library-i-didnt-know-i-was-building-e802d4907044?source=rss-ba994a0adb4d------2</link>
            <guid isPermaLink="false">https://medium.com/p/e802d4907044</guid>
            <category><![CDATA[nonfiction]]></category>
            <category><![CDATA[books]]></category>
            <category><![CDATA[reading-books]]></category>
            <dc:creator><![CDATA[Roy]]></dc:creator>
            <pubDate>Sun, 27 Sep 2026 17:22:56 GMT</pubDate>
            <atom:updated>2026-09-27T17:22:56.894Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OFGqyzRAH3LfyS1YvgFNXg.png" /></figure><p>Ten years ago, I wasn’t much of a reader.</p><p>I would pick up a business book here and there, usually around a topic that was relevant to work. Then I started reading Malcolm Gladwell. That led me to book recommendations in <a href="https://www.gatesnotes.com/books">Gates Notes</a>, and somewhere along the way, reading became less of an occasional activity and more of a habit.</p><p>I made a simple commitment, six to ten books a year. Nothing ambitious, just consistency.</p><p>Over the years, my interests also changed. I moved away from traditional business books and increasingly toward nonfiction. History, philosophy, social science, biology, technology and human behavior became my choices.</p><p>Malcolm Gladwell, Vaclav Smil, Yuval Noah Harari, Siddhartha Mukherjee and Walter Isaacson became some of my favorite authors. What attracts me to these books is not necessarily the subject itself. It is the attempt to understand <strong>how things work beneath the surface.</strong></p><p>How biological systems become incredibly complex. How energy shapes economies. How information changes societies. How culture and institutions influence human behavior. And ultimately, how information, biology, technology, energy and human choices combine to create what we call civilization.</p><p>For years, I simply enjoyed reading about these things. Only recently did I realize what reading had quietly been doing to me. I find myself in meetings trying to explain an idea, and an analogy suddenly appears. During a conversation, a historical example comes to mind. When I am writing, words and references that I don’t consciously remember learning somehow find their way into the argument.</p><p>It is almost as if ten years of reading created a library in the background that I didn’t know I was building. And that library has become surprisingly useful, because in almost every profession, the ability to <strong>think clearly and articulate an idea </strong>is an underrated advantage.</p><p>You can have a good idea. But if you cannot explain it, connect it to something people already understand, or create a compelling narrative around it, much of its value gets lost.</p><p>This matters even more in the age of AI. AI can summarize a book. It can draft an email, create a presentation or give us the right words. But having your own reservoir of ideas, analogies, perspectives and references changes the quality of the conversation you can have with people and, interestingly, with AI itself.</p><p>The future of education is going beyond learning tools. We have to now learn “how to learn”, which is “learn how to think”. Reading, I have realized, compounds your ability to think and articulate.</p><p>My only regret is that I didn’t start earlier. So, one simple takeaway from ten years of reading. Read more books. You may not know today where you will use what you read. But someday, when you need the thought, the analogy or the words, you may be surprised to find they are already there.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e802d4907044" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Outcome-Based AI Pricing: The Risk Cannot Be One-Sided]]></title>
            <link>https://imsroy.medium.com/outcome-based-ai-pricing-the-risk-cannot-be-one-sided-083f917b62f0?source=rss-ba994a0adb4d------2</link>
            <guid isPermaLink="false">https://medium.com/p/083f917b62f0</guid>
            <category><![CDATA[cost-attribution]]></category>
            <category><![CDATA[outcome-based-pricing]]></category>
            <category><![CDATA[outcome-over-output]]></category>
            <category><![CDATA[ai-pricing]]></category>
            <dc:creator><![CDATA[Roy]]></dc:creator>
            <pubDate>Tue, 22 Sep 2026 17:33:42 GMT</pubDate>
