Cost-Effectiveness of Sanku Food Fortification: An Analysis of Current Programs and a Framework for Expansion

Editorial note: We thank Sanku for commissioning and funding this research report. The views expressed herein are not necessarily endorsed by Sanku.

Summary and Key Takeaways

This report describes Rethink Priorities’ analysis of the present and hypothetical impacts and cost-effectiveness of Sanku’s food fortification program.

We model the health and economic impacts of Sanku-fortified maize and wheat flour in three countries (Tanzania, Kenya, and Ethiopia) where Sanku is active, and provide a quantitative and qualitative framework for assessing potential expansion into seven further countries (Burundi, Malawi, Mozambique, Rwanda, Uganda, Zambia, and Zimbabwe).

Key takeaways:

• Tanzania is the most cost-effective country in which Sanku is active, with a cost-effectiveness of 4.6x GiveWell’s cash transfer benchmark. Sanku’s program in Tanzania averts 20 deaths, 495,348 anemia cases, and an estimated 17,419 disability-adjusted life years (DALYs) annually, at a net cost of $207 per DALY. It causes ~492,399 IQ point losses averted for children and adults combined, which together contributes to a wage gain of ~$10.4 million annually. Iron was by far the greatest contributor to benefits from the four key nutrients we focused on, both in terms of health and economic benefits.
• We estimate that Sanku’s cost-effectiveness in Tanzania will rise to 8.2x, exceeding GiveWell’s funding bar, if they were able to make a full switch in their premix formulation from ferrous fumarate to sodium iron (iii) ethylenediaminetetraacetate (NaFeEDTA). NaFeEDTA has substantially higher bioavailability than ferrous fumarate, meaning that our estimate for cost-effectiveness increases by 80% even after accounting for its higher costs. Because these projected costs are only approximate, this estimate comes with greater uncertainty than our results for Sanku’s current cost-effectiveness. We discuss the operational implications and Sanku’s stated preference for a constant-volume formulation strategy. We note that the 8.2x figure assumes the switch applies across Sanku’s full flour volume, and should be treated as an upper bound. Sanku has acknowledged that NaFeEDTA is most practical for small maize mills, and that large wheat mills are not currently a feasible target for the transition given dosing and equipment constraints.
• Sanku’s programs in Ethiopia and Kenya have lower cost-effectiveness, at 1.2x and 0.8x GiveWell’s cash transfer benchmark, respectively. Kenya’s cost-effectiveness is constrained by high assumed counterfactual fortification among industrial millers. The assumed counterfactual fortification comes with substantial uncertainty, such that cost-effectiveness would be significantly higher under plausible alternative values for counterfactual fortification. Ethiopia’s cost-effectiveness is dampened by the absence of iron in its mandatory fortification standard (and therefore the premix used by Sanku), which eliminates modeled economic benefits.
• While our quantitative results suggest that Mozambique and Zimbabwe offer the greatest potential cost-effectiveness for potential Sanku programs, given high uncertainty, these should be considered alongside qualitative factors. Key drivers influencing cost-effectiveness that are not considered in our model include the likely extent of scale-up, the requirement to build new premix factories, and malaria prevalence.
• There are substantial uncertainties in parts of our model. The most important are likely to be: (i) the estimate for counterfactual fortification; (ii) the impacts of fortification on relative risks for key health outcomes; and (iii) our assumption of a linear dose-response relationship between intake and health impacts. Other approaches exist for estimating health impacts from fortification, which involve different assumptions to the ones presented here. We have tried where possible to be conservative in our estimates, but our assumptions may either upwardly or downwardly bias our headline results. In particular, we exclude the indirect impacts that Sanku’s work may have on fortification by non-Sanku partner mills (e.g., through network effects and support for government enforcement); this may drive our total estimated benefits downwards.

Model Structure, Inputs, and Assumptions

Data Collection

The data underpinning our modeling work was sourced from:

• Sanku’s operational data: Sanku shared with us their internal data on flour production and reach, as well as costs broken down into auditable line items.
• Desk research: Using peer-reviewed academic papers, with a preference for Cochrane studies reporting on randomized controlled trials (RCTs);
• Expert interviews: We completed interviews with John Kangarua, Tom Mboya, Brian Mwandau, and Edson Justin (all Sanku), as well as Riz Yusufali (TechnoServe), Marvin Jaeger (Lignopure), Ty Beal (Global Alliance for Improved Nutrition), and Demewoz Woldegebreal (University of Washington).

Modeling Framework

Our general modeling approach involves estimating impact metrics, including health impacts and economic impacts, and then assessing Sanku’s operational costs in generating these impacts. Dividing the impacts by the costs allows us to quantify overall cost-effectiveness. Figure 1 summarizes this general framework.

Figure 1: (left) Overview of modeling approach, from impacts to overall cost-effectiveness; (right) Countries included in our analysis, showing active countries in green and expansion countries in purple.

The impact and cost-effectiveness metrics we calculate are listed in Table 1. These are calculated at the country level: both for the three countries in which Sanku primarily operates, and for seven expansion countries to which Sanku is considering expanding its programs in the future (Figure 1).

Table 1: Impact metrics estimated by our model.

Health Impacts

Our model estimates health impacts from Sanku’s food fortification program for each country in the following way (these steps are summarized in Figure 2).

Our model’s general structure is designed to:

1. Calculate (or for expansion countries, estimate) the monthly output of Sanku-fortified maize and wheat flour, using Sanku administrative data;
2. Use consumption data broken down by age and sex to estimate consumption of Sanku-fortified flour for each demographic group;
3. Estimate additional micronutrient intake from consumption of Sanku-fortified flour, after subtracting estimated nutrient intake under the counterfactual that Sanku was not active;
4. Translate this additional intake into proportional risk reductions using meta-analytic estimates for the impacts of fortification on the risk of deficiency and related outcomes, scaled linearly by the ratio of per-capita Sanku-delivered intake to the intake in the underlying studies;
5. Multiply the resulting risk reductions by national burden estimates for deficiency and key outcomes of interest from the Institute for Health Metrics and Evaluation (IHME)’s Global Burden of Disease study, to obtain country-level health impacts.

Figure 2: Main stages in health impact modeling.

Further detail on these stages, along with specific data inputs and assumptions made in each stage, is provided below.

Consumption

For active Sanku countries, we use Sanku’s administrative tracking data (internal link provided by Sanku) to estimate the monthly output of Sanku-fortified maize and wheat flour. This is based on sales of Sanku’s nutrient premix, which is sold to millers both on its own and—to smaller maize millers in Tanzania—in a bundle with flour bags manufactured by Sanku. We assume that the appropriate concentration of premix is added to flour to estimate the total amount of flour fortified using the Sanku premix sold that month: for instance, we assume that for premix where the appropriate concentration is 500ppm or 200ppm, millers do in fact add the premix at the correct concentration. Given that the dosage of premix to flour is also monitored by Sanku, we are confident in this assumption. We take the average across the most recent six months (October 2025 to March 2026) to mitigate lumpiness in monthly output while also accounting for Sanku’s recent growth.^[2] We divide by 30 to calculate daily output, to align with the nutrition data.

We assume that all Sanku-fortified flour is consumed by humans, specifically within the country where it is produced. This assumes that no flour or premix, or any of its products, are wasted, fed to animals, or used for brewing, and that no flour is exported to other countries. This is a standard approach used also by Millers for Nutrition (interview with Riz Yusufali).

Our estimates for hypothetical flour consumption from a Sanku program in expansion countries are inherently highly uncertain. In our default setting, we assume that consumption of Sanku-fortified flour will be in proportion to the country’s population, with the ratio of flour consumption to population size equivalent to the average across the three Sanku countries. This parameter is discussed further in the Framework for Expansion Countries section.

We then combine this with estimates of the breakdown of maize and wheat flour consumption by specific age-sex groups, listed in Table 2. These groups are based on categories for recommended nutrient intake, adjusted to match the constraints of IHME age groups. We use data on calorie requirements for individuals in each group sourced by Sanku,^[3] interpolating between age groups where necessary, and estimate wheat and maize consumption for individuals in each group by assuming that the share of calories contributed by wheat and maize flour does not vary across groups. It is likely that for infants and very young children, a smaller share of calories is contributed by flour. In our final results, we found that infants do not contribute greatly to our estimates of impact, so this assumption is not very influential.

