<?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 Stephanie Santana on Medium]]></title>
        <description><![CDATA[Stories by Stephanie Santana on Medium]]></description>
        <link>https://medium.com/@stephanie.santana?source=rss-8d298b002793------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*dmbNkD5D-u45r44go_cf0g.png</url>
            <title>Stories by Stephanie Santana on Medium</title>
            <link>https://medium.com/@stephanie.santana?source=rss-8d298b002793------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 23:33:19 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@stephanie.santana/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[From data to disruption: Reinventing compliance through applied AI]]></title>
            <link>https://medium.com/zs-associates/from-data-to-disruption-reinventing-compliance-through-applied-ai-53f018e9bf4f?source=rss-8d298b002793------2</link>
            <guid isPermaLink="false">https://medium.com/p/53f018e9bf4f</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[pharmaceutical]]></category>
            <category><![CDATA[compliance]]></category>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[data-analysis]]></category>
            <dc:creator><![CDATA[Stephanie Santana]]></dc:creator>
            <pubDate>Tue, 26 May 2026 19:03:17 GMT</pubDate>
            <atom:updated>2026-05-26T19:03:17.672Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Compliance and risk leaders discuss how to turn a flood of risk signals into compliance intelligence you can actually act on at the 2026 Pharmaceutical Compliance Congress.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*sbXJq36KoeyN5UYo" /><figcaption>Photo by <a href="https://unsplash.com/@goumbik?utm_source=medium&amp;utm_medium=referral">Lukas Blazek</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p><em>By: </em><a href="https://www.linkedin.com/in/michaelshawcompliance/"><em>Michael Shaw</em></a><em> and </em><a href="https://www.linkedin.com/in/stephanie-l-santana/"><em>Stephanie Santana</em></a></p><p>The pharmaceutical industry is in the midst of a significant shift in how AI is applied across business functions. As adoption accelerates, compliance programs face a widening gap between the tools the business is deploying and the monitoring capabilities designed to oversee them. Keeping pace requires compliance functions to rethink not just their tools, but how they detect, prioritize and act on risk altogether. At the 2026 Pharmaceutical Compliance Congress, a panel of compliance and risk leaders took a different approach to that challenge. Rather than debating whether AI belongs in compliance, the discussion went straight to what it actually changes, what it takes to act before a violation is confirmed, and how to convert a flood of low-level signals into a pattern worth escalating.</p><p>The panel was co-moderated by Michael Shaw, Principal and Global Head of Risk and Compliance at ZS, and Stephanie Santana, Manager in ZS’s Compliance, Privacy and Risk practice. Joining them were Marcus Green, VP of Risk Analytics and Monitoring at GSK; Brent Boeckmann, Senior Director and Compliance Officer for U.S. and Global at Sanofi; and Christine Giglio, Senior Director of Compliance and Ethics at BMS. The session drew an audience of nearly 100 compliance professionals, reflecting how urgently the industry is looking for practical guidance on what AI-enabled compliance actually looks like in practice, not in principle.</p><h3><strong>Connected intelligence starts from disconnected data</strong></h3><p>Two polls were run at the start of the session to understand the audience landscape. The first poll reflected a monitoring landscape that will feel familiar to most compliance professionals: the dominant model is still reactive, with rule-based alerts, threshold triggers and siloed dashboards reviewed after the fact. Twenty-eight percent of respondents reported pattern-based monitoring already in production, though the majority are still building cases from the outside in. The second asked what the audience considered their biggest barrier to connected risk intelligence, with data quality and access leading the results, followed by legal and organizational comfort with AI-driven decisions and change management. Together, the responses mapped closely to what the panelists had encountered in building their own programs and grounded the discussion in a shared reality.</p><p>Santana framed the gap between where compliance programs operate today and where connected intelligence could take them: from fragmented signals, reactive detection, and manual bottlenecks that cannot scale, toward an integrated view that surfaces patterns invisible in isolation and enables earlier, more targeted intervention.</p><p>Risk signals already exist across enterprises, spread across compliance; commercial; medical; clinical and people data; speaker program records; territory activity; investigation findings; adverse event trends; HR role changes; and training completion rates, though what has largely been missing is the connectivity across it.</p><p>Green offered an analogy to ground the problem statement that compliance functions are navigating and the real challenge that follows:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/672/1*42kqBSk_Boi80co8lCJ4-g.png" /></figure><h3>Progress over perfection</h3><p>A theme that surfaced repeatedly across panelists was to not wait for a perfect data environment or a fully mature AI program to get started. Find one problem, show value and build from there.</p><p>For Green, the spark was a single meals analysis for one business unit in the U.S. He described presenting the findings to the business for the first time and watching the response in the room. They took the data and ran with it. That moment made clear to him and his colleagues that this was the future of compliance analytics. The program that followed became one of the more advanced in the industry, and it started with just one use case.</p><p>Boeckmann drew a clear distinction between chasing the shiniest AI capability and actually solving unmet compliance needs. The right question is not which tool is most advanced. It is where the biggest blind spot is, and whether AI can close it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/652/1*JKNdwmjUBMXWssqXp_u6bA.png" /></figure><p>Giglio described the data quality concern that tends to stall early conversations. Her organization addressed it directly by using AI to help with data stewardship, always with a human check in the loop. Rather than treating data quality as a prerequisite, the team used AI to accelerate the work of getting there. The 80/20 principle showed up in practice: AI handled the bulk of data preparation work, and human experts reviewed and corrected the remainder.</p><p>Shaw noted that while the panel represented compliance programs at some of the largest pharmaceutical companies in the world, the capabilities and approaches being discussed are not reserved for organizations of that scale. The pace of advancement in available tools means that smaller companies can move quickly too, often without the organizational complexity that slows larger programs down.</p><p>To ground the discussion in practice, the panel walked through two case studies: one focused on commercial risk monitoring, the other on investigation pattern recognition. Both illustrated the same core shift, from compliance programs that react to what they can see, to programs that surface what has been invisible.</p><h3>Case one: when ‘all-green’ monitoring is not actually ‘green’</h3><p>Santana opened the first case study with a scenario most compliance leaders will recognize. A large multinational pharmaceutical company was actively monitoring commercial risk across speaker spend, HCP engagement, sales lift versus promotion, expenses and third-party payments, and territory activity. All monitoring was operating as designed with everything technically within policy.