By Christian Meyer zu Natrup
Founder, MzN International | Managing Partner, Human Planet

Here’s the challenge, 4 concrete use-cases and 1 big warning about our AI work

Not too long ago, I stood in a crowded tent on the outskirts of Erbil, northern Iraq. The camp had grown overnight, again, and the local UN team, already stretched, was struggling to track who had arrived, what they needed, and which donors might help fund the next phase of support. A young programme officer sat cross-legged on the floor, toggling between WhatsApp messages, paper notes, and three open spreadsheets. I asked what she needed most. She said: “Someone to help me think.”

Her response stuck with me. Across 15 years of aid and development work, from South Sudan to Cox’s Bazar, I’ve seen incredible people doing extraordinary work with inadequate tools. Not because they lack talent or dedication, but because our systems still assume that impact is delivered manually, case by case, one grant at a time. That model no longer scales. The world moves faster than our spreadsheets.

That’s why this article is about something different: How we bring AI agents into the hands of those doing the work, not to replace them, but to help them breathe, think, and lead with more clarity and confidence.

The Challenge: Doing More with Less

In an era where expectations are rising faster than budgets, mission-driven organisations are being asked to deliver more impact with fewer resources. Leaders face relentless pressure to stretch limited teams across growing programmatic demands, funding requirements, and compliance protocols. Traditional models of growth no longer suffice. Something fundamental has to change.

In 2023, over 80% of NGOs reported struggling with administrative complexity and staff overload¹. Meanwhile, UNDP notes that funding fragmentation has increased by 27% over the past decade², multiplying tasks without harmonising requirements. The result? Missed opportunities, slow delivery and growing burnout.

This is not a failure of purpose, it ist’s a failure of systems. We must shift away from constant firefighting.

Innovative finance and AI offers a path not just to efficiency, but to scale and better decision-making, but not overnight and only if done right. From scanning global funding landscapes to flagging early programme risks, it allows teams to act earlier and with greater clarity. Like any major technological shift, the outcome depends not on the tech itself, but on how thoughtfully we manage its integration.

The stakes are high, as is the potential. Done right, this is not about machines replacing missions. It i’s about letting people do their best work more often. Here are four specific use cases we currently work with UN agencies, NGOs and social enterprises on:

A New Colleague: AI Enters the Workforce

Until recently, we’ve treated technology as a set of tools, helpful, yes, but clearly separate from the human heart. But what happens when software can write, analyse, predict, and learn? We are not just adding new tools. We are adding new team members, digital ones, that shift how we work, collaborate and lead.

AI is not a tool, it is a team member. This shift changes how we assign responsibility, structure collaboration, and even define accountability. When AI agents handle donor matching, reporting, or risk scanning, they do not merely automate, they enable the team to do better things. AI agents we currently trial with MzN partners already take on discrete, high-cognitive-load tasks and return insights, allowing human colleagues to focus on judgment, empathy, and leadership. In that aspect the AI colleagues are just like any new teammate: their value grows when its role is well-defined and integrated into daily workflows.

Capabilities beyond automation. AI Aagents are fast, but they do not just speed things up, they bring up what humans might miss. Pattern recognition across thousands of records, databases, predictive analytics for project risk, or multilingual donor engagement are not just efficiencies, done right they’re new superpowers for programme management teams. They also support localisation as the knowledge gap between the HQ and the country offices is bridged.

Building Better Programmes

At MzN, we have now begun designing data analysis agents specifically for our partners. These agents are built to interpret programme data dynamically, making sense of fragmented inputs to inform decisions. So far so good.

This summer, we will match this intelligence with data from the World Bank, major funders, and our impact investor community (where the data is public only). The goal: to build better programmes, aligned from day one with both community needs and viable funding pathways. Our data selection to achieve this is very careful here and not necessarily less work than designing the programme manually. But this approach ensures that programmes are designed with ALL lessons learned (not just those we know about) embedded from the start and with fundability at their core. The results so far are impressive. We have seen a real quality increase in programme design work, which helps for more great ideas for impact to become reality, not just proposals on paper.

It is helpful not just for designing, but running programmes too. After all, programmes do not fail because we lack good intentions. They falter because we often act too late. AI can turn real-time data into responsive decisions, helping programme teams adjust course before problems escalate. It is not hyperbole to state that the era of post-project evaluation is giving way to dynamic, insight-driven delivery because AI enables real-time feedback loops, scenario modelling, and early warning alerts. It basically transforms monitoring from retrospective to responsive

Access to more funds: Agentic AI for Funding That Finds You

Access to funding should never depend on expensive consultants or insider networks. Yet many of the most promising ideas stall, not because they lack substance, but because people cannot find the right funders.

A real game-changer here has not been the usual general AI chatbot, which offered more hype than precision, but agentic AI: purpose-built digital teammates trained to understand the funding ecosystem, country and organisation’s strategy. These systems are custom built and do more than text generation. Carefully set up, they can listen, learn, search, and match continuously. This marks a shift from chatbots to agents acting with contextual relevance, IF they are designed to serve the real needs of both funders and implementers.