            <atom:updated>2026-09-22T17:33:42.380Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LV1KR0sJ-43I8O-B1JnTkg.png" /></figure><p>Outcome-based pricing is increasingly positioned as the ideal commercial model for AI transformation programs. The provider gets paid only when the client realizes business value.</p><p>It sounds logical. It is also frequently misunderstood and rests on a dangerous assumption that the solution provider controls the outcome.</p><p>In reality, AI outcomes depend on much more than the technology, for example data quality, process maturity, user adoption, management decisions, market conditions and the client’s ability to implement change have huge roles to play in the final outcome.</p><p>These factors may be influenced by the provider, but they are rarely controlled by the provider alone.</p><p><strong>The Myth of Outcome-Based Pricing</strong></p><p>The common belief is that outcome-based pricing creates accountability by transferring delivery risk to the solution provider.</p><p>But delivery risk and business-outcome risk are not the same.</p><p>A provider can be accountable for building a solution that performs as designed. It cannot independently guarantee that customers will buy more, employees will adopt the solution, operating costs will be removed, or management will make different decisions.</p><p>If the provider is expected to assume those risks, it must also have sufficient control over the decisions and conditions that produce the outcome.</p><p>Accountability without control is not alignment. It is simply one-sided risk transfer.</p><p><strong>Output is not the same as Outcome</strong></p><p>The distinction becomes clearer when we compare the primary commercial models:</p><ul><li><strong>Time and materials</strong> pays for the people, expertise and effort applied.</li><li><strong>Fixed price</strong> pays for completing a defined scope within an agreed timeline.</li><li><strong>Output-based pricing</strong> pays for measurable units of work, such as claims processed, applications migrated, documents reviewed or automations deployed.</li><li><strong>Managed services</strong> pays for operating a function against agreed service levels.</li><li><strong>Outcome-based pricing</strong> pays for business impact, such as revenue growth, lower operating costs, improved conversion or reduced losses.</li></ul><p>Output-based models are particularly relevant to AI because outputs are usually easier to measure and attribute.</p><p>For example, an AI solution may review 50,000 loan applications, reduce average handling time by 40% and maintain an agreed accuracy level. Those are measurable outputs.</p><p>But whether the solution increases lending revenue depends on several other factors: credit policy, product pricing, market demand, funding capacity, customer experience and the effectiveness of the sales channel.</p><p>The provider may deliver the output successfully without being able to control the ultimate financial outcome.</p><p><strong>Sales compensation offers a useful analogy</strong></p><p>Salespeople are commonly compensated through outcome-based incentives. But companies do not simply ask them to “grow revenue” and decide later how much credit they deserve.</p><p>They define the territory, quota, product portfolio, eligible revenue, commission rate and attribution rules. They also clarify what happens when multiple teams influence the same sale.</p><p>Even then, sales compensation is rarely 100% variable. A base salary recognizes that the salesperson contributes effort and capability while the final outcome is affected by product quality, pricing, marketing, competition and the broader economy.</p><p>An AI provider operates in a similar environment. If its compensation is linked to outcomes, the rules of attribution must be equally clear.</p><p><strong>The CFO must help define the Baseline</strong></p><p>This is where businesses, and particularly CFOs must play a more active role. Before asking a provider to commit to cost savings or incremental revenue, the organization must establish how that value will be calculated.</p><p>What does the process cost today? Which costs are fixed, and which can genuinely be removed? Does a productivity improvement create financial savings, additional capacity or merely more available time? How will AI-generated revenue be separated from price increases, market growth, marketing campaigns or sales activity?</p><p>The baseline cannot be constructed after the solution has been deployed. By then, every improvement becomes debatable and every shortfall becomes someone else’s responsibility.</p><p>The CFO organization should help define the measurement methodology, attribution rules, validation process and treatment of external factors before the commercial agreement is signed.</p><p><strong>A Hybrid model is often more practical</strong></p><p>For many AI transformation programs, the most balanced structure is a hybrid model:</p><ul><li>A base fee covering the platform, talent and delivery commitment</li><li>An output-based component tied to measurable operational performance</li><li>An outcome-based incentive linked to clearly attributable business value</li></ul><p>This structure creates shared commitment without pretending that one party controls every variable.</p><p>Outcome-based pricing can work very well for AI. But the outcome must be measurable, attributable and reasonably within the parties’ control.</p><p>If the business cannot define how value will be measured, it is not yet ready to buy AI based on that value.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=083f917b62f0" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>