Scaling Human Impact with AI: Four Lessons from Social Entrepreneurs on the Frontlines

Personalized social impact is finally affordable. So why are most ventures still struggling to deliver it? This is the question that kicked off our Masterclass, "Scaling AI for Social Innovation".

<iframe title="vimeo-player" src="https://player.vimeo.com/video/1145687989?h=9e3c4819de" width="640" height="360" frameborder="0" referrerpolicy="strict-origin-when-cross-origin" allow="autoplay; fullscreen; picture-in-picture; clipboard-write; encrypted-media; web-share" allowfullscreen></iframe>

When Bruce Elliott's CTO demonstrated that generative AI could recreate their entire product—5,000 personalized games for people with dementia—in a single weekend, Elliott faced an uncomfortable question: What was his venture's competitive advantage if the technology itself had become commoditized?

Memory Lane Games had spent years building an app downloaded 100,000 times across 100+ countries, helping people with dementia maintain connections through memory-based games. The platform showed clinical evidence of improving socialization, not just entertainment. But if AI could replicate their core offering so easily, were they suddenly irrelevant?

Elliott's anxiety reflects a broader moment of reckoning. What cost $100 to process with AI in mid-2023 now costs roughly $10—a 90% drop in 18 months. Personalized education for millions, 24/7 healthcare guidance in resource-constrained settings, career coaching for underserved populations—services that would have required $3-5 million in infrastructure just years ago are now accessible to small organizations with limited budgets.

For social entrepreneurs, this represents an unprecedented opportunity. The gap between mission and capability has collapsed. Yet most ventures struggle to capitalize on this moment. The bottleneck is no longer technology or capital—it's deployment wisdom.

The Paradox: Accessible Technology, Elusive Impact

Three social entrepreneurs recently participated in a masterclass, hosted by MovingWorlds and featuring world-renowned entrepreneur Claudio Sassaki to explore this deployment challenge. Beyond Elliott's dementia care platform, the group included Geraldine Kyazze, whose MyMedikoz platform addresses medical record fragmentation for 1.2 billion uninsured Africans, and Agnieszka Czmyr-Kaczanowska, whose Talenti provides AI-powered career guidance to women navigating systemic barriers in Central and Eastern Europe.

All three had functioning AI solutions. All three faced adoption challenges that had nothing to do with accessing technology. Their struggles revealed a critical gap: as AI capabilities democratize, success depends less on technical sophistication and more on strategic deployment—understanding user needs, building trust, allocating resources effectively, and knowing when to orchestrate versus build.

Claudio Sassaki, who built Geekie (a Brazilian edtech company that used AI to personalize learning for 12 million students) and now serves as entrepreneur-in-residence at Stanford University, identifies the shift: "Innovation isn't building the hammer anymore. It's knowing exactly where to swing it." In the Masterclass

The implications extend beyond individual ventures to the entire social impact ecosystem. Impact investors, accelerators, and capacity builders designed to support innovation still largely evaluate ventures based on proprietary technology rather than strategic deployment. This misalignment leaves entrepreneurs with the wrong kind of support at precisely the moment they need practical guidance most.

Four Insights Emerging from the Frontlines

1. Strike While the Economic Window Is Open

The cost trajectory of AI capabilities isn't just declining—it's collapsing exponentially. What this means in practice: ventures that establish user bases and prove models now will see margins expand dramatically as costs continue dropping, while those who wait for "perfect" infrastructure risk missing the first-mover advantage entirely.

Elliott initially asked, "How can we keep our personalized AI offering affordable to deliver maximum impact globally?" The question presumed that current costs were prohibitive. But deeper analysis revealed a different dynamic. His AI companion currently costs roughly $6 monthly per user. According to Sassaki, processing costs, currently about 2 cents per minute of conversation, will likely drop to 0.005 cents within 18 months.

The strategic implication: rather than over-optimizing current costs, Elliott should focus on user acquisition and proving value. As costs decline, margins will naturally expand—but only if he has established market position. The window for differentiation is now, before every competitor can easily replicate core functionality.

This economic reality creates urgency that many social entrepreneurs underestimate. Geraldine Kyazze spent months perfecting technical features for MyMedikoz before launching, wanting the product "ready." But in rapidly evolving AI markets, perfectionism becomes procrastination. The entrepreneurs gaining traction are those deploying imperfect solutions now and iterating based on user feedback, rather than waiting for technical perfection that will be outdated within months.

The challenge for accelerators and funders: traditional milestone-based support (build MVP, then pilot, then scale) may be too slow for AI-driven markets where capabilities evolve monthly. Ventures need support structures that enable rapid iteration and deployment.