</p><p>Despite that, leadership had growing uncertainty. Signals were being reviewed in silos, threshold-based alerts only triggered after a clear rule breach, and subtle patterns across domains remained invisible. Santana posed the central question the case was designed to answer: could siloed, rule-based monitoring miss early signs of emerging commercial risk until it was too late to intervene?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/670/1*jzT74AMutoB-IuNM9txcYQ.png" /></figure><p>Santana walked through the solution: pattern-based analytics across more than 40 commercial variables. Instead of reviewing each indicator against its own threshold, the approach evaluated interaction effects across domains and an anomaly detection engine was layered on top to identify outlier sites, data that was too uniform, unusual number patterns, and unexplained data shifts, comparing behavior relative to peer baselines and surfacing statistically unlikely combinations of otherwise-explainable signals.</p><p>Early warning intelligence depends on connecting both internal signals and external landscape sensing, understanding what is shifting in the broader environment, not just within the enterprise. Giglio described how her team operationalized this, building business rules to identify key external signals, grading them through AI and deliberately curating what surfaced to avoid overwhelming users with noise.</p><p>Individually explainable signals in a subset of high-performing territories had collectively formed an unlikely pattern. AI generated data-driven, explainable, and actionable insights, strengthening organizational trust in the analysis and improving regulatory readiness.</p><p>Santana framed the harder decision that followed: leadership had to determine whether early warning signals justified intervention without a confirmed violation. That meant weighing revenue sensitivity in high-performing territories, potential commercial disruption, enforcement and reputational risk and the need for defensible documentation. None of those factors individually cleared a threshold, but collectively they pointed to a pattern that warranted action.</p><p>Shaw reminded the audience that the anxiety many compliance functions feel about AI surfacing unmanageable volumes of risk is grounded in a misreading of what government guidance has always required. OIG and DOJ frameworks have never demanded perfection. From their earliest iterations, those frameworks acknowledged that compliance programs must be tailored to organizational resources and risk priorities — that not every identified risk requires the same response, and that reasonable prioritization is not a compliance failure but a compliance design principle. AI changes the volume of information available to compliance functions. It does not change the standard by which those functions will be judged.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/681/1*Wd3aamtJKG4T8AbgHVprFw.png" /></figure><h3>Case two: When emerging risk look minor in isolation</h3><p>Shaw introduced the second case study, pressing into a different kind of invisible risk: the patterns buried inside high-volume investigations that look minor when reviewed on an individual basis.</p><p>The company in question managed a substantial volume of commercial allegations annually, drawn from hotlines, HR referrals, monitoring alerts and manager escalations. The triage model was reactive, manual and inconsistent across regions, with allegations reviewed individually, severity assessed case by case and no mechanism to connect similar behaviors across regions or track narrative patterns over time, leaving investigators to manage each file in isolation. The risk was not hiding, rather it was distributed across hundreds of closed files that no one had ever looked at together.</p><p>AI solutions were able to surface repeated low-severity allegations with similar behaviors that had already been closed, narrative patterns that were individually minor, and clusters of cases across regions that converged without a single high-severity trigger. It was a risk that would have remained invisible when cases were reviewed one at a time.</p><p>This problem statement raised the discussion: how does AI practically impact day‑to‑day investigations work? From the compliance perspective, Giglio described the process of building trust with the investigations team, who were initially hesitant about data exposure and sensitivity. The approach: partner with the team to understand what they were already looking for, use AI to search for those patterns at scale, and build in a human review loop at every stage. Over time, as consistency and accuracy were validated, the team became one of the strongest internal advocates of applied AI use.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/650/1*wum_32AXIEvT6fZLXZelpA.png" /></figure><p>Santana closed out the case by framing the operational tension leadership faced: avoiding unnecessary case reopenings, managing investigator capacity, ensuring fairness to employees and producing defensible documentation, punctuating that while AI surfaced the pattern, humans still had to make the judgement call.</p><h3>AI changes the horizon, not the fundamentals</h3><p>Shaw brought both case studies back to the same five-step cycle that ran through the session: detect signals, prioritize by likelihood and impact, decide ownership, act through automation and learn through feedback loops. The mechanics are not new, though what AI changes is the scale, speed and connectivity across those steps.</p><p>Green reinforced this by noting that when he stripped the word “automation” from the cycle, everything else still applied, underscoring that the process is one compliance professionals already understand and apply regularly and that AI should be viewed as a new tool in their toolkit.</p><p>Across both case studies, the same directional shifts emerged consistently:</p><p><strong>From rules-based to pattern-driven monitoring: </strong>The most important signals are often not the ones that breach a threshold. They are the combinations that look benign in isolation and only resolve into a risk when connected across domains.</p><p><strong>From compliance enforcer to business partner: </strong>For Giglio, the shift from reactive oversight to proactive partnership was not incidental. Using AI to bring data and insights to the business before issues surfaced changed the dynamic of how compliance was received across the organization.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/695/1*4ivBmFcqTLZUo-qUkhfECw.png" /></figure><p><strong>From hesitation to iteration: </strong>Every panelist emphasized that the first step does not need to be large. Piloting one use case, validating the output manually alongside AI for a limited period of time and building credibility before scaling is the approach that earns the next step.</p><h3>Where to go from here</h3><p>Shaw closed by asking each panelist to share what their one piece of advice would be for them to act on:</p><p><em>Green: Find where your biggest blind spot is, the area you have not been able to crack, and make that your first AI project.</em></p><p><em>Giglio: Do not get left behind. The pace of change is real. Start small, show value, and trust the process.</em></p><p><em>Boeckmann: A year from now, I want to be able to show progress. Inertia is not the answer.</em></p><h3>The compliance function is at an inflection point</h3><p>For decades, compliance has operated as a function that responds to hotline complaints, to monitoring alerts, to enforcement actions, to audit findings. That model was built for a world where data was scarce, connections between risk signals were manual, and the best a compliance team could do was review what it had and escalate what looked wrong. That world is gone, and the data and tools needed to do something fundamentally different now exist. The only remaining question is whether compliance professionals will choose to lead with them or continue to manage the lag.