Table 2: Age-sex groups and daily calorie intake used in our model.

Nutrient Intake

Sanku’s premix is designed to meet national standards, which vary by country. The nutrient levels in premix for the three active Sanku countries are listed in Appendix 2. The premix sold in Kenya contains nine micronutrients in total, while the premix sold in Ethiopia contains seven of these micronutrients, and Tanzania only four. The concentration of nutrients in fortified flour varies between maize and wheat. It is noteworthy that Ethiopia’s fortification program does not include iron, because the Ethiopian government’s mandatory standard excludes iron on the basis that Ethiopian diets contain meaningful iron-bearing meat and teff.

We only model the health impacts of four nutrients: iron, zinc, Vitamin A, and folic acid. In our review of the evidence, impacts of these nutrients were supported by the strongest evidence, based on two factors: whether a systematic review demonstrated clear evidence for impacts from strong underlying studies; and whether the impact was likely to be of substantial size, given the context of East Africa and the doses provided through Sanku fortification. This selection was also informed by the availability of country-level burden estimates from Global Burden of Disease (GBD) 2023 study, which formed the foundation of our modeling approach.

While there is evidence for impacts from the other five nutrients (Vitamin B1, Vitamin B2, Vitamin B3, Vitamin B6, and Vitamin B12), there was not sufficient evidence from high-quality systematic reviews for effects large enough to substantially contribute to our estimates of effectiveness. There was also substantial overlap with the primary pathways in our analysis, risking double-counting of benefits.^[4] We therefore do not include the potential benefits of these impacts in our model. This downwardly biases our estimate for Sanku’s total health impact; however, we think that it is almost certain that benefits from the four nutrients we model capture the majority of the benefits from Sanku fortification, and that they likely capture more than 75% of the benefits.^[5]

To calculate nutrient intake per kilo of Sanku flour consumed for active Sanku countries, we use the nutrient content target provided in internal specification data from Sanku (Appendix 2). For expansion countries, we input (as is part of Sanku’s strategy) that Sanku-fortified flour would meet national standards for the four key micronutrients we focus on; for this purpose, we use data on national standards from the Global Fortification Data Exchange (GFDx).^[6]

Counterfactual Fortification

Based on conversations with Sanku and Riz Yusufali (TechnoServe), we concluded that the status quo of flour fortification in East Africa is best modeled as a dual system (Figure 3). Small-scale mills, defined by the Global Fortification Data Exchange (GFDx) as those with a grain processing rated capacity below 20 metric tonnes per day, engage in negligible fortification. This is due both to low enforcement of regulatory standards against these millers and to their tighter profit margins, making the additional costs of purchasing and dosing premix difficult or impossible to bear. For industrial-scale mills, the level of fortification varies by country: some share of mills will fully comply with national standards; others will not fortify at all; and the rest will conduct sub-standard fortification, resulting in flour with some nutrient content but not meeting national standards.

To model flour nutrient levels under the counterfactual that Sanku was not active, we first estimate the share of Sanku-fortified flour that is (or for expansion countries, would be) produced by industrial-scale mills compared to small-scale mills.^[7] For the three active Sanku countries, we use estimates provided by Sanku, although these are necessarily approximate as Sanku does not have precise information on the milling capacity of its premix customers. For expansion countries, we take a simple average of the values for active countries, on the assumption that Sanku would work with a relatively similar mix of industrial-scale and small-scale mills. We also provide an option for users to adjust this parameter as more information becomes available.

We model the flour being produced by small-scale mills as having zero nutrient content under the counterfactual that Sanku was not active. While maize and wheat both contain naturally occurring nutrients, John Kangarua emphasized that the milling process removes most of these nutrients, resulting in negligible nutrient levels in the absence of fortification.

Estimating counterfactual nutrient intake for industrial-scale mills is more challenging. Recent progress in fortification means that in Kenya, for example, a recent report found relatively high compliance with fortification standards: 86% of wheat samples were adequately fortified, while 62% of maize samples were adequately fortified. We use these figures as our baseline for the Kenyan industrial-mill counterfactual, but the observed compliance rate in the QuImpact report may already incorporate Sanku’s effect on other mills. The compliance figure in the genuine “absence of Sanku” counterfactual would plausibly be substantially lower, reflecting issues with non-Sanku premix supply chains (in interviews, Sanku mentioned issues such as counterfeit premix, stockouts, formulation that targets minimum rather than midpoint nutrient levels, and inconsistent dosing at the mill). We therefore place a 20% haircut on estimates for Kenya: assuming that the level of fortification for industrial mills in the absence of Sanku would be 66% for wheat and 42% for maize in Kenya. This assumption may understate Sanku’s marginal impact at large mills, so we conduct sensitivity testing to test it.

John Kangarua (Sanku) told us that before their intervention in Ethiopia, fortification was essentially nonexistent, and that this is still the case for the seven expansion countries. We therefore choose a value of 5% compliant fortification among industrial mills for these countries.

The picture in Tanzania is complex. In our conversation with John Kangarua, he gave the example of a dominant large-scale miller that has 90–95% market share, and according to Sanku staff, would likely continue with fortification in the absence of Sanku, albeit with potentially lower quality premix (Sanku estimates ~50% compliance in this scenario). This suggests that a large majority of premix in Tanzania would be counterfactually fortified to ~45% in the absence of Sanku, so we use this figure in our model (we caveat the high uncertainty of this figure).

We then calculate a residual between Sanku-fortified flour and the counterfactual flour in the absence of Sanku, on a per-country basis for the active Sanku countries. For expansion countries, by default we assume a counterfactual compliance of 5%, since little to no fortification is currently undertaken. Our model allows changes to “User Input” so they can see a simple average of the values for active countries, if this is preferable. Sanku’s marginal nutrient levels are then calculated as the difference between nutrient levels in Sanku-fortified flour and counterfactual flour.

An assumption here is that Sanku does not have an indirect effect in causing non-Sanku partner mills to begin fortification. Sanku’s work in training millers and educating them on the benefits of fortification may have the effect of prompting fortification in general, for example, through network dissemination of this information and training. It is also possible that in some cases these millers will continue to fortify as a result of Sanku’s work, but not buy from Sanku’s premix explicitly: our model ignores the benefits from this fortification. In our conversation with John Kangarua (Sanku), he told us that fortification in Tanzania was minimal before Sanku’s entry, and that their influence has led other mills that do not partner with Sanku to fortify as well. Central to this is their government relations work: by supporting government enforcement and therefore driving mills to fortify in general, this work is likely to be a significant part of Sanku’s overall impact. Quantifying this indirect impact is highly challenging, and we therefore do not account for it in our model, but consider that this may downwardly bias our estimates for Sanku’s health impacts.

Figure 3: Approach for estimating counterfactual nutrient levels of Sanku-fortified flour.

Ferrous Fumarate vs. NaFeEDTA

Currently, Sanku’s flour premix is formulated using ferrous fumarate for the iron component. However, Sanku has signaled that they plan to switch to using sodium iron (iii) ethylenediaminetetraacetate (NaFeEDTA) instead, which is significantly more bioavailable, but typically comes at a higher cost. When faced with the choice of switching to NaFeEDTA, a fortifier could choose to either:

1. Reduce the volume of premix delivered to mills, holding absorbed iron per consumer constant (thus not changing health benefit).
2. Hold the volume of premix constant and use the higher bioavailability to deliver more absorbed iron per consumer. This would increase the health benefit, but at a higher cost.

When asked which of the two options they will realistically adopt when switching to NaFeEDTA, Sanku staff told us they would select the second option (hold the volume of premix constant). Sanku’s delivery model is built around standardized dosing across a highly diverse network of mills, so adjusting premix dosage would introduce significant complexity, including the need for widespread recalibration of dosifiers, retraining of field teams, and increased risk of underdosing or non-compliance at the mill level. Maintaining a consistent dosing approach ensures stability within existing operations, aligns with regulatory nutrient ranges, and minimizes disruption across the system.

To estimate the effect of switching to NaFeEDTA on health impacts, we modify the dose-scaled relative risks (discussed in the next section) to account for its higher bioavailability. The literature suggests that NaFeEDTA is absorbed 2.3 times as well as ferrous fumarate in wheat,^[8] so the discounted relative risk reductions for our iron impacts are 2.3 times greater under the scenario of NaFeEDTA (at the same concentration) than ferrous fumarate. The cost implications of switching to NaFeEDTA in our model are described in the costs section.