2. Lead with Outcomes, Hide the Infrastructure

When Kyazze won a presidential award for AI innovation in Uganda, she initially led with technology in her communications—showcasing "AI chatbot capabilities" and "handwriting-to-PDF conversion"—to build credibility. Her inbox soon filled with warnings that AI was "demonic" and messages steeped in superstition about technology.

The lesson wasn't about technology skepticism—it was about positioning. Users don't care about your LLM architecture. They care whether you solve their problem. Instead of "AI-powered chatbot," Kyazze now says "your personal health assistant." Instead of "AI converts handwritten records," she says "take a photo, get it organized instantly."

This principle extends beyond avoiding negative associations. Even in contexts without AI skepticism, leading with technology distracts from value. When Elliott described Memory Lane Games, he emphasized personalized games and conversational AI. But families don't pay for technology—they pay for connection. His platform has the potential to promise something different: weekly audio clips of a loved one's most treasured memory, captured by an empathetic companion that adjusts to emotional cues. The AI enables this outcome, but the outcome is the value proposition.

Czmyr-Kaczanowska faces skepticism from another direction: impact investors who question whether using Google's Gemini + RAG constitutes "real innovation." Her response reveals the positioning challenge many ventures face. Building a proprietary language model would cost millions and take years. Using Gemini costs $0.002 per conversation, enabling her to serve thousands of women for the cost of a single engineer's salary.

Sassaki pointed out that her innovation lies elsewhere: proprietary career data from 10,000 women, bias-aware prompt engineering that counteracts gender bias in LLM training, human-AI hybrid models that route complex cases to mentors, and regional labor market integration. Netflix didn't build the internet. Apple doesn't manufacture chips. Strategic orchestration of existing infrastructure is innovation.

But communicating this requires different language for different audiences. For users: outcomes and benefits. For investors: defensible advantages and strategic positioning. For partners: complementary capabilities. The failure mode is using the same technology-forward language for all audiences, assuming sophisticated stakeholders care about infrastructure rather than strategic deployment.

3. Obsessive User Focus Despite Infinite Possibilities

Perhaps the most counterintuitive insight: AI's infinite possibilities become a liability without ruthless focus on specific user needs.

Elliott's platform could generate games on unlimited topics, in multiple languages, with varying difficulty levels, for different cognitive abilities. The temptation was to build features showcasing this flexibility. But what actually reduces churn? What makes the service irreplaceable to families?

Through user research, Elliott discovered the answer wasn't more games or features—it was emotional connection. Families needed to feel connected to loved ones they couldn't visit daily. They needed relief from guilt. They needed reassurance that their relative was engaged and experiencing moments of joy. The 45-second weekly audio clip of treasured memories addressed this need. Additional game variations didn't.

This focus becomes especially critical given resource constraints. Social entrepreneurs face the universal challenge of searching for product-market fit before running out of money. The instinct when uncertain is spreading efforts across multiple approaches, hoping something resonates. But this guarantees mediocrity. With AI enabling rapid feature development, the discipline of saying "no" becomes paramount.

Kyazze struggled with this. MyMedikoz started as a personal health records platform but added consultation features and health forums to increase engagement. Different clinic partners wanted different modular features. The temptation was offering everything to everyone. But fragmented offerings dilute brand, complicate positioning, and prevent doing any single thing exceptionally well.

Sassaki’s guidance that was the most pertinent to all – even if scary to implement: pick one flagship feature, validate that clinics will pay for it, then expand. Don't launch five modules before proving one. Choose your best bet, focus obsessively on that user need, validate the model, then expand. Otherwise limited resources get spread too thin to achieve excellence in anything.

For accelerators, this suggests a different support model. Rather than encouraging ventures to pilot multiple approaches simultaneously, effective support might mean helping entrepreneurs identify which single bet to focus on, then providing intensive resources to validate that quickly.

4. Build Trust Before Features—The Human-AI Hybrid Imperative

The ventures achieving traction aren't choosing between AI or humans. They're building hybrid models where AI handles scale while humans provide wisdom, nuance, and trust.

Czmyr-Kaczanowska's Talenti illustrates this. Their AI assistant Zoe provides 24/7 initial career guidance, answers common questions, and helps women articulate goals. But career breaks, imposter syndrome, and wage negotiation require human mentors who understand cultural context and power dynamics. The community validates AI advice—"Zoe suggested this path; here's my experience with it"—creating feedback loops that improve recommendations while building trust through peer experience.