At MzN, we are currently designing specialised funding and investor-matching agents for all our Funding Support partners. These agents tailor searches to each organisation’s funding history, programme focus, and strategic priorities, dramatically lowering the cost of identifying opportunities while increasing their relevance and timing. Crucially, this helps level the playing field between large, well-resourced organisations and smaller, often overlooked ones.

Our aim is to make our funding support more accessible, regardless of size or visibility. This is urgent. Only 12% of small NGOs receive direct bilateral funding, compared to 68% of large INGOs¹. Many are excluded not for lack of impact, but due to limited bandwidth, visibility or sheer lack of time to look and network with funders.

By designing specific agentic AI systems customised for each of our non-profit partners, we make the search for funding smarter and more strategic. And a lot cheaper too, enabling us to lower our prices further.

From Reports to Results: Automating the Bureaucracy

Reporting obligations in the nonprofit sectors are increasing, yet the resources to meet them rarely follow. A 2022 Humentum study found that 59% of NGO staff report spending less time on actual programme delivery due to the growing weight of donor compliance demands.

At MzN, our own research shows that over 60% of partners see proposal and report writing as their greatest operational stress point.

Part of the solution is not generic automation, but agentic AI systems designed not simply to follow instructions, but to understand, learn, and act within the funding ecosystem from the start. Unlike traditional tools that act as glorified digital secretaries, agentic AI can independently draft reports, format logframes, and align programme data with donor requirements in real time. Our early trials with some of our partners indicate a 20% reduction in staff time spent on reporting in the first few months. This is set to rise as the AI colleague learns to do more, reducing the administrative burden while increasing both accuracy and relevance. After all, what we really need to invest our time in is learning, adaptation and delivering real impact.

Learning While Doing: Real-Time Insight for Real Impact

The true value of learning lies in its ability to improve decisions. It would be great to not just learn after the action, but during it. This is particularly true in development and humanitarian settings, where conditions shift quickly and stakes are high, learning must be immediate, iterative, and embedded. With the right agentic AI systems in place, organisations no longer have to wait for end-of-project evaluations to understand what’s working.

Agentic AI acts as an embedded evaluator, constantly analysing programme data, surfacing insights, and closing the feedback loop in real time. These systems can detect emerging patterns across project documents or other data, enabling benchmarking, peer learning, and strategic adaptation at speed.

This is not about replacing human judgment, but about enhancing it. It offers frontline teams the clarity to course-correct while they work, not after it is too late.

Our approach to Responsible AI: Proceed with Purpose, Not Hype

AI promises transformation, but history teaches us that powerful tools often come with unintended consequences. In the rush to embrace the latest breakthrough, many overlook the risks: hidden biases, misplaced trust, and decisions made without context. Agentic AI, while offering tremendous potential, also raises the stakes. These systems can act on their own initiative, amplifying not just insight, but also any embedded flaws.

A 2023 Stanford study found that some AI models used in healthcare and humanitarian analysis exhibited bias in over 30% of test cases[6]. When deployed at scale without scrutiny, AI can unintentionally reinforce inequality, prioritise speed over fairness and widen gaps in access and power.

At MzN, we believe that trust and inclusion are essential, so this is not a risk we can afford. We therefore need to test carefully and not rush to market as soon as it looks good.

Larger organisations are particularly exposed. With more data, broader reach and public visibility, the impact of a flawed AI deployment can be profound. Rushing to implement agentic systems without rigorous testing or ethical safeguards turns innovation into a liability. Imagine a grants officer relying on an AI-generated forecast, only to realise too late that it was trained on biased or incomplete data. Our caution is not conservatism; it is responsibility.

At MzN, we treat responsible design as a core principle. Our “Responsible AI Integration” protocol requires on the ground stakeholder engagement, transparent logic, continuous audits, and ethical governance from day one. Human-centred design, much like our human centred consulting approach, is not an add-on, it is the basis of everything we do.

Beyond the Buzz: A Measured Path to AI-Native Impact

That said, I believe that we are at a crossroads. Not only in how non-profit agencies use technology, but in how they are structured and led. The future NGOs, foundations, UN-agencies or social enterprises may be AI-native. But the real challenge is for it to remain mission-led and values-based. No artificial system replaces human values. Otherwise, we risk building systems optimised for efficiency but blind to values.

McKinsey’s 2023 report notes that social sector organisations using AI well can achieve a 20–40% improvement in process efficiency and data-driven decision-making [7]. But those numbers mask the deeper truth: technology alone does not create impact. Leadership, accountability, and design choices do. AI will not save us from bad strategy or poor judgement. Done wrong, it may in fact amplify it.

The promise is real, but only if pursued with humility, purpose and discipline.

References

[1] Humentum Annual Report, 2023. “Administrative Overload and the Capacity Crisis.”
[2] UNDP Report, 2022. “Fragmented Aid Delivery and Its Impact.”
[3] Devex, 2023. “The Equity Gap in Donor Funding.”
[4] MzN AI Pilot Results, Internal Memo, 2024.
[5] Humentum Survey Results, 2022. “NGO Staff Time Use and Stress.”
[6] Stanford University, 2023. “Bias in Machine Learning for Social Applications.”
[7] McKinsey & Company, 2023. “The State of AI in the Global Development Sector.”