</p><p>Applied AI does not change the principles that define effective compliance. What it changes is the horizon. Risks that were previously invisible until they materialized can now be seen forming, and patterns that once required a seasoned investigator to surface through manual review can now be identified systematically. For compliance professionals who embrace that shift, the result is not just more effective risk management but a fundamentally different, and more powerful, seat at the table, backed by the kind of enterprisewide risk intelligence that no compliance function has ever had access to before.</p><p><a href="https://www.zs.com/insights?utm_source=medium&amp;utm_medium=organic&amp;utm_campaign=insights-medium-promotion&amp;utm_term=20230926&amp;utm_content=insights">Read more insights from ZS.</a></p><p><em>This article reflects our personal views. They do not necessarily represent any official position of ZS.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=53f018e9bf4f" width="1" height="1" alt=""><hr><p><a href="https://medium.com/zs-associates/from-data-to-disruption-reinventing-compliance-through-applied-ai-53f018e9bf4f">From data to disruption: Reinventing compliance through applied AI</a> was originally published in <a href="https://medium.com/zs-associates">ZS Associates</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Beyond oversight: Why agentic systems will redefine third-party risk management for pharma]]></title>
            <link>https://medium.com/zs-associates/beyond-oversight-why-agentic-systems-will-redefine-third-party-risk-management-for-pharma-5340fd56baef?source=rss-8d298b002793------2</link>
            <guid isPermaLink="false">https://medium.com/p/5340fd56baef</guid>
            <category><![CDATA[responsible-ai]]></category>
            <category><![CDATA[compliance]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[risk-management]]></category>
            <category><![CDATA[pharma]]></category>
            <dc:creator><![CDATA[Stephanie Santana]]></dc:creator>
            <pubDate>Wed, 14 Jan 2026 14:27:56 GMT</pubDate>
            <atom:updated>2026-01-16T16:50:01.805Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*NyUb7HL65wfM6P0E" /><figcaption>Photo by <a href="https://unsplash.com/@campaign_creators?utm_source=medium&amp;utm_medium=referral">Campaign Creators</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>By:<strong> </strong><a href="https://www.linkedin.com/in/michaelshawcompliance/">Michael Shaw</a> and <a href="https://www.linkedin.com/in/hrishikesh-prabhune-5715a83b/">Hrishikesh Prabhune</a></p><p>Even with mature frameworks and significant investment, many organizations still find themselves blindsided by vendor-related incidents.</p><p>Risk data lives in silos across procurement, compliance and IT. Subcontractors and fourth parties sometimes operate in the shadows. Assessments remain periodic, inconsistent and retrospective — telling leaders what went wrong, but rarely what might happen.</p><p>In ZS’s recent discussions with risk and procurement executives, <strong>many shared that at least one major vendor issue</strong> <strong>was identified too late to prevent disruption</strong>. The issue isn’t visibility — its rapid interpretation of risk signals.</p><p>As risks grow more dynamic — cyber disruptions, regulatory shifts, geopolitical shocks — traditional third-party risk management (TPRM) approaches are reaching their limits. The challenge is no longer collecting data; it’s interpreting it fast enough to act.</p><p><strong>What’s at stake:</strong> The cost of missed signals is rising. Overnight, a single vendor failure can halt operations, trigger regulatory action or damage trust. To keep pace, organizations must evolve from oversight to intelligent action.</p><p>That’s where <strong>agentic systems</strong> can redefine how enterprises manage risk.</p><p><strong>From monitoring to intelligent action</strong></p><p>Agentic systems bring together data, context and workflow intelligence to help teams see risks sooner and respond faster.</p><p>Unlike static dashboards, they continuously gather signals from multiple systems — procurement, compliance, audit and external sources — and surface what actually matters. They don’t replace expert judgment; they enhance it by reducing noise, connecting insights and prompting timely action.</p><p><strong>“We don’t need more dashboards — we need systems that tell us what to care about today” </strong>— <strong>Chief Risk Officer, Global Pharma Company</strong></p><p><strong>Where agentic systems create the most impact</strong></p><p>· <strong>Procurement (<em>20–30% spend rationalization opportunity</em>)</strong></p><p>Managing supplier risk today is a balance between cost efficiency and resilience. Agentic systems help procurement teams maintain that balance by connecting data across financial health, performance metrics and external signals. Early warnings — like shipment delays, credit downgrades or Environmental, Social and Governance (ESG) red flags — can surface before they escalate. They also summarize exposure across regions and categories, enabling faster, better-informed sourcing and contracting decisions.</p><p>This isn’t automation for efficiency — it’s foresight for continuous vigilance.</p><p>· <strong>Internal audit (<em>cycle time reduced from months to days</em>)</strong></p><p>Agentic tools can continuously monitor vendor transactions, contract compliance and issue-closure timelines. They help audit teams focus where control breakdowns are most likely, generate evidence summaries and flag recurring <strong>themes across business units</strong>.</p><p>The result: fewer surprises during audits and greater confidence in vendor assurance.</p><p>· <strong>Oversight (<em>one client achieved 90% accuracy using AI mapping of parent vendors vs. conventional matching</em>)</strong></p><p>Regulations change faster than review cycles. Agentic systems can track evolving requirements and map them to vendor operations, highlighting where policy gaps or new obligations might arise.</p><p>This ensures compliance teams stay proactive, not reactive — and maintain readiness across jurisdictions.</p><p>· <strong>GxP and quality</strong></p><p>In highly regulated environments, agentic monitoring can detect early signals from deviations, Corrective and Preventive Action (CAPA) trends and supplier-performance data. Instead of reacting to failed audits or complaints, teams can intervene sooner with data-backed insight.</p><p>It’s about preventing issues before they reach patients or disrupt production.</p><p><strong>The real return on investment (ROI): Confidence</strong></p><p>Ask any risk leader what keeps them up at night, and the answer rarely starts with “efficiency.” It’s <em>uncertainty</em>.</p><p>Uncertainty about whether a supplier will meet obligations when markets shift.<br> Uncertainty about whether the next audit will reveal a control gap no one saw coming.<br> Uncertainty about how quickly the organization can respond when something breaks.</p><p>Agentic systems don’t eliminate risk — but they <em>reduce the unknowns</em>.</p><p>They give teams the confidence that someone — or rather, something — is always watching the data, connecting signals and alerting them when attention is truly needed.</p><p>It’s the confidence that comes from moving beyond hindsight into foresight.</p><p>Confidence that when a vendor stumbles, you’ll know before it becomes a major disruption.<br> Confidence that decisions are grounded in facts, not assumptions.<br> And confidence that your governance framework is not a static checklist but a living, adaptive system that learns with you.