How to Advance Innovation Beyond Pilot Phase

Why AI Pilots Don’t Scale in Healthcare: A Framework for Thinking Clearly Many AI pilots in healthcare start with promise and end in silence. It’s not because they lack accuracy or ambition. It’s often because they lack architecture. Or alignment. Or adoption. McKinsey’s Five Frames of Transformation can help us think clearly about how to move from isolated AI pilots to scalable, embedded solutions. When adapted to healthcare, this becomes less about technology and more about systems. Here’s how I’ve come to see it: 1. Aspire Start with purpose, not tools. The right question isn’t “How can we use AI?” It’s “What patient outcomes are we trying to improve, and why?” For example: “Reduce diabetes-related amputations through AI-enabled early detection in polyclinics.” Meaning gives AI relevance. 2. Assess What is our starting point? Do we have usable data, engaged clinicians, and the right infrastructure? Readiness is often overestimated. Before building anything, diagnose the system honestly. 3. Architect Design with the end in mind. Create a cross-functional operating model. Align procurement, governance, ethics, and clinical workflows. Involve clinicians early. Build explainability in from day one. Think platform, not pilot. 4. Act Deploy gradually. Use shadow mode. Pair AI tools with human judgment. Refine based on user feedback and unintended consequences. Implementation is not a one-time event. It is continuous calibration. 5. Advance Sustain what works. Monitor for model drift. Institutionalize training and feedback loops. Move AI from innovation to infrastructure. Build learning systems, not just smart systems. In Singapore, as we deepen Healthier SG and move toward a team-based, preventive model of care, this framework reminds us: AI is not the transformation. It is a tool within it. The transformation is cultural, clinical, and structural. What frameworks or insights have helped you scale AI or digital innovations in healthcare? #HealthTech #AIinHealthcare #PublicHealth #HealthierSG #ClinicalAI #SystemsThinking #NHG #DigitalHealth #MBBS #MPH #MBA #INTP #Enneagram5

A successful NHS pilot is not a product launch. Getting a pilot in the NHS feels like a win. But the problem is pilots don’t scale. Not unless you plan for scale from day one. I’ve seen it too many times: ✅ Great clinical engagement ✅ Strong early outcomes ✅ Excited feedback from frontline teams Then… nothing. Why? Because success in one Trust doesn't equal system readiness. If you want to move beyond the pilot, here's what really matters: Budget lineage - Who funds this after the pilot? Can they? Will they? Integration credibility - Did it really work integrated in the EPR… or did someone babysit it into production? Multi-level buy-in - It's not enough to have a champion. You need operational, financial, and IT sign-off before you scale. System fit, not just user love - Pilots thrive on enthusiasm. Rollouts depend on infrastructure. Evidence in the language that matters - Think CQUINs, KPIs, workforce impact - not just UX love or NPS. The NHS doesn't scale innovation because it likes you. It scales what solves real, recognised pain - with minimal risk. So ask yourself early: Can this be embedded? Funded? Owned? If the answer is "not yet" - great. Start fixing that, not just the feature list. Because in the NHS, pilots prove promise. But procurement only follows a plan.

The healthcare landscape is filled with brilliant insights and promising pilots that never scale. As human-centered designers, we excel at uncovering needs and creating compelling solutions—yet implementation remains our greatest challenge. Transforming promising pilots into widespread practices represents a profound opportunity to shape healthcare's future. When innovative approaches successfully scale, they create ripple effects—enhancing patient experiences, improving outcomes, and often reducing burden on care teams. Our opportunity lies in developing implementation approaches as thoughtful as our initial designs. Institutional inertia often presents the first major hurdle. Overcome this by starting with targeted 8-week interventions that demonstrate immediate value. Identify informal leaders who shape culture—the veteran nurse or respected physician whose opinions influence others. Create visual artifacts that make pain points undeniable and build emotional connection to the need for change. Regulatory concerns require thoughtful navigation. Invite compliance partners into design sessions from day one, giving them ownership in finding solutions. Distinguish between actual requirements and accumulated practices—you'll often find more flexibility than assumed. Consider modular implementation where less-regulated components can advance first. Address the human element of implementation. Design changes that reduce workload in visible ways—for every new step added, eliminate two. Create a "change budget" that acknowledges the cognitive costs and limits concurrent initiatives. Develop frontline champions who receive dedicated time for implementation support. For measurement challenges, create simple dashboards that include both traditional and experience measures. Develop visual data stories showing impact through multiple perspectives to build a compelling case. Establish 30-day feedback cycles where users shape refinements. When moving from pilot to scale, build solutions with a stable core and flexible edges that adapt to different contexts. Document "implementation recipes" with specific steps and resource requirements. Connect implementation teams across sites to share adaptations and solutions. By addressing these barriers with practical strategies, we can accelerate human-centered innovation in healthcare—moving from isolated bright spots to transformative change at scale.

