The U.S. Department of Agriculture is taking on a larger role in U.S. international food assistance at a moment when expectations are high.
Food for Peace is now being administered through USDA’s Foreign Agricultural Service, and USDA is actively considering what its future non-emergency programming should look like. They want to know how programs can strengthen resilience, reduce future aid needs, and improve evidence on what works.
Those are the right questions.
The renewed focus on helping countries become more self-reliant also creates an opportunity to build agricultural programs differently. This is also in the U.S. national interest: stronger agricultural economies abroad can improve global food security, create new markets for American goods, and reduce the need for costly humanitarian assistance over time. Emergency food assistance will remain essential in crises. But where the goal is to help smallholder farmers move toward greater productivity, income, and resilience, decades of research point to several practical lessons.
Evidence from Mathematica’s evaluations of agricultural programs across Africa and Asia, combined with MFAN’s longstanding focus on effective, accountable U.S. assistance, point to three principles to guide USDA design and implementation of agriculture investments for durable results.
First, fund the whole pathway out of dependency—not isolated pieces of it
Smallholder farmers rarely face just one constraint. A farmer may lack access to improved seed, but seed alone will have limited impact if the farmer cannot afford fertilizer, does not know how to use the new variety, lacks reliable weather information, cannot irrigate during a dry spell, or has nowhere profitable to sell the additional crop.
That is why agricultural programs built around a single input or training often disappoint. Addressing one, or even two, constraints simply exposes the next one.
The stronger evidence points toward integrated approaches that tackle several barriers at once. That can mean combining access to quality inputs with extension, finance, irrigation or other infrastructure, farmer organizations, and links to buyers. Mathematica’s evaluations across sub-Saharan Africa have repeatedly found that these complementary investments are more likely to produce adoption and durable gains than technical assistance alone.
This matters especially if USDA’s goal is to reduce long-term aid dependency. Success should not be measured only by whether farmers received U.S.-supported inputs or attended training. It should be measured by whether they can keep producing, financing, and selling when the investment ends.
Second, put partners with deep local knowledge in the driver’s seat
Agricultural interventions do not travel perfectly from one context to another. A model that works in one country—or even one district—may perform very differently elsewhere because farmers face different markets, social norms, production systems, infrastructure, and binding constraints.
As evaluators, we see this repeatedly. Technologies that appear promising on paper can stall when they do not fit how farmers actually farm, buy inputs, sell products, or make decisions. Training can be technically sound but ineffective if the recommended practices require land, finance, labor, or services that farmers do not have. And approaches that rely on unfamiliar technology or delivery channels may struggle to build the trust needed for adoption.
That is why locally rooted organizations should be more than advisers to externally designed programs. Partners with a long-term presence often understand how markets work, which institutions people trust, where implementation is likely to break down, and how programs need to adapt as conditions change. They also have a stake in what remains after donor funding ends.
Government agencies, social enterprises, local community groups and faith-based organizations can all play this role in different ways. Government actors understand public systems, policy constraints, extension structures, and what can realistically be sustained at scale. Social enterprises combine local relationships with revenue streams that allow services to continue after grant funding ends. Faith-based organizations often have longstanding community networks and trust.
The evidence does not suggest that one organizational model will always work best. It does suggest that programs are more likely to endure when the institutions implementing them understand the context deeply, have incentives to stay, and have real authority to shape the intervention—not just validate decisions made elsewhere.
Third, create a portfolio-wide learning partner so USDA can manage for results, not just compliance
Good program design is only the beginning. USDA also needs to know which investments are working. That requires more than relying on individual implementers’ monitoring reports and it requires broadening a focus beyond simply preventing waste, fraud and abuse.
A portfolio-wide Learning and Evidence Partner for future non-emergency Food for Peace awards could support USDA and implementers from the outset by synthesizing existing evidence before new awards are designed, helping define a common set of meaningful indicators and targets, establishing the data systems needed to measure them, planning independent evaluations where they will add value, and pulling lessons together across the full portfolio.
Where feasible, independent evaluations should compare outcomes against a credible counterfactual—what would have happened in the absence of the program. These do not always require expensive randomized trials. Depending on the investment and the question USDA needs to answer, quasi-experimental designs, geospatial impact evaluations using satellite data, or mixed-methods studies may be more appropriate. Planning for these approaches from the beginning makes them both stronger and less costly.
Just as important, a learning partner can help test whether enthusiasm for a new approach is supported by evidence before it becomes a portfolio-wide bet. Digital agriculture is a good example. There is considerable excitement around AI-enabled weather forecasts, market information, and advisory tools because they may be able to deliver highly localized information at low marginal cost and enormous scale. That promise is real. But our evaluations also show that sophisticated tools can underperform when farmers cannot easily use them, do not trust the source, or prefer familiar channels and human intermediaries.
For USDA, a portfolio-wide learning function could answer questions that individual evaluations cannot by identifying patterns across investments. One project may tell USDA whether a particular irrigation program worked. Across 10 agriculture investments, however, a learning partner could ask: which combinations of finance, extension, market access, and infrastructure consistently produced income gains? Which approaches continued after subsidies ended? Where did programs fail because another binding constraint was left unresolved?
A chance to build for self-reliance from the start
As USDA takes on this new responsibility, it has an opportunity to make Food for Peace even more effective—delivering greater returns on U.S. foreign assistance investments by helping countries build more productive, resilient agricultural economies and reducing their need for future assistance. Achieving this will require designing programs for long-term sustainability from the outset.
Fund programs that address the multiple constraints farmers actually face rather than isolated pieces of the problem. Give partner governments and locally rooted organizations real authority to shape implementation and build toward a transition they can sustain after U.S. support ends. Invest in information, but deliver it through simple, trusted, user-centered systems. And establish an independent, portfolio-wide learning function that can identify which investments produce measurable results and value, which approaches are durable, and where programs need to adapt.
Those choices can increase the likelihood that agriculture programs do what long-term development assistance is ultimately supposed to do: help more households reach the point where they no longer need it.