</p><p><strong>In a world where uncertainty is the only constant, confidence becomes the most valuable form of return on investment.</strong></p><p><strong>From point solutions to an intelligent ecosystem</strong></p><p>For years, organizations have tried to modernize third-party risk management by layering tools — dashboards for visibility, analytics for insight and workflows for remediation. Each brought value, but often in isolation. The result: more data, not necessarily more intelligence.</p><p>The recent leap in AI capability changes that equation. It’s now <em>more efficient — and more impactful</em> — <em>to move directly toward agentic solutions</em> than to keep expanding point analytics or static dashboards.</p><p>Agentic systems don’t replace existing tools; they <strong>connect and amplify</strong> them. They unify fragmented risk signals across procurement, audit, compliance and quality into a single, intelligent fabric that learns and adapts over time.</p><p>Organizations can start by identifying their strongest existing analytical or monitoring solutions and letting agentic layers sit on top — interpreting, prioritizing and activating insights across teams.</p><p>What begins as a set of point solutions quickly evolves into an <strong>ecosystem of collaborating agents</strong> that make governance faster, sharper and far more responsive.</p><p><strong>The leap forward doesn’t come from adding more dashboards — it comes from enabling systems to think, connect and act together.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/964/1*3CwwVrT0BpTV_OrQbfoByg.png" /></figure><p><strong>What to do now: Building the intelligent risk organization</strong></p><ol><li><strong>Map signal sources.</strong> Identify where risk data already lives and where critical blind spots exist.</li><li><strong>Layer intelligence.</strong> Pilot agentic overlays that interpret and prioritize insights in real time.</li><li><strong>Measure confidence.</strong> Track reductions in detection time, uncertainty and disruption as your new ROI metrics.</li></ol><p>As business ecosystems expand, resilience will depend less on how much data organizations collect and more on how intelligently they act on it.</p><p>The future of third-party risk management isn’t about watching everything. It’s about <strong>knowing exactly where to look, when to act and how to adapt.</strong></p><p>Leaders who act now will redefine governance itself, moving from reactive control to proactive intelligence.</p><p><strong>At ZS, we help organizations reimagine third-party risk management for the age of intelligent systems. If your organization is ready to move beyond oversight and build adaptive, resilient risk management, ZS is ready to help.</strong></p><p><a href="https://www.zs.com/insights?utm_source=medium&amp;utm_medium=organic&amp;utm_campaign=insights-medium-promotion&amp;utm_term=20230926&amp;utm_content=insights">Read more insights from ZS.</a></p><p><em>This article reflects our personal views. They do not necessarily represent any official position of ZS.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5340fd56baef" width="1" height="1" alt=""><hr><p><a href="https://medium.com/zs-associates/beyond-oversight-why-agentic-systems-will-redefine-third-party-risk-management-for-pharma-5340fd56baef">Beyond oversight: Why agentic systems will redefine third-party risk management for pharma</a> was originally published in <a href="https://medium.com/zs-associates">ZS Associates</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Game changer: Shifting priorities and capabilities to mitigate risks in business Gen AI solutions]]></title>
            <link>https://medium.com/zs-associates/game-changer-shifting-priorities-and-capabilities-to-mitigate-risks-in-business-gen-ai-solutions-408dedf87378?source=rss-8d298b002793------2</link>
            <guid isPermaLink="false">https://medium.com/p/408dedf87378</guid>
            <category><![CDATA[responsibleaidevelopment]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[responsible-ai]]></category>
            <category><![CDATA[compliance]]></category>
            <category><![CDATA[compliance-strategy]]></category>
            <dc:creator><![CDATA[Stephanie Santana]]></dc:creator>
            <pubDate>Fri, 30 May 2025 20:16:20 GMT</pubDate>
            <atom:updated>2025-05-30T20:16:20.143Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MwAlqgQHiaWZtKWanJkBBA.jpeg" /></figure><p><em>By: </em><a href="https://www.linkedin.com/in/michaelshawcompliance"><em>Michael Shaw</em></a><em> &amp; </em><a href="https://www.linkedin.com/in/stephanie-l-santana"><em>Stephanie Santana</em></a></p><p>Across the life sciences sector, Gen AI business solutions are proliferating at a fast pace. Gen AI creates <em>original</em> content based on analyzing boundless amounts of data. While autonomous creation of polished content has undeniable benefits, it also has significant risks when moving away from traditional methods of human review and approval. Company compliance officers have a choice: stand in the way of innovation or embed practical compliance controls throughout the end-to-end process of AI solutions.</p><p>At the 2025 Pharmaceutical Compliance Congress, a panel of compliance and data science leaders brought a pragmatic perspective to the latest industry challenge and opportunity as it relates to AI: <strong>Responsible AI</strong>. The panel was led by Michael Shaw, Principal at ZS and Head of the Compliance, Privacy &amp; Risk practice, and included Ethan Yen Senior Manager, Risk Analytics at Gilead Sciences, Stephanie Santana, Manager in ZS’ Compliance, Privacy &amp; Risk practice, and Adnan Shaikh, a ZS Data Science Manager.</p><p>Rather than getting stuck in academic principles or regulatory hypotheticals, the panel’s goal was to demystify what Responsible AI looks like in practice and provide a toolkit of controls that compliance professionals can use to support business innovation without compromising ethical and regulatory standards. The focus wasn’t about AI for compliance, rather it was about compliance for AI. The difference? Many business units are already piloting Gen AI tools — often without knowing what risks have manifested or the options and corresponding trade-offs at play to mitigate those risks.</p><h3>Unlocking a responsible AI mindset</h3><p>Santana kicked off the discussion by level-setting what is meant by the term “Responsible AI” for the purposes of the panel discussion. Responsible AI is not a static checklist. She highlighted, “due to the evolving nature of technology along with the continuous changes across the regulatory and legal landscape as it relates to AI, a checklist could be outdated as soon as tomorrow.” AI governance, which is often used synonymously with Responsible AI, is just the tip of the iceberg when it comes to evaluating AI use cases. Leveraging a framework for responsible AI is more practical to navigate the nuances of each AI use case, weighing the considerations, options, and trade-offs for each decision point throughout the AI lifecycle. The panel’s advice: embed compliance from the start — during solution design, training, and deployment — rather than as solutions are ready to be piloted or after solutions are deployed.</p><h4><strong><em>Responsible AI is the practice of designing, developing, and deploying AI systems that are ethical, transparent, and trustworthy.</em></strong></h4><p>The discussion was grounded in a set of core principles based on standards such as CHAI (the Coalition for Healthcare AI) guidance as well as the NIST AI Risk Management Framework.