Rare moment... OpenAI’s new “From Experiments to Deployments” guide actually reflects how AI gets built and deployed in the real world. Their four-phase approach... foundations, fluency, prioritization, then build and scale, tracks almost exactly with what many of us have been talking about… what it actually takes to move AI from experimentation to production at scale. It all starts with the foundation. 1️⃣ The guide opens with something that can’t be emphasized enough… executive alignment, governance designed for speed, and data access come before the flashy deployments. Not as a blocker, but as the enabler for everything else. They’re explicit about this. Companies that skip this groundwork “move fast at first but stall when gaps appear.” And the governance framing is exactly right… it’s governance for motion, not gatekeeping. Create fast paths for experimentation while your frameworks mature alongside real use cases. Once you have that, then build fluency everywhere. 2️⃣ Phase 2 is about creating organizational literacy before expecting innovation to scale. Not theoretical workshops… hands-on practice, champion networks, daily office hours to answer questions. BBVA is a perfect example: 3,000 → 11,000 ChatGPT licenses, 83% weekly usage, 2,900 GPTs created. That only happens when fluency reaches critical mass. And OpenAI calls out something most reports miss. Those SMEs you develop in Phase 2 are the ones who later define what “good” looks like as use cases get more complex. Then comes structured intake and prioritization. 3️⃣ Phase 3 tackles the “now we have 500 ideas” problem. Open intake, but disciplined scoring around impact, effort, risk, and reuse. That reuse lens is critical. Every pattern you build once should accelerate multiple future projects. Without it, you end up with duplicate agents, duplicate data, duplicate context, and duplicate risk. Finally, incremental building with continuous evaluation. 4️⃣ Phase 4 gets into the real operating rhythm… gated checkpoints (MVP → Pilot → Production), constant evaluation, and iterative improvement. OpenAI is clear here: AI systems don’t improve by scaling infrastructure. They improve by iterating the product itself… refining prompts, workflows, evals, grounding data, failure modes, cost efficiency. The core message is the same many of us have been seeing in the field. --> Organizations stuck in pilots are usually waiting for conditions that will never be perfect. --> Organizations reaching production put structure around experimentation and learn by doing. And the guide is honest about what actually drives scale: not the latest model, but the less glamorous work… building skills, capturing learnings, and creating reusable systems. Definitely worth reading if you’re in the “how do we make this repeatable?” phase rather than the “should we even be doing this?” phase.

If you’re building GenAI pilots inside a bank or any large enterprise, they rarely fail because of the model. They fail because of how the organization works. Every pilot looks great in month one. By month four, it’s usually stuck between innovation and operations, waiting for someone to decide who owns it and who’s paying for it. You want it simpler? Here’s what actually keeps pilots alive: ✅ A business owner who owns the KPI you’re improving ✅ Controls and audit trails built in from day one ✅ A path from sandbox to production that Infrastructure already agreed to I’ve seen this from both sides, leading pilots inside a global bank and advising founders trying to land there. And the issues are almost always the same. 1️⃣ Innovation runs experiments. Operations runs the business. And in In most financial institutions, innovation teams can greenlight a 3-month POC for ~$200K. But the moment you want to scale, the budget sits with Ops or Risk, not Innovation. If Operations isn’t in the loop early, the money disappears the day the pilot ends. That’s why 9 out of 10 pilots never make it past phase one. 2️⃣ Compliance always comes in late and stops everything. The demo may go well, and then audit walks in and will be asking: – Where’s the decision log? – Can we isolate outputs by region? – Who signs off when something goes wrong? Each one of those can add 4–6 weeks. If you start addressing them after the pilot, your timeline just doubled. 3️⃣ Data access is never what you expect. That “production data” everyone promises usually ends up partial, masked, or locked behind three approvals. Data owner, legal, and infosec. Each review can take 10–15 business days. So do the math, 2 to 3 sprints delay just there. By the time you actually get access, half your pilot window is gone. And even then, you’ll be working with 20–30% of the real data fields. 4️⃣ The handoff is where good ideas die. Innovation celebrates the POC, Ops hesitates because they weren’t part of the design, and Infra says, “not our environment.” Moving a pilot from dev to production in a global tier-1 bank can take 3–4 months assuming you have approval from change control and security review. The plans rarely take this into account, most startups never plan for that either, and that’s where the energy and momentum disappears. 5️⃣ Quarter-end comes, budgets reset, and your sponsor moves on. That happens in every large company. If you haven’t shown a measurable win like hours saved, risk reduced, false positives cut, it’s gone. Show me the money. Not the innovation. And you’ll be starting again with a new stakeholder who’s never heard of you. It’s not the model that decides if AI succeeds, but everything that happens after it runs like ownership, controls, data, and timing. That will separate a demo from a real deployment. 👉 I help financial institutions get that part right — from strategy to adoption, minus the hype.

