What it actually costs to serve a smallholder

Published costs of reaching a smallholder farmer in East Africa run from under one dollar a year to over four thousand. That is not a range with an average in the middle. It is a set of different products being priced as if they were the same thing, and it is why programme budgets and delivery reality keep parting company somewhere around month eighteen.

The numbers, on one page

Every figure below is from a published source. The point of putting them side by side is not to pick one.

Four things in that table matter more than the individual numbers.

1. The published number is usually the subsidy, not the cost

One Acre Fund publishes a core programme deficit of USD 20 per farmer for 2024. That is a real and useful number, and it is not the cost of serving a farmer. It is what remains after farmers have paid.

Working from the same published accounts: core programme costs of USD 131.6 million across 2,100,043 farmers give USD 62.67 per farmer per year. Farmers themselves contributed USD 90.5 million, or USD 43.08 each, which is 68.7 per cent cost recovery. The difference between USD 62.67 and USD 20 is not a discrepancy. It is the farmer’s own contribution, netted off before publication.

Inside that USD 62.67, the part that most budgets treat as overhead comes to USD 19.53 per farmer: field operations at USD 22.2 million, programme support at USD 16.2 million, overhead allocation at USD 2.7 million. In other words, the human infrastructure of last mile delivery costs almost exactly what the entire visible donor subsidy costs.

The same organisation’s cost per farmer was USD 90.14 in 2023 and USD 62.67 in 2024. It fell 30 per cent in a year purely on scale, with field operations per farmer falling from USD 19.09 to USD 10.56. Unit cost is a function of density, and density is the thing a two year programme in a new district does not have.

2. The distribution is skewed, so budgeting on an average underfunds the tail

IDH’s FarmFit data set is the largest published body of evidence on this question, covering more than 100 inclusive business models, most of them in Africa. Its service delivery cost per farmer runs from USD 2.50 to USD 4,382, with a median of USD 77 and a mean of USD 276.

A mean three and a half times the median is a heavily right skewed distribution. Programme budgets are almost always built on a single planning figure applied uniformly across a target population. Applied to a skewed distribution, that systematically overfunds the farmers who are easy to reach and underfunds the ones who are not, which are the same farmers the programme exists to serve.

The same data set records that businesses working with unorganised farmers spend 170 per cent more per farmer than those working with organised groups, and that models which charge farmers for services invest up to seven times more per farmer than models that give services away. It also finds that scaling to more farmers improves efficiency and cost recovery but frequently reduces service quality per farmer. Cheap per farmer and good per farmer are in tension, and the tension is measurable.

3. The line items that get left out

Field staffing, at a ratio that actually works. A fully loaded extension officer in the Tanzanian tea case costs USD 9,764 a year. At the observed ratio of one officer to 700 farmers that is USD 13.95 per farmer. At the programme’s own target of one to 300 it is USD 32.55. The ratio is the budget. Uganda’s public extension service runs at roughly one worker to 1,900 farming households against a recommended one to 500. Kenya’s 2023 extension policy targets one worker to 600 farmers by 2029.

And the projection is usually wrong in one direction. In IDH’s Uganda grain case, 89 agents were projected. 484 were ultimately needed to serve the target farmer base, a factor of more than five.

The first mile. Measured in Kenya and Tanzania, head loading costs 16 to 23 times more per tonne kilometre than a truck, and motorcycles cost around 12 times a truck on the second leg. First mile transport takes 10 to 20 per cent of farmer income in the Kenyan case and 20 to 30 per cent in the Tanzanian dry season, rising to 40 to 50 per cent in the wet. A separate study puts transport’s bite at 30 to 40 per cent of net farm income for Tanzanian crops. Regionally, EAC freight runs near USD 1.80 per kilometre per container against an international benchmark near USD 1.00, with transport and logistics at 35 to 42 per cent of production cost against roughly 8 per cent in Asian comparators.

A budget that funds training and inputs but not the movement of the resulting crop has funded the part that does not generate income.

Registry decay. There is no published attrition rate for a national farmer registry anywhere in East Africa, and no published recurring cost of keeping one current. That absence is itself a finding. The available proxies point the same way: in the Tanzanian tea case, 2,800 farmers were active out of 6,147 registered, 54 per cent dormant. In Kenya’s KAPP, about 60 per cent of common interest groups and 67 per cent of cooperatives were dormant after the project closed. In Ethiopia, of 6,486 farmer training centres built, 2,380 were fully functional. Budgets count registrations. Registrations decay at something like half over a few years, and nobody budgets the re-registration.

The cost of an outcome versus the cost of a contact. A Uganda market systems programme reported USD 31.05 per direct beneficiary and USD 63.08 per beneficiary achieving a real income increase. Exactly double. Programme logframes count the first number. Funders think they are buying the second.

4. Reach can be bought cheaply. Outcomes cannot.

AGRA’s PIATA programme spent USD 113 million between 2017 and 2021, with USD 84.5 million on systems development, roughly USD 10 per targeted smallholder. It reached over 10.1 million farmers against a 9 million target, through more than 30,000 village based advisors. The independent evaluation found that it “did not meet its headline goal of increased incomes and food security for 9 million smallholders, despite reaching over 10 million smallholders”. Maize yields improved in some countries and not in Kenya or Tanzania.