<a href="#_ftn1">[1]</a><a href="#_ftn2">[2]</a> These core principles should be considered table stakes to comply with relevant standards while also protecting an organization’s reputation and overarching effectiveness of the AI solution:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/571/1*Kek7KXd6I2vlFfSk5D4DUg.jpeg" /></figure><p>Shaw pointed out that a common concern voiced by legal teams is retention. For GenAI tools that answer free-text questions, how long is a user’s input retained? What audit trails exist? What happens when sensitive or inappropriate prompts are submitted? Organizations must determine whether to apply standard enterprise retention policies or adopt use-case specific ones. And just as critically, they should document those decisions transparently to demonstrate good faith governance during audits or investigations.</p><h3>Anchoring to the end-to-end AI lifecycle</h3><p>A methodical AI lifecycle is central to understanding and implementing a Responsible AI framework. Often times, compliance is not involved until the Pilot phase of the AI lifecycle, which can cause timeline delays to retroactively embed the controls to mitigate risks. Compliance should play a role from the define and plan phase through to the deploy and monitor phase. Yen punctuated the benefit of involving compliance at the forefront by sharing an example from a recent Enterprise Risk Management (ERM) project where the Gen AI agent would use non-sensitive, curated data and would answer questions based on documents in the knowledge repository with the ability to share chats with stakeholders. To ensure controls were appropriately in place, they “designed this AI solution with compliance from the ground up.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/576/1*D4OlCM6mEfKb2biQd48xeA.jpeg" /></figure><h3>How does this look in practice?</h3><p>To move from theory to substance, the panel walked through a case study illustration: A Gen AI agent to support field sales reps and MSLs during pre-call planning and post-call summaries. The proposed agent can answer questions based on internal structured and unstructured data, such as prescribing trends, call plan adherence, and engagement history. The business objective of the agent is identification of unmet medical and educational needs in pre-call planning, which remained consistent throughout the AI lifecycle for the purposes of the discussion.</p><p>This illustration is the kind of high-utility, high-risk use case advancing across life sciences and it is exactly why compliance needs to be at the table — early. This type of example also opens a minefield of risks: unauthorized promotion, inappropriate data access, privacy violations, and medical-commercial collaboration concerns. The panel used this scenario to show how different control options can be applied at each stage of the AI lifecycle along with trade-offs and considerations.</p><h3>1. Define and plan:</h3><p><em>“AI is an evolving field that is going to be super dynamic, its regulatory landscape is dynamic, its capabilities are dynamic and how people use them is going to be dynamic…if we get all parties beyond the compliance team…it will help an organization determine what level of risk they are willing to take on for the value of a given use case.” — Yen</em></p><p><em>Select Risks:</em></p><p>· Inappropriate recommendations generated by the agent (e.g., off-label)</p><p>· Data accessed and used by the agent</p><p><em>Select Control Options:</em></p><p>· Prompt Engineering: Design the system to respond only to permissible prompts, with stock messages triggered when off-limits topics like off-label use are queried.</p><p>· Governance Forums: Establish cross-functional AI governance committees, inclusive of compliance, legal, tech, and procurement to assess the risk and value trade-offs of an AI solution. This is an opportunity for compliance to work with the business to ensure both groups understand what is at stake in terms of value and risk to make an informed decision on the AI solution.</p><p>· Role-Based Access Controls: Limit system access and data visibility based on user role. For example, restrict MSL interaction history from sales rep views to avoid medical-commercial conflicts.</p><h3>2. Design:</h3><p><em>“The way we define guardrails is by providing detailed descriptions, instructions, examples of what good prompts look like and what bad prompts look like. .” — Shaikh</em></p><p><em>Select Risks:</em></p><p>· Careless documentation</p><p>· High-risk prompts going unflagged</p><p><em>Select Control Options:</em></p><p>· Trigger Words or Phrases: An AI agent can be trained using standards, instructions and trigger words to know what types of prompts may be violative.</p><p>· Risk Tiering: Build logic to classify user prompts by risk category. High-risk queries (e.g., referencing incentives or off-label use) trigger internal reviews.</p><p>· Monitoring Blueprints: Too often, monitoring is treated as a post-launch activity, but effective AI oversight begins in design. Compliance will have to monitor responses and prompts, pre-design log capture, red-flag indicators, and escalation flows — mapping out how to track what is asked, how it is answered, and who reviews it.</p><h3>3. Engineer:</h3><p><em>“[Pop-up policy reminders] is the most effective way to get users to review company policies in real-time, but it shouldn’t be the end-all-be-all in risk mitigation, it should be coupled with the other options to enhance risk mitigation.” — Santana</em><br> <br><em> Select Risks:</em></p><p>· Not detecting nuances in language</p><p>· Not connecting to relevant policies</p><p>· Lack of clear escalation path</p><p><em>Select Control Options:</em></p><p>· Automated Escalation: Configure AI to recognize and flag problematic inputs in real time. Define clear ownership of flagged content across compliance, IT and legal.</p><p>· Flag problematic dialogue and permit the user to correct: Allow the system to give real-time compliance nudges (e.g., ‘This response may violate policy X. Would you like to revise your input?’), while enabling users to correct their input — reducing downstream violations and promoting self-guided learning.</p><p>· Automated pop-ups: Connecting users to Company policy in response to a potentially violative prompt is the most effective way to share policies with users in real-time.</p><h3>4. Assess, pilot, deploy and monitor:</h3><p><em>“There [needs to] be some way to streamline the monitoring — a Gen AI agent to monitor a Gen AI agent” — Shaw</em><br> <br> <em>Select Risks:</em></p><p>· Insufficient pilot use cases</p><p>· Not effectively applying feedback</p><p>· Lack of a robust monitoring plan</p><p><em>Select Control Options:</em></p><p>· Root-cause analysis: Begin with small groups to validate controls, gather feedback, and observe behavior in a low-risk setting.</p><p>· Simulation of Use Case Scenarios: Run simulations of edge-case prompts before launch to test how the system handles borderline or intentionally noncompliant queries.</p><p>· Real Time Monitoring: Ongoing oversight should combine manual review with automation — like using one AI to review another — to scale risk detection and mitigation.</p><h3>Compliance as a culture of partnership, not policing</h3><p>The panel closed with a reminder that none of these controls exist in a vacuum. Tradeoffs are continuous: tighten scope, and you lose business value. Increase monitoring, and you risk chilling adoption while overloading burdened compliance resources. Rely on disclaimers alone, and you’ll miss the mark.<br> <br> The panel emphasized compliance partnership over enforcement. Rather than positioning compliance as a reactive gatekeeper, the panelists encouraged embedding compliance as a proactive collaborator. This means engaging early in ideation sessions, co-designing prompts with engineering teams, and helping business leaders understand not just what AI can do, but what it should do. By influencing decisions at inception and throughout, compliance can reduce the burden of risk mitigation later in the lifecycle. The most effective teams will be those who understand how to collaboratively shape the solution — not just sign off on it.