4 ways to move your GenAI project from pilot to production—that no one's talking about! Are you stuck in the same loop: launching promising GenAI pilots but never quite making the leap to full-scale production. From my experience of helping leaders scale their GenAI projects, here’s what I think you should do differently. Beyond the usual focus on data quality, governance, and scalability, these overlooked strategies can accelerate your GenAI journey from pilot to production: 1️⃣ Leverage Shadow Deployments: Run your GenAI models in the background, comparing predictions against actual outcomes. This helps refine accuracy without disrupting live systems. 2️⃣ Adopt a “Minimum Viable Model” Mindset: Start small, with just enough functionality to prove value. Scale up based on real-world feedback, not assumptions. 3️⃣ Create a Sandbox for Business Users: Build a safe environment where non-tech teams can experiment with GenAI applications and identify high-impact use cases. 4️⃣ Design for Simplicity: If users can’t understand it, they won’t use it. Build experiences that feel intuitive—because complexity kills adoption. Remember, the difference between a pilot and a breakthrough is how boldly you’re willing to explore beyond the usual playbook.

Scaling is not a bigger pilot. It’s a new product and a new way of working. Your AI Strategy should change the workflow, not just optimize it. Update SOPs, KPIs, and incentives so people use the system by default. Train at scale. Communicate like adults. Close the loop on feedback weekly. Harden the backbone as you scale: - stable data pipelines, - security and privacy checks, - performance SLOs and on-call, - monitoring with alerts that someone owns. Right-size controls to impact: - meeting-summary bot → light checks, fast rollouts, - pricing recommender → validation, rollback plan, - credit-limit engine → rigorous testing, audit trail, human override. Ship with a one-page factsheet: purpose, data sources, known limits, and owner. Use human-in-the-loop when the stakes are high. Keep logs. Prove control. Good governance doesn’t slow you down; it removes debate and speeds you up. Want speed or rework? → Subscribe to my newsletter starting this week: https://lnkd.in/dEwSKCYw #AIstrategy