This hybrid approach addresses a fundamental challenge: trust building takes time that resource-constrained ventures often lack. The shortcut is leveraging existing trust networks. Kyazze partners with Rotary International, community health workers, and local clinics—institutions that already have credibility. These intermediaries become the face of the service, not the technology.

When Sassaki built Geekie, his team partnered with famous Brazilian educators who had established public trust. They worked with religious institutions, influencers, and community groups that had no economic gain to advocate for their product. Once these influencers endorsed the solution, school adoption accelerated dramatically. The pattern holds across contexts: identify who users already trust, position those intermediaries as champions, and let them vouch for outcomes rather than explaining technology.

This has implications for measuring success. Traditional metrics focus on user acquisition and engagement. But trust-building metrics might be more predictive: How many trusted intermediaries actively refer? How strong is community validation of AI recommendations? What's the ratio of human touchpoints to AI interactions? Ventures optimizing for rapid user growth without trust infrastructure often achieve impressive numbers but struggle with retention and word-of-mouth growth.

The deepest insight about trust and features came from Sassaki's coaching: "Build trust before you build features." Ventures instinctively want to prove value through feature sophistication. But users don't trust you because of features—they trust you because you consistently deliver on a simple promise, then expand. Elliott focused on one irreplaceable thing for families. Kyazze validated one modular feature clinics would pay for. Czmyr-Kaczanowska proved value through a free tier before optimizing pricing.

Systems-Level Implications: Rethinking Support for AI-Driven Ventures

These insights reveal a misalignment between how social entrepreneurs need support and how impact ecosystems typically provide it.

For Impact Investors: The ventures to bet on aren't those building proprietary AI models—they're those demonstrating strategic deployment. Questions to ask: What's your defensible advantage beyond technology? How are you building trust with users? What makes you irreplaceable even as AI capabilities commoditize? Can you articulate why strategic orchestration is more valuable than infrastructure ownership?

For Accelerators: Traditional programs emphasize either technical AI training ("here's how to build models") or general business fundamentals ("here's how to create a business model canvas"). But entrepreneurs with functioning AI solutions need strategic deployment guidance: positioning, pricing strategy, trust-building approaches, go-to-market tactics. The gap is practitioner wisdom from those who've navigated similar challenges, not more technical training or generic business frameworks.

For Capacity Builders: As AI democratizes technical capabilities, the scarce resource shifts from technology access to deployment wisdom. Effective support increasingly means connecting entrepreneurs with experienced practitioners who can provide strategic guidance on their specific context—not just training modules or consulting frameworks, but hands-on coaching through actual deployment challenges.

For the Field: We need new frameworks for evaluating innovation. Proprietary technology made sense when infrastructure was expensive and capabilities were scarce. But in an era of commoditized AI, innovation lies in strategic resource allocation, deep user understanding, trust-building capability, and knowing what to build versus what to orchestrate. Our evaluation criteria should reflect this.

The New Scarce Resource

When costs of powerful capabilities drop 90% in 18 months, what becomes valuable? Not access to technology. Not capital to build infrastructure. The scarce resource becomes wisdom about strategic deployment, or as Sassaki said, innovation is not building another hammer, “it’s knowing where to swing the hammer”.

This wisdom is inherently contextual. It comes from having navigated similar challenges in adjacent spaces. Sassaki brought value to these entrepreneurs not because he knows more about AI than they do, but because he built an AI-powered edtech company in a developing market, dealt with skeptical users, wrestled with B2B and B2C models simultaneously, and struggled through pricing and positioning challenges.

The implication for building support systems: as AI capabilities democratize, the highest-value intervention shifts from providing capital or technology to connecting entrepreneurs with practitioners who have relevant deployment experience. Not training. Not consulting frameworks. Strategic coaching through actual implementation challenges from people who've been there.

The entrepreneurs scaling AI-driven social impact aren't building proprietary models or waiting for perfect infrastructure. They're deploying existing tools strategically, building trust effectively, focusing obsessively on user needs, and proving value relentlessly. The real innovation isn't in the technology itself—it’s becoming more human in an AI-driven world.

The three organizations featured—Memory Lane Games (dementia care), MyMedikoz (medical records in Africa), and Talenti (women's career advancement in Europe)—participated in a masterclass examining AI deployment for social impact, facilitated by MovingWorlds. The session was led by Claudio Sassaki, entrepreneur-in-residence at Stanford University and former CEO of Geekie. While their specific challenges differed across sectors and geographies, common patterns emerged about strategic deployment, trust-building, and the support systems social entrepreneurs need as AI capabilities democratize.

Social Entrepreneurs are encouraged to join MovingWorlds to access support from skilled professionals around the world.