Our model provides the user the ability to model the effect of switching from ferrous fumarate to NaFeEDTA via a switch in the “User Input” tab. Our primary results use the status quo of ferrous fumarate. Sanku clarified that the planned switch to NaFeEDTA is targeted at small maize mills, rather than applying across their whole network, for two reasons:

1. The nutritional case is stronger for maize, with maize presenting a higher phytate constraint than wheat. The incremental bioavailability gain from NaFeEDTA is therefore substantially larger in maize flour than in wheat flour.
2. There are practical barriers in large wheat mills, which currently operate at around 200 ppm, whereas NaFeEDTA would likely require dosing to be increased to approximately 250–300 ppm alongside changes to existing dosing equipment and processes. Sanku expects that large millers would reject these changes. Our modeled switch to NaFeEDTA applies the higher bioavailability across all Sanku-fortified flour, and therefore represents an upper bound on the impact of the transition rather than Sanku’s realistic implementation plan. We discuss the implications of this in the sensitivity testing section.

Dose-Scaled Relative Risks From Additional Nutrient Intake

Our approach to estimating the reduction in the risk of key health outcomes caused by the additional nutrient intake from Sanku’s fortification program is summarized in Figure 4. We draw on estimates from the literature for the impacts of fortification and supplementation, selecting rigorous meta-analyses of randomized controlled trials, preferring Cochrane reviews where available due to their high quality.

We searched for estimates that mirror Sanku’s intervention as closely as possible. Where available, we use meta-analytic estimates specifically for the impact of fortification with our key micronutrients accompanied by other vitamins. Where these estimates are not available (for Vitamin A, folic acid, and the impact of iron on IQ and physical capacity), we use meta-analyses of supplementation studies instead. We collect information on the meta-analytic estimates here.^[9] The relative risks we choose are shown in Table 3.

Table 3: Relative risks used in model.

Key information that we collected was the dosage and compound used in the studies underlying each estimate.^[10] Because Sanku-fortified flour only represents a fraction of the total flour consumed within each country,^[11] and because we are only accounting for the additional nutrient intake in Sanku after subtracting counterfactual nutrient levels, the values for additional intake per person in the country are very small. Dosages per person per day for participants in fortification studies are much larger, and dosages in supplementation studies even more so. To adjust the relative risks in these studies to our modeling, and to be able to combine these relative risks with national health burden estimates in the next step, we must adjust for the much lower nutrient intake delivered by Sanku-fortified flour at the population level.

To do so, we multiply the relative risk reduction in the meta-analyses by the ratio of the dose delivered by Sanku to the dose delivered in the underlying studies. We adjust the dose in the studies by the relative bioavailability of the compound used compared to the compound used by Sanku^[12] (see related discussion of ferrous fumarate vs EDTA above) to make sure that the ratio captures the relative uptake of each nutrient by the body.

This approach makes the assumption that benefits of micronutrients scale linearly with intake. The direction of bias introduced by our assumption of a linear dose-response is uncertain. As pointed out by Demewoz Woldegebreal, there is a lack of evidence on the dose-response relationship of micronutrient intake. Some studies suggest that fractional absorption of iron decreases with greater doses (Moretti et al. 2015), suggesting a concave dose-response relationship where impacts decrease as individuals reach sufficiency. However, this evidence is from high-dose supplementation trials, which may not translate to the much smaller doses provided by Sanku.

If a concave (or S-shaped) curve represents the dose-response relationship well, the direction of bias resulting from our linear dose-response assumption would depend on the country’s deficiency levels compared to the study populations underlying the relative risk estimate. If our populations have higher deficiency than the populations in the underlying studies, then Sanku’s impact will be at the lower, steeper section of the dose-response curve, and their impact will therefore be underestimated by the assumption of linearity. In general, our relative risk estimates are for similar settings (e.g., for low- and middle-income countries [LMICs]), but there may still be significant variation in deficiency. Given the uncertainty both regarding the shape of the dose-response curve and this deficiency alignment, we do not think there is sufficient evidence to move away from the parsimonious assumption of linearity.

Figure 4: Approach for estimating national relative risk reductions from Sanku fortification.

National Burden of Micronutrient Deficiency

To estimate the burden of our metrics of interest attributable to deficiency in the four key nutrients for each country, we draw on the Global Burden of Disease (GBD) 2023 study. This includes the most comprehensive effort to quantify the burden attributable to risk factors, including micronutrient deficiencies, and the research underlying it is likely to be considerably more thorough than our rapid review of the micronutrient literature. The GBD database includes estimates for a number of health outcomes attributable to iron, zinc, and Vitamin A deficiency across all countries and relevant age groups, although not for folic acid deficiency.

We retrieve estimates of the following metrics attributable to iron, zinc, and Vitamin A deficiency directly from the GBD study: deaths, DALYs, maternal deaths,^[13] and anemia cases. We then multiply these by the dose-scaled relative risk estimates for the impact of Sanku’s fortification on deficiency in these nutrients to calculate the estimated reduction in attributable health burden through Sanku fortification.

The GBD does not provide an estimate for neural tube defects attributable to folic acid deficiency. We therefore retrieve the number of neural tube defect cases from the GBD, as well as the deaths and DALYs associated with these cases; we then multiply these by the dose-scaled relative risk estimate of Sanku’s fortification on the risks of neural tube defects directly, to estimate the burden reduction through Sanku fortification.^[14]

This assumes that rates of deficiency for those who consume Sanku flour mirror the national population: if Sanku recipients have a lower level of deficiency than the national population, for example, then the additional nutrient intake will have a smaller impact on the health outcomes linked to deficiency.

An additional assumption is that Sanku fortification does not have harmful effects. While there is established evidence for harmful effects from excessive intake of certain nutrients that are present in Sanku fortification, we expect that given generally low nutrient intake levels in East Africa, this impact is likely to be low to moderate.

A potentially more significant issue is that some evidence suggests that increased iron intake can raise malaria incidence in high-transmission settings. The evidence on this is mixed: a recent Cochrane review (Itzkovich et al. 2026) found that iron supplementation results in little to no effect on clinical malaria even in endemic settings. Given this mixed evidence and the relatively low prevalence of malaria in the active Sanku countries, we do not include this impact in our model, although we do include it as a qualitative factor for Sanku’s expansion into new countries in our‘Framework for Expansion section. However, this may slightly upwardly bias our estimates for the health benefits from Sanku’s fortification programs.

Alternative Estimation of Averted Anemia

There is some debate as to how to best model averted anemia cases from iron deficiency. In our main methodology described above, we calculate cases averted as the anemia attributable to iron and Vitamin A deficiencies, multiplied by their respective dose-scaled iron relative risks (RR) using GBD data.

However, discussions with GiveWell highlighted that the GBD “dietary iron deficiency” category is imputed primarily from US data, and offers limited country-specific validity. We therefore calculate anemia cases averted using a secondary methodology, which follows GiveWell’s latest iron-fortification methodology to provide a sensitivity test of our main method. This more “direct” method applies the Cochrane meta-analytic relative risk for iron fortification on anemia prevalence (≈0.60, corresponding to a 40% reduction based on discussions with GiveWell) directly to country anemia prevalence, rather than passing through GBD’s iron-deficiency-attributable-anemia category.

The direct approach gives an anemia-cases-averted estimate that is broadly consistent with the indirect method but does not depend on iron-deficiency attribution. The direct approach theoretically produces a higher estimate for anemia cases averted relative to our method, because it applies the RR to all anemia, not just iron-deficiency-attributable anemia, and does not separately credit the share of anemia driven by inflammation, malaria, or other non-iron causes, leading to a potentially meaningful upward bias in high inflammation settings (Hurrell, 2022). We use our main method as the bottom-line measure of anemia (using the iron-deficiency intermediate for the cognitive and maternal-mortality channels, where the relevant outcome is mediated by iron status rather than anemia per se), but still present results from the secondary direct method for comparison.

Economic Impacts

We estimate the economic benefits of Sanku’s fortification program through three channels, all linked to the iron component of Sanku’s premix:

1. Lifetime wage gains from improved cognitive development in children;
2. Annual wage gains in working-age adults from improved adult cognitive function;
3. Annual wage gains from productivity benefits linked to reduced physical work limitation in working-age adults.