𝐀𝐈 𝐩𝐢𝐥𝐨𝐭𝐬 𝐝𝐨𝐧’𝐭 𝐟𝐚𝐢𝐥 𝐛𝐞𝐜𝐚𝐮𝐬𝐞 𝐭𝐡𝐞 𝐦𝐨𝐝𝐞𝐥𝐬 𝐚𝐫𝐞 𝐰𝐞𝐚𝐤. 𝐓𝐡𝐞𝐲 𝐟𝐚𝐢𝐥 𝐛𝐞𝐜𝐚𝐮𝐬𝐞 𝐭𝐡𝐞 𝐞𝐧𝐭𝐞𝐫𝐩𝐫𝐢𝐬𝐞 𝐢𝐬. The whitepaper I wrote for the The Data Institute, University of San Francisco makes this plain: C-suites are discovering that AI isn’t a technology upgrade. It’s a business transformation. And that’s where the gap lies. 𝐓𝐡𝐞 𝐩𝐫𝐞𝐝𝐢𝐜𝐭𝐚𝐛𝐥𝐞 𝐟𝐚𝐢𝐥𝐮𝐫𝐞 𝐩𝐚𝐭𝐭𝐞𝐫𝐧𝐬: 1. 𝐏𝐢𝐥𝐨𝐭 𝐭𝐫𝐚𝐩. Models in the lab, never in production. 2. 𝐒𝐭𝐫𝐚𝐭𝐞𝐠𝐢𝐜 𝐦𝐢𝐬𝐚𝐥𝐢𝐠𝐧𝐦𝐞𝐧𝐭. AI because “others are doing it,” not because it solves a business problem. 3. 𝐎𝐫𝐠𝐚𝐧𝐢𝐳𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐫𝐞𝐬𝐢𝐬𝐭𝐚𝐧𝐜𝐞. Systems that work, but people who won’t. 4. 𝐈𝐧𝐟𝐫𝐚𝐬𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞 𝐢𝐧𝐚𝐝𝐞𝐪𝐮𝐚𝐜𝐲. Data stacks built for reporting, not real-time AI. 𝐓𝐡𝐞 𝐩𝐥𝐚𝐲𝐛𝐨𝐨𝐤 𝐟𝐨𝐫 𝐥𝐞𝐚𝐝𝐞𝐫𝐬 𝐰𝐡𝐨 𝐰𝐢𝐧: 1. 𝐒𝐭𝐚𝐫𝐭 𝐰𝐢𝐭𝐡 𝐛𝐮𝐬𝐢𝐧𝐞𝐬𝐬 𝐢𝐧𝐭𝐞𝐠𝐫𝐚𝐭𝐢𝐨𝐧. Anchor AI in one critical business problem before chasing use cases. 2. 𝐓𝐫𝐞𝐚𝐭 𝐭𝐫𝐚𝐧𝐬𝐟𝐨𝐫𝐦𝐚𝐭𝐢𝐨𝐧 𝐚𝐬 𝐨𝐫𝐠𝐚𝐧𝐢𝐳𝐚𝐭𝐢𝐨𝐧𝐚𝐥. Train, adapt roles, reset culture. 3. 𝐁𝐮𝐢𝐥𝐝 𝐩𝐥𝐚𝐭𝐟𝐨𝐫𝐦𝐬, 𝐧𝐨𝐭 𝐩𝐨𝐢𝐧𝐭 𝐬𝐨𝐥𝐮𝐭𝐢𝐨𝐧𝐬. Reuse infrastructure, governance, and data across functions. 4. 𝐋𝐞𝐚𝐝 𝐟𝐫𝐨𝐦 𝐭𝐡𝐞 𝐂-𝐬𝐮𝐢𝐭𝐞. CEO for patience and vision. CTO for scalable architecture. CFO for funding with accountability. CHRO for workforce evolution. The first 120 days decide everything. Either you build alignment, assess capabilities, plan realistically, and start building systematically… Or you drift into the same wasteland where 95% of enterprise pilots go to die. AI advantage compounds. Those who scale early will lock in moats that laggards cannot cross. 𝐂𝐡𝐨𝐢𝐜𝐞 𝐟𝐨𝐫 𝐞𝐯𝐞𝐫𝐲 𝐞𝐱𝐞𝐜𝐮𝐭𝐢𝐯𝐞: Commit to transformation, or accept decline. Half-measures won’t save you. 𝐒𝐡𝐨𝐰 𝐮𝐩 𝐰𝐢𝐭𝐡 𝐀𝐈 𝐭𝐡𝐚𝐭 𝐰𝐨𝐫𝐤𝐬 𝐚𝐜𝐫𝐨𝐬𝐬 𝐭𝐡𝐞 𝐛𝐮𝐬𝐢𝐧𝐞𝐬𝐬, 𝐧𝐨𝐭 𝐣𝐮𝐬𝐭 𝐢𝐧 𝐭𝐡𝐞 𝐥𝐚𝐛. 𝐎𝐫 𝐝𝐨𝐧’𝐭 𝐬𝐡𝐨𝐰 𝐮𝐩 𝐚𝐭 𝐚𝐥𝐥. If you want to move beyond experiments, explore the AI-First Organizations program at University of San Francisco Data Institute. And if your enterprise is struggling with this transition, reach out to me directly. Paul Intrevado Thomas Maier Ph.D Jamie Wheeler Elisabeth Merkel Baghai