</p><h3>Conclusion: A strategic inflection point</h3><p>AI and Gen AI solutions are not a future state — they’re here. Whether compliance functions choose to lean in or hold back will determine their value in AI-driven initiatives. compliance professionals have a unique opportunity to not just be risk mitigators, but strategic enablers — embedding ethical design, trustworthy governance, and real-time guardrails into the next generation of enterprise solutions.</p><p>Watch the full session recording <a href="https://icnk.io/u/GDp6xfuN6xIZ/"><strong>here</strong></a><strong>.</strong></p><p><em>References:</em></p><p><a href="#_ftnref1"><em>[1]</em></a><em> </em><a href="https://chai.org/"><em>https://chai.org/</em></a></p><p><a href="#_ftnref2"><em>[2]</em></a><em> </em><a href="https://www.nist.gov/itl/ai-risk-management-framework"><em>https://www.nist.gov/itl/ai-risk-management-framework</em></a></p><p><a href="https://www.zs.com/insights?utm_source=medium&amp;utm_medium=organic&amp;utm_campaign=insights-medium-promotion&amp;utm_term=20230926&amp;utm_content=insights">Read more insights from ZS.</a></p><p><em>This article reflects our personal views. They do not necessarily represent any official position of ZS.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=408dedf87378" width="1" height="1" alt=""><hr><p><a href="https://medium.com/zs-associates/game-changer-shifting-priorities-and-capabilities-to-mitigate-risks-in-business-gen-ai-solutions-408dedf87378">Game changer: Shifting priorities and capabilities to mitigate risks in business Gen AI solutions</a> was originally published in <a href="https://medium.com/zs-associates">ZS Associates</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Substance over form: Practical approaches to Enterprise Risk Management (ERM)]]></title>
            <link>https://medium.com/zs-associates/substance-over-form-practical-approaches-to-enterprise-risk-management-erm-5d3052784bf8?source=rss-8d298b002793------2</link>
            <guid isPermaLink="false">https://medium.com/p/5d3052784bf8</guid>
            <category><![CDATA[corporate-governance]]></category>
            <category><![CDATA[pharmaceutical-compliance]]></category>
            <category><![CDATA[pharma]]></category>
            <category><![CDATA[pharmaceutical]]></category>
            <category><![CDATA[enterprise-riskmanagement]]></category>
            <dc:creator><![CDATA[Stephanie Santana]]></dc:creator>
            <pubDate>Tue, 10 Dec 2024 22:41:02 GMT</pubDate>
            <atom:updated>2024-12-10T22:41:02.267Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*jnlfqsCXI5gAX0YK" /><figcaption>Photo by <a href="https://unsplash.com/@jasongoodman_youxventures?utm_source=medium&amp;utm_medium=referral">Jason Goodman</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p><em>By: </em><a href="https://www.linkedin.com/in/michaelshawcompliance/"><em>Michael Shaw</em></a><em> &amp; </em><a href="https://www.linkedin.com/in/stephanie-l-santana/"><em>Stephanie Santana</em></a></p><p>The approach to Enterprise Risk Management (ERM) has become over-engineered, resulting in focus on process vs. targeting the risks that count. At the 25th Annual PCF-sponsored Pharmaceutical and Medical Device Ethics and Compliance Congress, ZS’s Global Head of Compliance, Privacy &amp; Risk, Michael Shaw, moderated a panel to discuss how companies can shift away from the academic to the practical approach to ERM. The panel included Bill Benvenuto, SVP Chief Legal Affairs, Compliance &amp; Privacy Officer Travere Therapeutics, Inc., Susan Barsky, VP, Corporate Compliance, BioCryst, and Ed Sleeper Chief Ethics and Compliance Officer, Veloxis Pharmaceuticals. You can watch the full discussion in the linked video and explore the key takeaways from the session below:</p><p><strong>1.</strong> <strong>Sustainable: Creating a sustainable process to consistently identify, prioritize and measure risks being raised</strong></p><p>Companies often approach ERM as a box to tick rather than a comprehensive approach to identifying, assessing, and mitigating risk across the organization:</p><p>ERM is an ongoing process where assessing risk and reviewing systems for managing risk are continuous. The panel shared the importance of embedding a mindset that instills a continuous expectation of accountability to identify and manage risk across all employees. Ed emphasized the importance of ensuring ERM does not become an exercise that <em>“boils the ocean.”</em> ERM should be targeted, focused, and practical. As one audience member shared, <em>“ERM is never truly ‘done’…ERM should be pragmatic to evolve with the business to be continually relevant.”</em></p><p>The ultimate goal is to have a process that minimizes risk and maximizes the possibility of success to achieve business objectives. Providing a holistic view of risks across the organization breaks down silos and ensures all potential threats are considered.</p><p><strong>2.</strong> <strong>Integrated: Integrating ERM with the overall business strategy and culture to break siloes and effectively enable cross-functional triangulation of risk</strong></p><p>By integrating ERM into their corporate strategy, companies can better anticipate risks, ensure compliance, and maintain operational resilience. Bill emphasized this by sharing, <em>“Cross functional [and] holistic is the only way to do ERM…we need buy in from the whole company to help us identify risk.”</em></p><p>A bottoms-up approach works well in many companies where business units identify risks that are then brought up for discussion and action planning at the business unit LT level who then determine risks to be brought to the ERM Committee. To get to this point, a high performing leader should hold their teams accountable to be meeting business objectives <em>and</em> managing risk by imploring them to not only define business objectives but consider the biggest risks that could prevent them from achieving success of those objectives.</p><p>Although ERM is often owned by the Compliance function within the organization, it relies on Board and Executive buy-in to solicit input from its cross-department stakeholders. When considering membership and involvement in the ERM process or an ERM Committee, the right approach will vary company to company. A few options are:</p><p>- The ERM committee serves as your compliance committee</p><p>- Fully separate committees and different members</p><p>- Separate committees, but same members</p><p>Regardless of the structure for your ERM committee, Susan highlighted that membership should include stakeholders who have <em>“the power and authority to navigate risk and take action.” </em>ERM is not a one-time exercise and fostering a risk-aware culture across an organization result in broad accountability for the holistic identification and mitigation of risks.</p><p><strong>3.</strong> <strong>Prioritized: Prioritizing risks and resources based on likelihood and impact, while leveraging both qualitative and quantitative data to derive the full value of risk insights</strong></p><p>Prioritization of identified key risks in both the internal and external environment unlock focused and targeted mitigation. A common approach is to assign a risk score indicating likelihood and impact of risk in alignment with defined risk appetite:</p><p>The panel discussed that this approach should be informed by leveraging both qualitative and quantitative data. This enables organizations to more effectively manage and oversee risk impacting strategic objectives and/or business activities. The panel discussed qualitative and quantitative data can be used to inform whether a risk continues to require the attention of the Board, Executive Leadership and the ERM Committee or if it has been navigated to the point to be managed at the business unit level. But, of course, the process should allow for the risk to move back to the ERM Committee level if the data indicates it warrants that level of attention.