We follow GiveWell’s iron-fortification cost-effectiveness analysis (CEA) as our methodological anchor for the cognitive benefits channels, but our formulation differs from their published methodology in three ways. First, we translate the supplementation-trial IQ effect to a fortification-dose effect using the ratio of Sanku-delivered iron intake to the supplementation-trial dose, rather than GiveWell’s published (1−RR_fortification)/(1−RR_supplementation) scalar. We adopt this linear-dose-response framing because it applies the same dose-scaling logic we use throughout the health-impacts model, avoids the need for a beneficiary-count denominator that is hard to define cleanly given Sanku’s overlapping maize and wheat consumer base, and exposes the dose-response assumption in a single transparent parameter. We acknowledge this implies a stronger linearity assumption than GiveWell’s approach.

Second, we monetize the wage benefit by multiplying through by each country’s per-capita wage component (GDP per capita × labor share of GDP), rather than using GiveWell’s “value per unit ln(consumption)” moral-weights framework. This produces a dollar-denominated benefit figure that maps directly to our cost-side calculations and to standard external benchmarks (such as GDP per capita per DALY averted).

Third, we add a productivity-benefits channel for working-age adults (Channel 3), following the methodology outlined in the Micronutrient Initiative’s Economic Consequences of Iron Deficiency (p. 20). GiveWell does not explicitly include this channel in its iron-fortification CEA.

We define economic benefits narrowly as wage gains accruing to program beneficiaries via cognitive and physical-productivity channels mediated by iron status. Unlike for health impacts, we do not estimate economic benefits from the other nutrients in Sanku’s premix (zinc, Vitamin A, folic acid, Vitamin B12), nor from non-wage economic channels such as averted healthcare costs, caregiver-time savings from reduced child morbidity, downstream productivity gains from averted neural tube defects, or wage gains from averted vitamin-A-related under-five mortality. We restrict to the iron channel because adding the other nutrient-specific economic channels would require strong dose-response evidence that we do not currently believe is available. The consequence is that our modeled economic benefit is a downward-biased estimate of total economic benefit accruing to beneficiaries.

Channel 1: Cognitive Benefits for Children

For each beneficiary child with iron deficiency, we estimate the IQ point gain from Sanku fortification by combining the meta-analytic effect of iron supplementation on cognition in deficient children (4.55 IQ points per fully supplemented deficient child, drawn from Low et al. (2013) with internal and external validity adjustments of 0.7 each), and the ratio of Sanku-delivered iron intake to the supplementation-trial dose.^[15] This linear scaling reflects that Sanku fortification delivers a fraction of the supplementation dose, and assumes the cognitive effect scales proportionally.

The IQ gain is then translated into lifetime wage benefits via four further adjustments: a 0.67% wage gain per IQ point (drawn from observational labor-market studies, GiveWell’s central estimate), a 33% retention factor reflecting the share of childhood cognitive gains preserved into adulthood, a resource-sharing multiplier of 2 capturing within-household consumption smoothing, and an amortization adjustment of 0.1. This amortization adjustment, which approximates that one year of program exposure produces roughly one-tenth of the trial-measured cognitive improvement, reflects GiveWell’s framing that the IQ effect size from Low et al. (2013) was measured in supplementation trials of finite duration, whose cognitive impact is presumed to persist after the trial ends. Without this adjustment, attributing the full trial-measured IQ effect to each year of Sanku fortification would overstate the per-year benefit, since the trial-equivalent effect size is delivered cumulatively over a period of exposure rather than immediately within a single year of intervention.

We discount the resulting wage stream at 4% per year, with an 11-year lag from intervention to first wage benefit and a 30-year duration of benefit accrual. We follow GiveWell’s assumption of log-utility of consumption, meaning a 1% consumption gain has roughly constant utility regardless of base income.

The result, expressed as the present value (PV) of lifetime benefits per child cohort, is multiplied by the country’s per-capita wage component (GDP per capita x labor share of GDP) and total population to obtain a total dollar value for the children’s cognitive benefit channel. We also report the total and per-capita averted loss in IQ points for children.

Channel 2: Cognitive Benefits for Adults

For working-age adults, we apply a parallel structure but with adult-specific parameters: a smaller IQ effect size (0.35 standard deviations (SD), equivalent to 5.25 IQ points after the SD-to-points conversion), different validity adjustments per GiveWell’s method (0.85 internal, 0.6 external), and the same wage-per-IQ-point and resource-sharing multipliers. We use a similar supplementation dose assumption.^[16] We assume adult cognitive benefits accrue contemporaneously with iron sufficiency (no lag, no retention adjustment, no PV calculation), since iron stores deplete in months when adults stop consuming fortified flour (i.e., there is no permanent capacity gain). Therefore, the dollar benefit is the year-1 wage gain across the working-age adult population, computed as the per-capita ln-consumption gain multiplied by the country’s wage component (GDP x labor share of GDP).

We also report the total and per-capita averted loss in IQ points for adults.

Channel 3: Productivity Benefits in Adults

Following the Micronutrient Initiative’s method, we assume anemic adults experience productivity losses estimated at approximately 5% in blue-collar work and 12% additional in heavy manual labor (which is a subset of blue-collar work). We multiply these productivity losses by the country’s wage component (GDP x labor share of GDP), employment composition (blue-collar share, heavy-manual share), country anemia prevalence, and the dose-ratio scaler that we apply throughout the model.^[17] Note that the Micronutrient Initiative report was published in 1998, but we could not find a credible, more recent source for estimating productivity given our research problem.

The result of the calculation is an annual wage gain (i.e., loss averted) via reduced anemia-driven productivity loss, which the Micronutrient Initiative notes may “favor rich over poor and men over women” (p. 215). We do not estimate physical-productivity benefits in non-blue-collar work, owing to uncertainties around how baseline productivity suffers from anemia across labor markets, which is a downward bias on our adult productivity estimate.

Combining the Channels

Our headline economic-benefit figure sums the children’s cognitive PV (annualized over the 30-year benefit period for comparability with the adult flows), the adult cognitive year-1 wage gain, and the adult productivity year-1 wage gain. We report all three separately, plus a combined total. The magnitudes of the channels for children and adults are in theory not directly comparable in raw dollar terms, because children’s benefit is expressed as a PV over a 30-year future stream and adults’ benefit is expressed as a single-year flow. We nonetheless choose to report a combined figure because GiveWell treats the cost-effectiveness ratio as an estimate of how many dollars of total future-value benefit are achieved per dollar spent today; however, this combined total should be treated with some caution.

A consequence of restricting our economic-benefit calculation to iron-mediated channels is that Ethiopia has zero modeled economic benefit through any of our three channels. The Ethiopian government’s mandatory fortification standard for wheat and maize does not currently require iron, and Sanku’s premix in Ethiopia accordingly contains no iron. This means our headline economic-impact estimate for Ethiopia is structurally zero, with all of Ethiopia’s quantified benefit accruing on the health-impact side of the model.

Costs

We model Sanku’s monthly operating costs at the country level, with cost estimates organized into four blocks:

1. Direct programmatic operating expenditure (OpEx), captured at the country level and net of intercompany double-counting.
2. Government relations and advocacy expenditure, which we present separately so it can be included or excluded depending on the question being asked
3. Capital expenditure (CapEx), captured through the depreciation and amortization lines on the income statement
4. Revenue from premix, dosifier, and bag sales, which we treat as defraying programmatic cost

Our reference period is the second half of fiscal year 2026 (October 2025–March 2026), the period over which Sanku’s three active countries have operated in their current configuration with consistent country-level financial reporting. We use the results to feed Cost-effectiveness metrics.

Active Countries

Across the three active countries, total operational expenditure (OpEx), gross of revenue and before any government relations or capital expenditure adjustments, was ~$950,000 per month. Of this, Tanzania accounted for ~$612,000 per month (the largest share, reflecting both the size of the program and the bag-printing operation that is unique to Tanzania), Kenya ~$172,000, and Ethiopia ~$166,000. Ethiopia is unusual among the three in being close to operating break-even at country level, because of higher premix-sale revenue relative to a smaller cost base. Factory capital expenditure is treated as sunk for Tanzania.