At the other end of the spending scale, Kenya’s KAPP I cost about USD 814 per beneficiary and produced a 6.6 percentage point gain in hybrid maize adoption and a 4.3 point gain in fertiliser use on maize. KAPP II halved its beneficiary target mid project, from 400,000 to 200,000, and cancelled USD 16 million.

The systematic review evidence on farmer field schools puts the underlying problem plainly: they change practice and raise yields in pilots, they “have not been effective when taken to scale”, and non-participating neighbouring farmers do not benefit from knowledge diffusion. The assumed multiplier that makes a low cost per farmer budget arithmetic work does not appear in the evidence.

What happens when the subsidy is withdrawn

This is the question that matters most in 2026, with official development assistance down 23.1 per cent in real terms in 2025 and a further decline forecast.

The clearest published illustration is a Kenyan dairy lender in the IDH case series. With donor support its operating self-sufficiency was 121 per cent in 2016 and 91 per cent in 2017. Without that support the same years were 56 per cent and 52 per cent. Net margin moved from plus 7 per cent to minus 78 per cent. Cost to serve was rising at about 16 per cent a year with donor support and about 30 per cent a year without it. Break even required roughly 1,400 farmers, a fivefold increase on its then 200, or 600 farmers at market rate interest, or a 35 per cent cut in training cost.

A well run smallholder lender at 200 farmers covers about half its costs. That is not a failing organisation. That is the unit economics of the last mile, stated honestly.

Where the money in a programme actually sits

• ◆ In One Acre Fund’s 2024 core programme, field operations plus programme support plus overhead were 31 per cent of total cost.
• ◆ IFAD reported administrative expenditure to portfolio at 16.5 per cent against a 12.5 per cent target for 2024, with efficiency rated moderately satisfactory or better in only 73 per cent of completed projects.
• ◆ USAID’s localisation reporting put USD 2.1 billion, or 12.1 per cent, through local partners in FY2024, against a 25 per cent target, and that calculation covered only 43 per cent of total resources.
• ◆ In a Rwandan nutrition programme benchmarked against cash transfers, total cost was USD 141.84 per eligible household, of which about 40 per cent was village level expenditure on WASH and behaviour change rather than household level intervention. After a year it had not moved any primary outcome, while a comparable cash transfer had.

Six questions we would ask of any last mile budget

1. Is the cost per farmer figure gross or net of what the farmer pays? If it is net, it is a subsidy figure and it is roughly a third of the real cost.

2. What agent to farmer ratio does the budget assume, and what does it cost to hold it? Not the target ratio in the design document. The one the budget actually funds in year three.

3. Is the first mile funded? Movement from farm to collection point, at wet season prices, not dry.

4. Is there a re-registration line? If the registry is assumed to hold its value for the life of the programme, the budget is wrong by something like half.

5. Is the target a contact or an outcome? If the target is farmers reached and the theory of change is income, the cost per unit of the thing you actually want is roughly double the cost per unit of the thing you are counting.

6. What is the plan at the point the subsidy ends? Stated as a farmer number and a price, not as a sustainability paragraph.

A note on the evidence

We would rather say what is missing than fill it in. There is no published figure we could verify for: the cost of quality assurance and grading per tonne in East Africa; warehouse receipt storage and handling fees per tonne; aflatoxin testing cost per sample; the cost of loan origination and collection per smallholder; interest and fee structures for the major input credit providers; or churn in farmer registries. ISF Advisors, reviewing the state of smallholder finance in October 2025, says it directly: “data on the financial performance of smallholder finance models is scarce, with even fewer insights into the relevance of revenue and cost drivers.”

Published cost per beneficiary figures are also not on a common denominator. An IFAD rural roads programme in Uganda reported USD 19 to 25 per beneficiary, spread across a large mostly indirect population. KAPP reported USD 314 to 814 per directly served farmer. Both are correct. Neither is comparable to the other, and both get quoted in the same sentence.

This is the gap HAGRO’s own work is aimed at. We publish what our trading and delivery operation costs because a sector that cannot price its own last mile cannot budget for it. The figures above are the public benchmarks. A future note will set our own against them.

Sources

One Acre Fund, 2024 Financial Performance, July 2025 · IDH FarmFit, service delivery cost per farmer, and the Ikanga Tanzania, Landmark Millers Uganda and ECLOF Kenya service delivery model cases · Precision Development, 2024 annual report · Mathematica, PIATA evaluation for AGRA, February 2022 · World Bank IEG, Kenya Agricultural Productivity Program I and II, January 2019 · IFAD, Report on IFAD’s Development Effectiveness 2024 · IFAD IOE, Uganda CAIIP-1 · TechnoServe, East Africa Coffee Initiative II impact audit · DLEC and Digital Green, Village Agent Model study, Uganda, September 2019 · Hine et al., Overcoming the First Mile, ReCAP, 2015 · 3ie, farmer field schools systematic review summary · ISF Advisors, Beyond the Frontier, October 2025 · McIntosh and Zeitlin, Gikuriro cash benchmarking, Rwanda · Publish What You Fund on USAID localisation, January 2025 · OECD, preliminary 2025 ODA data, April 2026 · Kenya Agricultural Sector Extension Policy, 2023.

The all in One Acre Fund figures are derived from published line items in its 2024 financial performance statement, not published as a cost per farmer by the organisation itself. Cost per beneficiary figures from different evaluations use different denominators and are not directly comparable. Several sources are dated and carry their year in the table above.