</p><p>Bill emphasized this point in response to an audience question regarding when a risk comes off of the ERM Committee list by saying, <em>“…You can measure quantitatively or qualitatively, for example, speaker programs you can measure quantitively, CMC qualitatively…we’ve retired some risks, not that nothing is happening, but it doesn’t require the attention of the Compliance Committee, [Executive Leadership Team] ELT, or the Board because we’ve navigated it for 4 quarters and if it bubbles back up we will navigate it at that level.”</em></p><p>An important aspect to doing this effectively is to consider the resources required to mitigate a risk. Compliance professionals should ask themselves, “does this risk warrant the dedicated resources to mitigate it or can it be effectively mitigated with limited resources at the business unit level?”</p><p>Most companies do not have dedicated resources for ERM, rather it is absorbed into the existing responsibilities of an employee’s role. Technology can help facilitate the ERM process and enable ongoing monitoring; with the complexity and effort tailored to meet the company’s requirements. Susan highlighted that regardless of whether resources are dedicated to ERM, <em>“it’s important to keep it simple for the business teams.”</em></p><p>ERM is an ongoing process where assessing risk and reviewing systems for managing risk are continuous. To further emphasize the takeaways from this session, ZS recommends Compliance professionals to ask themselves these key questions to evaluate whether their approach to ERM is practical and beyond a tick-the-box exercise:</p><p>- <em>Do we have a process in place to </em><strong><em>consistently</em></strong><em> identify, prioritize and measure risks being raised?</em></p><p>- <em>Do we have a </em><strong><em>cross-functional team </em></strong><em>in place to operationalize and integrate ERM?</em></p><p>- <em>Do we have a practice for </em><strong><em>monitoring and reviewing </em></strong><em>our ERM processes that leverages both qualitative and quantitative data?</em></p><p>- <em>How are we </em><strong><em>breaking siloes </em></strong><em>to effectively enable cross-functional triangulation of risk?</em></p><p>Explore their full discussion <a href="https://global-health-permalinks.s3.us-west-1.amazonaws.com/pharma_2024/mini19.mp4">here</a>.</p><p><a href="https://www.zs.com/insights?utm_source=medium&amp;utm_medium=organic&amp;utm_campaign=insights-medium-promotion&amp;utm_term=20230926&amp;utm_content=insights">Read more insights from ZS.</a></p><p><em>This article reflects my personal views. They do not necessarily represent any official position of ZS.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5d3052784bf8" width="1" height="1" alt=""><hr><p><a href="https://medium.com/zs-associates/substance-over-form-practical-approaches-to-enterprise-risk-management-erm-5d3052784bf8">Substance over form: Practical approaches to Enterprise Risk Management (ERM)</a> was originally published in <a href="https://medium.com/zs-associates">ZS Associates</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Harnessing the power of AI for responsible implementation by the business and compliance]]></title>
            <link>https://medium.com/zs-associates/harnessing-the-power-of-ai-for-responsible-implementation-by-the-business-and-compliance-4335d511625c?source=rss-8d298b002793------2</link>
            <guid isPermaLink="false">https://medium.com/p/4335d511625c</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[pharmaceuticals-industry]]></category>
            <category><![CDATA[healthcare-ai]]></category>
            <category><![CDATA[compliance]]></category>
            <category><![CDATA[responsible-ai]]></category>
            <dc:creator><![CDATA[Stephanie Santana]]></dc:creator>
            <pubDate>Mon, 02 Dec 2024 14:34:01 GMT</pubDate>
            <atom:updated>2024-12-02T14:34:01.022Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EShnJXydnzerw5M74O4xDw.jpeg" /></figure><p><em>By: </em><a href="https://www.linkedin.com/in/michaelshawcompliance/"><em>Michael Shaw</em></a><em> &amp; </em><a href="https://www.linkedin.com/in/stephanie-l-santana/"><em>Stephanie Santana</em></a></p><p>The meteoric adoption of Artificial Intelligence (AI) across the pharmaceutical industry necessitates both compliance experts and business counterparts to effectively navigate the convergence of innovation and risk. At the 25th Annual PCF-sponsored Pharmaceutical<br> and Medical Device Ethics and Compliance Congress, ZS’s Global Head of Compliance, Privacy &amp; Risk, Michael Shaw, moderated a panel before several hundred compliance professionals to engage on how to harness the power of AI in a practical and responsible way. The panel included Christopher Corallo, Deputy Chief Compliance Officer, Convatec, Don DiLanno, MBA, Director Global Healthcare Compliance, Insmed, and Nevada Heft, JD, Global Privacy Director, Merck.</p><p>The panel opened the discussion by referencing the <a href="https://chai.org/wp-content/uploads/2024/07/CHAI-Assurance-Standards-Guide-6-26-2024.pdf">Assurance Standards Guide for the Coalition of Health AI</a> (<strong>CHAI</strong>), published by a consensus group of patient advocates, technology developers, clinicians, and data scientists in September 2024, to reinforce the core principles for trustworthy health AI:</p><p>While these principles sound simple for compliance professionals, each one of them has a significant amount of context and associated circumstances in the art of developing technology and AI. To facilitate the discussion, the panel anchored to the evolving ZS Responsible AI Framework pillars of:</p><p>A robust interdisciplinary AI evaluation and development approach serves to foster greater organizational confidence in AI tools — in particular, tools using Generative AI (GenAI) which creates new content based on patterns it has learned from existing data. Working through the Responsible AI Framework pillars is a simple and straightforward path to thoroughly evaluating and mitigating the risk to support innovation in alignment with a company’s risk tolerance. Although various guidance documents may refer to these pillars by different names, the intent of the framework remains consistent. The panel shared the following practical takeaways for each pillar during the session:</p><p><strong>Pillar 1: Define</strong></p><p>Evaluating the intended purpose of AI, data sensitivity, and intended audience is the first step in the path to responsibly leveraging AI. Key mitigation solution considerations for Legal and Compliance functions at this stage that were highlighted by panelists include:</p><p>· <strong>Adhering to laws</strong> (e.g., EU AI Act) that designate some use cases as inherently risky and mandate specific additional requirements; and</p><p>· Evaluating the <strong>kind of data</strong> an AI solution it is handling and differentiating the risk associated with <strong>different audiences</strong> that AI tools are addressing.</p><p>Heft highlighted the criticality of including <strong>contract language</strong> to protect an organization’s data beyond standard confidentiality terms as well as establishing a <strong>federated review model</strong> that incorporates stakeholders from across the organization to govern AI solutions. Heft highlighted the importance of establishing risk criteria and differentiating between “<em>“inherent risk that cannot be reduced and has automatic additional obligations due to the type of activity” and “risk that can be reduced by proper safeguards.”</em></p><p>DiLanno shared how establishing an <strong>executive steering committee</strong> can help review AI proposals at a high-level to evaluate use cases appropriateness, compliance, and feasibility. The panel was aligned on how confirming an organization has <strong>sufficient resources</strong> to implement and monitor an AI solution is critical to the responsible and effective development of the solution.