We make some adjustments to the OpEx values for active countries. For Tanzania, we exclude costs relating premix sold from the Tanzania blending plant to Kenya and Ethiopia, which avoids a double-count. We also allow the user to model the effect on OpEx of Sanku’s planned switch from ferrous fumarate to NaFeEDTA in the flour premix, using a switch in the USER INPUT tab that multiplies the premix cost by 40–60%. This is based on Sanku’s internal calculations for the net increase in switching, accounting for both carrier displacement and bioavailability.^[18] By default, the model assumes the use of ferrous fumarate (and thus no additional OpEx).

Government relations and advocacy expenditure across the three countries amounts to approximately $25,000 per month. We exclude this from our headline cost-effectiveness ratio on the grounds that Sanku does not consider it directly attributable to programmatic delivery. Government relations work plausibly improves enforcement of fortification standards among non-Sanku partner millers, so excluding it from our headline figures may upwardly bias our results. We facilitate the user’s ability to assess the effect of including government-relations costs by providing a switch in the USER INPUT tab.

Capital expenditure is captured through the depreciation and amortization lines on the income statement (~$63,000 per month for Tanzania, ~$11,800 per month for Kenya, and ~$11,700 per month for Kenya), which represent the annualized cost of equipment that has already been deployed. By default, we exclude CapEx for the results presented in this report, but we provide a toggle in the USER INPUT tab for the user to include or exclude CapEx from the final gross and net cost calculation. We exclude the one-off cost of building the central premix factory in Tanzania, which Sanku told us cost approximately US$3.5 million, on the grounds that this expenditure is sunk: the marginal decision facing donors today is whether to continue funding ongoing operations rather than whether to fund the original factory build.

Revenue is reported as a single line item and used to convert gross costs to net-of-revenue costs for use in some cost-effectiveness metrics.

Expansion Countries

For the seven expansion countries, we forecast costs using a combination of fixed country-presence costs and flour-volume-variable costs. The fixed country-presence component dominates at a low operating scale: Sanku’s Kenya and Ethiopia operations each cost ~$170,000 per month, despite Ethiopia’s flour volume being ~75% higher, suggesting that the marginal cost per additional kilogram of fortified flour is small in the premix-only operating model.

To estimate gross costs in expansion countries, we take a weighted average cost per kilogram from the three active countries (equating to ~$0.01/kg) and apply it to each expansion country’s flour volume. To this, we add a fixed expansion cost of $12,500 (Sanku estimates $750,000 representing travel, office registration, and consultancy services, and we amortize this over 60 months). Flour-bag printing operations such as Tanzania’s add a ~50% premium to the per-kilogram gross cost. We default to a “no bag model” assumption for all expansion countries, although a switch in the USER INPUT tab allows the user to toggle individual countries to bag-model treatment.

Where required, a premix-factory capital expenditure can be added to any given country by the user, equating to ~$19,400 per month (US$3.5 million, amortized over 180 months). The amortization period is a judgment call about plant lifetime; shorter or longer assumptions move the per-beneficiary cost meaningfully. Sanku’s advisor Marvin Jaeger indicated that two or three of the seven expansion countries (most plausibly Mozambique, Uganda, and Zambia) would require their own factory.

Revenue for expansion countries is forecast by taking a flour-production-weighted average of revenue per kilogram of flour in active countries and multiplying it by the total forecasted flour production in each expansion country. This is used to convert gross costs to net-of-revenue costs, for use in some cost-effectiveness metrics.

Cost-Effectiveness

We report cost per life saved, cost per DALY averted, cost per maternal death averted, cost per neural tube defect averted, and cost per anemia case averted, both gross of revenue and net of revenue. We default to the gross figures as the headline (consistent with how GiveWell reports its analyses), with net figures available as a sensitivity check (One Acre Fund tends to estimate cost-effectiveness using net revenue figures). We compute these for each country using annual cost figures (the income-statement monthly figures multiplied by 12) divided by annual impact figures.

We calculate an additional measure of cost-effectiveness that may be relevant to potential funders, namely a multiple of the GiveWell benchmark. The benchmark represents the cost-effectiveness of improving lives by giving cash to people living in poverty in low-income countries to purchase things they need, and it provides a way to compare interventions across different sectors. As of May 2026, GiveWell’s bar for funding programs is 6x their benchmark.

For simplicity, given limited time to conduct a detailed GiveWell-style moral weights analysis, we opted to convert our health and economic impacts into mortality units of value (UoV) and economic UoVs (respectively). For mortality UoV, we avoid double-counting deaths by simply multiplying DALYs averted in our model by 2.33, as per GiveWell’s recommendation. For economic UoV, we multiply total wage gain by 0.001, since GiveWell assumes $1 of wage gain is worth approximately 0.0005 to 0.001 UoVs. To calculate the multiple of the benchmark, we sum the mortality and economic UoVs, divide by net annual revenue, and then divide by 0.0031 (per GiveWell).

Key Assumptions and Directions of Bias

We have detailed many of the assumptions our model makes in the description of our approach above, as well as the uncertainty around many parameters. In Table 4, we list what we deem to be the major assumptions in our approach, as well as our view on whether each assumption is likely to upwardly or downwardly bias our estimates for the effectiveness of Sanku’s flour fortification program.

Table 4: Summary of key model assumptions, and likely direction of bias for effectiveness estimate.

¹ Assumption upwardly (🔼) or downwardly (🔽) biases estimates of effectiveness.

Model Results for Active Countries

In this section, we provide model results for Sanku’s current programs in the three active countries. Health and economic impacts are reported annually unless stated otherwise.

Note that these results are for Sanku’s current premix formulation (i.e., using ferrous fumarate rather than NaFeEDTA). Projected impacts and cost-effectiveness after the planned switch to NaFeEDTA are presented in the sensitivity testing section.

Health Impacts in Active Countries

Sanku’s fortification program resulted in an estimated 17,419 disability-adjusted life-years (DALYs) averted annually in Tanzania, with an order of magnitude fewer DALYs averted in Kenya (1,314) and Ethiopia (1,040) (Table 5).

Iron was by far the greatest contributor to the number of DALYs averted, accounting for more than 90% of DALYs averted in Tanzania and Kenya. Our primary method for estimating anemia cases averted gave 495,348 cases in Tanzania and 38,673 in Kenya, with the alternative method giving 447,045 cases in Tanzania and 34,528 in Kenya.

The total health impacts were lowest in Kenya, probably owing to the fact that the milling sector is most developed in terms of fortification, which reduces the scope for Sanku to have a large counterfactual impact.

Table 5: Health impact metrics from model analysis, reported annually.

¹Includes maternal deaths averted.

Economic Impacts in Active Countries

Total annual economic benefit was approximately ~$10.4million for Tanzania and ~$842,000 for Kenya, comprising wage gains through three channels, all attributable to iron fortification (Table 6). Ethiopia did not have any reportable gains owing to zero iron in its premix.

Since Sanku’s daily delivered iron dose is a small fraction of the supplementation-trial doses that anchor the underlying IQ effect size, our linear dose-response assumption scales the per-child IQ gain to 0.006 IQ points per capita, and the per-adult IQ gain to 0.009 IQ points per capita in Tanzania (lower in Kenya). The cognitive children channel (Channel 1) is smaller than the adult cognitive channel (Channel 2), partly because of how its effects are adjusted in our methodology based on GiveWell: namely, a 33% adjustment for maintaining effects into adulthood, and a 10% amortization effect, which both drive estimates downwards.

In wage gain terms, the physical productivity channel (Channel 3) is about three times as large as the cognitive channel for adults, which is to be expected given productivity loss from anemia is a contemporaneous effect on labor market output, while adult cognition is a smaller second-order effect.

Table 6: Economic impact metrics from model analysis, reported annually.

Cost-Effectiveness in Active Countries

We report both gross and net-of-revenue cost-effectiveness here for transparency, but focus our analysis mainly on net figures (Table 7). Cost per life (net) varied by a factor of ~7.5 across the active countries, ranging from $57,524 for Ethiopia to $446,186 for Kenya. Cost per DALY varied by a factor of ~6.2 across the countries, ranging from $207 for Tanzania to $1,273 for Kenya. Ethiopia’s net figures are particularly strong, whereas it is weakest in gross terms, suggesting the revenue model for Ethiopia may be particularly sustainable.

Tanzania also had significantly more cost-effective results for anemia and wage benefits than Kenya, with a net benefit-to-cost ratio of 2.9 for Tanzania versus 0.5 for Kenya.

Table 7: Cost-effectiveness metrics from model analysis.

Comparison to GiveWell Benchmark

We convert our DALYs averted and estimated wage gains into mortality and economic units of value (UoV) using standard GiveWell methodology to ultimately quantify the multiple of cash transfers represented by each active country’s operations (Table 8).

Tanzania achieves the most effective multiple at 4.6x, which is commendably close to GiveWell’s benchmark of 6x. Operations in Ethiopia have a multiple of 1.2x, and in Kenya 0.8x. These results reflect both the maturity and scale of the Tanzanian operation, as well as its relatively low-compliance environment. In Kenya, where fortification is likely to be higher in the absence of Sanku, the counterfactual effect is notably weaker, which suggests that focusing on low-fortification settings for expansion will yield higher cost-effectiveness.

Table 8: Multiples of cash transfers for comparison to GiveWell benchmark.

Sensitivity Testing

Ferrous fumarate vs. NaFeEDTA

By default, our model assumes that Sanku’s premix contains ferrous fumarate for the iron component, as this is what Sanku currently uses in Tanzania and Kenya. However, Sanku plans to switch to NaFeEDTA, which has higher bioavailability but also higher costs. In this sensitivity analysis, we consider the effect that this switch will have on Sanku’s impact and cost-effectiveness.

We estimate that switching to NaFeEDTA could result in a significant increase in health and economic impacts in both countries, with the benefits from NaFeEDTA’s higher bioavailability substantially outweighing its increased cost (Table 9). The number of DALYs averted would increase by 124% in Tanzania and 120% in Kenya, to 39,022 in Tanzania and 2,888 in Kenya. Total wage gains would increase by 130%, to $28.8 million in Tanzania and $1.9 million in Kenya. The cost per DALY would decrease by 44% to $116 in Tanzania, and by 50% to $631 in Kenya. Importantly, switching to NaFeEDTA could lift Tanzania’s operational model by 79% to 8.2x cash transfers, which is above GiveWell’s benchmark of 6x.

Full results for the switch to NaFeEDTA for the active Sanku countries are provided in Appendix 3. We note that these estimates come with greater uncertainty than our primary results for two reasons:

1. The projected costs for the switch to NaFeEDTA are approximate projections rather than actual financials.
2. Our model applies the NaFeEDTA bioavailability uplift across all Sanku-fortified flour, whereas Sanku’s actual plan is to switch small maize mills only. Sanku has advised us that large wheat mills are not a practical target for the transition: the nutritional case is weaker in wheat given its lower phytate content, and switching would require increasing dosing from around 200 ppm to 250–300 ppm along with changes to dosing equipment, which Sanku expects large millers would reject. The figures reported here and in Appendix 3 should therefore be read as an upper bound on the effect of the transition. The realized gain will be smaller, and its size will depend on the share of Sanku’s flour volume produced in small maize mills. In Tanzania, where our model assumes that 0% of maize flour and 100% of wheat flour is produced in industrial-scale mills, the realized uplift would apply to the maize segment only.

Table 9: Impact metrics assuming the use of NaFeEDTA in premix. Value in brackets show proportional change in value relative to fumarate. Impacts are reported annually.

Cost Inclusions

By default, our model only considers OpEx costs in the denominator of cost-effectiveness calculations. We test the effect of including CapEx and government relations costs on the headline figures for cost-effectiveness (Table 10). Gross costs increase by 10–15%, while net costs increase by 13–47%, with corresponding increases in cost figures. Net benefit-to-cost ratios decrease by 20% for Tanzania and 11% for Kenya when CapEx and government relations are included.

Table 10: Cost-effectiveness metrics when including CapEx and government relations costs. Value in brackets show proportional change in value relative to OpEx costs only.

Industrial-Scale Fortification in the Absence of Sanku

In our model, we assume counterfactual fortification levels by industrial-scale mills (≥20 tonnes per day) in the absence of Sanku vary by country. As explained in Counterfactual Fortification, we have high uncertainty about the values we rely on for this, so we test the sensitivity of our model results to the fortification value by separately modeling 0% and 100% fortification counterfactuals (Table 11).

In Tanzania, and particularly in Kenya, where counterfactual fortification is known to be relatively high, lowering the fortification counterfactual to 0% has a large positive impact on health, impact, and cost-effectiveness metrics. Tanzania’s operations reach 6.9x cash transfers in cost-effectiveness under this scenario. Conversely, increasing the counterfactual to 100% fortification leads to vastly reduced impacts across all three countries, with cost per DALY increasing by 158% in Tanzania and 164% in Ethiopia.

These results emphasize the sensitivity of our model on the counterfactual fortification scenario; as fortification compliance gets increasingly enforced across East Africa, the uncertainty bounds on the counterfactual will likely decrease.

Table 11: Impact metrics assuming 0% and 100% fortification in industrial-scale mills in the absence of Sanku. Value in brackets show change in value relative to fumarate. Impacts are reported annually.

How to Move the Needle on Cost-Effectiveness

While our estimates for Sanku’s current cost-effectiveness are under GiveWell’s benchmark of 6x cash transfers, a few things are worth noting. First, there is large uncertainty over several key parameters, both from the assumptions in the Key Assumptions and Direction of Bias section and general uncertainty in the specific parameters we choose. Upper-bound estimates for Sanku’s impact are likely to be above the 6x bar, and further information may push our estimate of Sanku’s impact towards this higher bound.

In particular, we currently do not factor in any “indirect,” industry-level impact of Sanku’s work in prompting fortification among other millers. Sanku supplies premix, but its work also includes government relations and advocacy with regulators, technical support for national fortification standards setting, capacity-building and training for the broader milling industry, laboratory upskilling for quality-control infrastructure, and convening across the fortification ecosystem in each operating country.^[19] These activities plausibly affect compliance and fortification quality at mills that do not buy Sanku premix by improving the regulatory environment, raising industry-wide technical competence, and improving the quality and reliability of non-Sanku premix supply chains. Since cost-effectiveness analysis excludes any positive externalities at non-Sanku-served mills, there may be a meaningful downward bias in our estimates of Sanku’s impact at the industry level, particularly in countries where industrial-scale mills constitute a large share of the national flour supply.

Second, there are changes to Sanku’s programs that would raise their cost-effectiveness further. As noted in the sensitivity analysis section, switching from ferrous fumarate to NaFeEDTA is estimated to increase the multiple of cash transfers figure by ~80%, pushing their program for Tanzania above the 6x bar. This is a strong argument for switching to NaFeEDTA, even after factoring in its higher costs.

By contrast, Sanku’s program in Kenya is estimated to be significantly less cost-effective, owing mostly to a significant share of Sanku’s Kenya operations being with industrial-scale mills, where the counterfactual fortification level is high. We suggest that focusing on countries with low counterfactual fortification (our model assumes this to be true of all seven expansion countries, but further information may change this) is a very effective way to raise cost-effectiveness.

Framework for Expansion Countries

In our model, we estimate impact and cost-effectiveness estimates for theoretical Sanku programs in seven potential expansion countries (Burundi, Malawi, Mozambique, Rwanda, Uganda, Zambia, and Zimbabwe). However, these estimates inherently come with substantially greater uncertainty than estimates for the three Sanku countries, as the scale and characteristics of hypothetical programs in these countries are highly challenging to determine ex-ante.

In this section, we therefore first discuss factors likely to influence impacts and cost-effectiveness that are included in our model with substantial uncertainty, or are not included at all; we also describe some general takeaways from our model results. Where available, we present quantitative and qualitative data on these factors. This complements our description of the Key Uncertainties and Assumptions in our model. The purpose is principally to highlight information that will be useful for Sanku to collect and consider as they decide where to expand, and only secondarily to provide provisional insights about the most suitable country, based on present information.

We begin by discussing factors relevant to expansion decisions—including key uncertainties in our model and contextual information not captured by our qualitative estimates—before describing the implications of the model results.

Forecasting Flour Consumption in Expansion Countries

Volume estimates for premix sold (and therefore consumption of Sanku-fortified flour) influence not just impact but cost-effectiveness estimates, since greater scale results in fixed costs becoming relatively smaller, improving cost-effectiveness. Because consumption is likely to scale over time, any static cost-effectiveness estimate carries inherent limitations: our estimates are designed to give some sense of the cost-effectiveness of Sanku’s program once established in a country.