</p><p>Corallo emphasized, <em>“The other thing that’s important when you’re defining is [determining whether AI] is a nice to have or is it need to have…based on your business model, beyond the person that’s creating the business case, do you have the resources that are technically capable of implementing and monitoring and supporting the AI?”</em></p><p><strong>Pillar 2: Design</strong></p><p>Ethical tool design helps to establish appropriate boundaries and controls to detect tool errors and vulnerabilities. Laying out step by step how the AI is to be used while <strong>balancing system capabilities</strong> and risk mitigation is a critical first step to designing AI responsibility. Panelists highlighted the following additional risk mitigation considerations:</p><p>· Confirming a <strong>reporting policy</strong> is in place to address AI solution deviations, as appropriate, and a retention plan is in place for queries or other data generated by an AI solution;</p><p>· Assessing how an organization’s <strong>other systems</strong> will interact with an AI solution;</p><p>· Using <strong>prompt engineering</strong> to set boundaries of AI solution scope; and</p><p>· Having capabilities within the compliance function to <strong>interact with IT</strong> department throughout the design of the AI solution.</p><p>A tangible and publicly available example of the importance of a targeted, interdisciplinary approach to the design of AI can be seen in the <a href="https://www.myeloma.org/">International Myeloma Foundation</a> (IMF) chatbot, Myelo. This chatbot was designed to capture volumes of information about multiple myeloma disease state education and clinical research, publications, etc. Panel moderator, Shaw, acknowledged <em>“this may sound scary, but with prompt engineering, you’re able to make the boundaries so when a patient asks, ‘Of all the clinical research, what are the significant gaps that still have not been figured out?’ The chatbot will respond with a synthesis of all of the clinical data in a safe, responsible way while being empathetic and including the appropriate disclaimers.”</em> Shaw highlighted that this is the type of innovation and tool that compliance officers will need to get accustomed to and strengthen their capabilities to enable.</p><p><strong>Pillar 3: Develop</strong></p><p>Compliance officers need to consider methods to evaluate AI tool performance and assess for risk mitigation needs<strong><em> </em></strong>as<em> </em>the AI is developed. Risk can still be mitigated with disclaimers as an initial form of defense, but only as part of a bigger framework of risk mitigation controls.<strong> </strong>It is during this stage that the mindset of compliance shifts<strong> </strong>from seeing AI solutions as pure risk to appreciating how AI solutions may be able to mitigate risk better than traditional ways that business activities are monitored. AI solutions can offer a more sustainable approach to continuous compliance. For example:</p><p>· Generative AI is currently being used to offer pharmaceutical sales representatives a way to safely and confidently engage with a virtual healthcare professional and practice interactions that deliver educational messages about company medicines. This Virtual Sales Rep Trainer offers an alternative to live role plays normally conducted between sales representatives at national and regional meetings. While the Virtual Sales Rep Trainer can offer an effective way to practice approved education and messages, it could also address non-compliant interactions more effectively than traditional role playing if controls are built within tool covering various tiers of risk:</p><p>o In response to a low level of risk, a sales rep could get a tailored response at the end of their session</p><p>o Next level of risk could be routed to a sales rep’s first line manager to reinforce proper messaging and company policies</p><p>o Highly problematic interactions could be triaged directly to the compliance function</p><p>Corallo highlighted, <em>“It’s important [to] really set it up in the right way that all stakeholders, before it’s even launched, have a real appreciation for why are we doing this in the first place? How is it going to help you, but also [how it makes] us be more efficient in providing services or support.”</em></p><p><strong>Pillar 4: Monitor</strong></p><p>Monitoring for AI is not just ongoing monitoring that all compliance professionals are familiar with; it’s the monitoring conducted as the product is developed. Enabling a process for user feedback and human interventions in AI tool management enhances tool capabilities and ensures compliant use during piloting, deployment, and an ongoing basis. A few practical examples of monitoring considerations discussed during the panel were:</p><p>· Using <strong>piloting</strong> to assess real world impact as you develop the product (e.g., development testing, overseeing people trying to constructively “break” the AI solution);</p><p>· Mitigating <strong>false-negatives</strong> by checking small subsets of data; and</p><p>· Recognizing the challenges and <strong>resource burdens</strong> associated with monitoring:</p><p>o Projects will need to be re-reviewed to ensure compliance with laws as laws change;</p><p>o Transcripts will need to be reviewed as part of periodic auditing of logs; and</p><p>o Cross-functional oversight will need to be leveraged to further refine and mitigate ongoing risks.</p><p>Heft emphasized, <em>“There are so many different aspects of monitoring during development and after launch that I think it has more administrative burden than the other three steps combined.”</em></p><p>Alignment to <strong>Enterprise Risk Management</strong> and proportionate allocation of resources will be critical factors to enabling AI. DiLanno shared, <em>“When leveraging the power of AI, we must always remember the importance of governance, ethical practices, and transparency in AI implementation, while also highlighting the need for interdisciplinary collaboration, continuous monitoring, and iterative improvement to ensure responsible AI use”.</em></p><p>AI — how it’s built, used, and evolves- intrinsically carries risk and it is impossible to mitigate 100% of the risk. Accordingly, Compliance and Legal functions should not view AI use cases with a zero-risk mentality. As Heft shared during the panel, “risk is not inherently evil; it is something to be evaluated and mitigated.” This is a unique crossroads for compliance functions where compliance teams can choose to get scared and pull back from leveraging AI or can choose to acknowledge the opportunity to become an effective and strategic business enabler as an expert in responsible AI. AI and GenAI solutions are here, and instead of rejecting innovation, compliance professionals need to embrace it, build capabilities, and learn how to best develop practical mitigation controls throughout development and monitoring processes.</p><p><a href="https://www.zs.com/insights?utm_source=medium&amp;utm_medium=organic&amp;utm_campaign=insights-medium-promotion&amp;utm_term=20230926&amp;utm_content=insights">Read more insights from ZS.</a></p><p><em>This article reflects my personal views. They do not necessarily represent any official position of ZS.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4335d511625c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/zs-associates/harnessing-the-power-of-ai-for-responsible-implementation-by-the-business-and-compliance-4335d511625c">Harnessing the power of AI for responsible implementation by the business and compliance</a> was originally published in <a href="https://medium.com/zs-associates">ZS Associates</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>