Our default model input (‘population-weighting’) only considers population when estimating the likely scale of Sanku’s program in the respective country. We also provide a user-input option to allow for control over this input by estimating impacts and cost-effectiveness under the scenario that Sanku reaches 1, 5, or 10 million kg/month of both maize and wheat flour for each country. Our model also attempts to account for regulatory and enforcement differences between countries using data from the GFDx (described in the Health Impacts section).

In our interview with Marvin Jaeger (Sanku), he pointed to other factors influencing Sanku’s ability to scale up within expansion countries. These include: openness of the milling sector to foreign vendors; transportation infrastructure, including access to ports and road infrastructure; and geographic and cultural proximity to Sanku’s base in Tanzania. Table 12 summarizes information on these factors for each country, drawing on information provided by Claude as well as our interview with Marvin Jaeger.

This information highlights Mozambique and Rwanda as countries with the opportunity to scale up quickly. This suggests that consumption in these countries could be quite large in the medium term, especially through building relationships with large millers. On the other hand, as discussed in the Counterfactual Fortification section, large mills are also more likely to be already practicing fortification, even if to a poor standard. This means that while consumption may be higher, the additional nutrient intake resulting from this consumption may be lower (per kilo) than small mills where counterfactual fortification is likely to be negligible.

Malaria Prevalence

Malaria-driven inflammation is known to attenuate iron absorption and could materially reduce the realized iron benefit in high inflammation settings such as much of sub-Saharan Africa. Additionally, there is some evidence that the risk of malaria and other infectious disease can be increased due to iron intake (Gera & Sachdev, 2002).

In our primary model results, we do not include either of these impacts, primarily due to the mixed evidence on each. In Table 12, we show malaria prevalence for the seven expansion countries. Of the seven countries, Mozambique and Uganda show significantly higher malaria prevalence. We consider this a factor to be cautious about when considering expanding into these countries.

Cost Forecasting for Expansion Countries

As highlighted in the Costs section, costs are dominated by fixed country-presence components. This arguably suggests that Sanku should focus on expanding within countries as much as possible before considering expansion markets. Another influential factor is whether new factories are required for premix production. Sanku’s advisor Marvin Jaeger indicated that Mozambique, Uganda, and Zambia were most likely to require new factories, which would bring with them substantially higher costs in the short term.

A significant uncertainty in our cost forecasts for expansion countries is that they depend on parameters estimated from only three active countries. With more observations across countries, we would have more data to reliably separate fixed and variable components. Revenue dynamics in particular likely differ from one country to the next, based on purchasing power and competitor pricing, so it is difficult to predict in which direction our naive calculation (based on revenue per kilogram of flour) may bias revenue figures in any given country. Further research into these factors in the proposed expansion countries would improve Sanku’s understanding of net costs for a hypothetical program there.

Default Model Parameters for Expansion Countries

To apply our model to the seven expansion countries, we make the following assumptions:

• Ferrous fumarate is used as the iron compound in the premix
• The counterfactual fortification level for industrial-scale mills without Sanku is 5%
• The default share of Sanku-fortified flour from large mills is equal to the average across the three active Sanku countries
• Flour is forecast using the population-weighted method
• Only OpEx costs are included (CapEx and government relations costs are excluded)
• No country adopts the bag model
• Mozambique, Uganda, and Zambia require a new factory (Marvin Jaeger interview)

Model Results for Expansion Countries

Key characteristics of expansion countries are shown in Table 12, and selected modeling results are shown in Table 13. These are provisional results that depend on the (sometimes uncertain) parameters described above. We note that these results assume the use of ferrous fumarate, and that the switch to NaFeEDTA would increase estimated impacts and cost-effectiveness.

Our results show that Mozambique and Uganda have the potential for the largest numbers of DALYs averted (5,085 and 4,543, respectively) and lives saved (15 and 14, respectively). The largest number of anemia cases averted would likely be in Uganda (116,817), Zambia (98,925), and Mozambique (98,827). All countries would likely achieve 0.002–0.004 IQ point losses averted per capita in children, and 0.002–0.005 IQ points in adults. The largest total wage gains would be achieved in Uganda and Zimbabwe (~$1.8 million and ~$1.7 million, respectively).

In terms of cost-effectiveness, expansion countries perform well. Mozambique and Burundi are estimated to have the lowest cost per DALY averted ($188 and $189, respectively), with Zambia ($205) and Zimbabwe ($223) also showing high cost-effectiveness. In terms of economic effects, Zimbabwe has by far the highest benefit-to-cost ratio from wage gains (3.9). Estimates for multiples of cash transfers range from 2.6 for Rwanda to 4.6 for Zimbabwe; a key difference with active countries (except for Ethiopia) is that the assumed level of counterfactual fortification is very low, so the additional intake from Sanku’s fortification is greater.

Inspecting our model, the key drivers behind the differences between countries seem to be:

• National standards for iron fortification: Countries like Uganda and Rwanda have low standards for fortification, meaning that iron content in Sanku flour would also be low, reducing the impact of iron, which is the primary driver of averted burden in our estimates.
• Estimates for DALYs attributable to micronutrient deficiency: Mozambique and Uganda have much higher DALY burdens attributable to deficiency (especially iron deficiency) than Rwanda, for example, meaning there is more potential benefit from additional nutrient intake.

While these results point to Mozambique and Zimbabwe as the most promising countries for additional Sanku programs, they should be considered alongside other information provided in Table 12. For example, Mozambique has the highest malaria prevalence of all seven countries (which may reduce the benefits from a Sanku fortification program, as discussed above), while Zimbabwe has the weakest connection to Sanku’s base in Tanzania.

Table 12: Expansion country characteristics.

¹ Source: Our World in Data (share of population with malaria in 2023).

² Sanku’s operational base.

³ Based on an interview with Marvin Jaeger.

Table 13: Impact and cost-effectiveness metrics from model analysis for expansion countries, based on default parameters. Impacts are reported annually.

What We Would Do With More Time

Given more time, we would seek to undertake the following work:

• We think that the most important cause of downward bias is that we do not include indirect effects of Sanku’s work in encouraging fortification among non-partners (discussed in Counterfactual Fortification section). Our conversations with John Kangarua and Felix Brooks-church suggest that Sanku’s entry into a country can prompt fortification generally, but these impacts are extremely difficult to model. Nonetheless, with more time we might explore ways to capture these impacts.
• There is a different approach taken in some literature to estimating the health impacts of micronutrient interventions. Instead of drawing from estimates from the literature for the impact of nutrient fortification or supplementation on deficiency levels, this involves modeling how additional micronutrient intake moves the beneficiary population’s overall intake to their Estimated Average Requirements (EAR): the intake required on average to not be deficient, which varies by population. This may require assumptions around current intake levels and how bioavailability varies between populations, as well as modeling how intake for an intervention (in our case, the consumption of Sanku-fortified flour) is distributed across a population. If we had more time, we might explore how estimates from this approach compare to those using our approach.
• Our method for adjusting relative risks from the literature for nutrient intake did not take into account how nutrient needs of the study population may be different to our own. For example, if more studies underlying estimates were for adults, then it is likely that applying these relative risk estimates to the children reached by Sanku would underestimate the benefits from their consumption of Sanku-fortified flour, as children require less nutrient intake to mitigate deficiency. With more time, we would explore the study populations underlying the relative risk estimates, and potentially make adjustments to the relative risks before applying them to our population. In general, our search for the most appropriate relative risks was also not especially thorough: with more time we would look for a wider range of estimates and spend more time considering which is most relevant to our model.
• We did not have time to run a full range of sensitivity tests. Key uncertainties that we would model explicitly with more time include the uncertainty in the relative risk estimates from the literature; alternative estimates for the burden from micronutrient deficiency; the impact of changing assumptions around the counterfactual; and varying forecasted costs, which come with high uncertainty. More sophisticated sensitivity testing would allow for a plausible range of confidence intervals to be established.

Appendices

Appendix 1: Rapid Evidence Review of Micronutrient Health Impacts

To estimate health impacts from Sanku’s fortification program, we began with a rapid review of the literature on health impacts from the nine micronutrients in Sanku’s premix. This review identified the impacts with the strongest evidence for substantial effects in the East African context, and informed our selection of focus nutrients and outcomes.

Drawing on this review, we concluded that the most rigorous and efficient approach to modeling the majority of health impacts was to build on the Global Burden of Disease (GBD) study’s country-level estimates of burden attributable to micronutrient deficiency. GBD’s estimates draw on an extensive evidence base and sophisticated modeling that goes well beyond what would be replicable within the scope of this engagement, and using them avoids reinventing the wheel where high-quality estimates already exist.

However, for three health impacts not covered by GBD—the impact of folic acid on neural tube defects, the impact of iron on maternal mortality, and the impact of iron on cognitive development—we drew directly on our own evidence review to develop estimates. We did so because our evidence review indicated that there is considerable evidence for them having substantial effects.

Selecting Focus Impacts

Sanku’s maize and wheat flour premix each contain nine nutrients. These are listed, along with their concentration in the maize and wheat flour premixes, in Appendix 2. Given this range of nutrients, each of which has been linked in the scientific literature to several health outcomes, the number of plausible health impacts from being a Sanku beneficiary is likely greater than 50. Due to the scope of this engagement, we followed a prioritization process to identify the impacts with the strongest evidence for substantial effects.

We used Claude to support an initial literature mapping of the impacts from the nine micronutrients. We collected information on a longlist of 30 impacts, including: the likely causal mechanism for the impact, key academic references, characteristics of the underlying studies, and information relevant to translating effect sizes in the literature to the case of Sanku’s wheat and maize fortification in East Africa.

We reviewed this evidence to narrow down the health impacts for our analysis. Our decision was holistic and focused on two factors. First, whether a systematic review (focusing on Cochrane reviews, given their high quality) demonstrated clear evidence for impacts from strong underlying studies. Second, whether this impact was likely to be of substantial size in our context; this depended not just on effect sizes in the literature, but also whether the particular health outcome had a high burden in the context of East Africa, and whether the doses provided through Sanku’s fortification were likely to deliver a substantial impact.

We also excluded impacts where we considered the effect to be mostly captured by another impact, as including both impacts would risk double-counting benefits. For instance, we excluded the impact of iron on iron deficiency, as we concluded that the impact of iron on anemia would capture the majority of the benefits of reducing iron deficiency. There were several cases where more than one nutrient was plausibly linked to the same outcome. If we decided to include the primary nutrient for this outcome in our shortlist, then we also preliminarily included impacts from other nutrients on that outcome.

Through this process, we found that the impacts with the strongest evidence for large effects in our context all came from four nutrients: iron, zinc, folic acid, and Vitamin A. These were the nutrients that we ultimately decided to include in our model, primarily using estimates of their impact from the Global Burden of Disease (GBD) study. However, we also drew on our evidence review directly for three impacts (the impact of folic acid on neural tube defects, the impact of iron on maternal mortality, and the impact of iron on cognitive development), which are not available directly in the GBD.

Appendix 2: Nutrient Content of Sanku Fortification Premix

Table A1: Nutrient content of Sanku fortification premix.

Appendix 3: Modeled Impacts After Switch to NaFeEDTA

Note: the figures in this appendix apply the NaFeEDTA bioavailability uplift across all Sanku-fortified flour. Sanku’s actual implementation plan targets small maize mills, as large wheat mills face dosing and equipment barriers, making the transition impractical (Sensitivity Testing). These results should be interpreted as an upper bound on the effect of the switch.

Table A2: Health impact metrics from model analysis after switching to NaFeEDTA, reported annually.

¹ Includes maternal deaths averted

Table A3: Economic impact metrics from model analysis after switching to NaFeEDTA, reported annually.

Table A4: Cost-effectiveness metrics from model analysis after switching to NaFeEDTA.

1. While we originally defined this metric as “number of children with improved cognitive development,” we think this definition requires likely arbitrary decisions about the threshold of “improved cognitive development,” and have therefore shifted to this definition. ↑
2. Sanku has only recently started to sell premix for maize in Ethiopia, so for that country we use only the values for February and March. ↑
3. These figures were for Tanzania. With more time, we would find estimates that are applicable across East Africa, although we think that age-group shares will not vary drastically across countries. ↑
4. For instance, Vitamin B12 has been linked to risk reductions for anemia and neural tube defects. However, much of the literature on Vitamin B12 is of interventions involving supplementation with B12 accompanied by other nutrients like iron or folic acid supplementation. Disentangling the impacts of these nutrients to identify a B12-specific estimate is highly challenging. ↑
5. In our model, iron accounts for around 90% of the DALYs among our four key nutrients. This suggests that the impact of nutrients is highly concentrated in the most critical nutrients. ↑
6. If the GFDx has multiple standards for a given country, we take the simple average for these standards. ↑
7. Where possible, this is based on the threshold in the Global Fortification Data Exchange (GFDx) of a >=20 MT/day grain processing rated capacity. As we explain, however, in practice this level of precision is not possible so our parameter here is only approximate. ↑
8. NaFeEDTA is absorbed 2.3 times as well as ferrous sulphate in wheat (Bothwell and Macphail 2004). Literature suggests that bioavailability for ferrous sulphate is similar to ferrous fumarate. ↑
9. Note that the choice of relative risk estimates is highly influential to our final result. For instance, an estimate used in GiveWell’s former modeling for the impact of iron fortification on iron deficiency assumes a significantly stronger impact than the parameter we use (relative risk estimate of 0.48 vs. 0.72). Our choice is based on our estimate from a more recent Cochrane review for the impact of iron fortification accompanied by other micronutrients, which we consider to better match our setting. ↑
10. We take the simple mean of bioavailability-adjusted doses from the underlying studies where this information is reported. Ideally, we would take a weighted average of all dosages (including those that are not reported in the paper), but given time constraints had to settle for this approach. ↑
11. Note that for Tanzania, Sanku has a larger representation: a rough estimate we conducted suggested that the fraction of flour fortified with Sanku premix approaches 50% of all flour consumed in the country. ↑
12. We use the relative bioavailability comparisons produced by the World Health Organization (WHO, 2006: p98). However, these are estimates and can vary between populations, especially due to diet and malaria-driven inflammation. Because the three active Sanku countries have relatively low malaria prevalence for Sub-Saharan Africa, we do not account for this in the model; however, we discuss it further with respect to the expansion countries in the Framework for Expansion section. ↑
13. In GBD, maternal deaths are modeled to be attributable to the risk factor “iron deficiency’,” rather than the cause-of-death category “dietary iron deficiency’.” However, Demewoz Woldegebreal, who is involved in micronutrient deficiency modeling for the GBD, told us that using dietary intake with these estimated maternal deaths is a valid approach. ↑
14. An additional complication with folic acid and neural tube defects is that the relevant consumers of folic acid (women of child-bearing age) are different to the beneficiaries of this consumption (their children). We therefore adapt our method by assigning all neural tube defect burden to the ‘women aged 15–49” group. ↑
15. For channel 1 (cognitive benefits for children), we assume a supplementation-dose figure of 63 mg/day, based on the following supplementation studies: Pollitt et al. (1989), Seshadri et al (1989). ↑
16. For Channel 2 (cognitive benefits for adults), we assume a supplementation-dose figure of 61 mg/day, based on the following supplementation studies: Elwood 1970, Beard 2005, Murray-Kolb 2007, Leonard 2014, Murray-Kolb 2017, Wenger 2017. ↑
17. For Channel 3 (physical productivity benefits for adults), we assume a supplementation-dose figure of 123 mg/day, based on the following supplementation studies: Edgerton 1979, Ohira et al., 1981, Basta et al. 1979, Li et al. 1994, Zhu and Haas. 1998 ↑
18. According to Sanku, ferrous fumarate costs ~$2.10/kg, while NaFeEDTA costs ~$3.81/kg. However, ferrous fumarate has a higher density of iron (~32%) than NaFeEDTA (~12%) so delivering the same level of iron requires roughly 2.7x more NaFeEDTA (~394 g/kg of premix vs ~147 g/kg for fumarate). This higher inclusion has a compounding effect because it displaces the carrier, which is relatively inexpensive at ~$0.82/kg, and thus removes a low-cost component that normally helps dilute the overall premix cost. In practical terms, this results in an increase of ~40% in premix cost for Kenya-type maize formulations and ~60% for Tanzania-type formulations. ↑
19. Note that we also do not include the costs of government relations work in our primary results. ↑