The World Food Programme used optimisation to redesign food assistance in Iraq, reducing monthly costs by 12% while maintaining almost the same nutritional value for around 800,000 people. The redesigned food basket was used for two years, saving more than $25 million, while the wider analytics programme developed around the model has since been applied across WFP operations.
Why is planning food assistance more complicated than choosing what food to distribute?
The World Food Programme operates food-assistance programmes in settings where needs can change quickly while funding, food prices and transport capacity are limited.
Planning an operation involves several connected decisions. Teams need to decide what foods will provide the required nutrition, where those foods should be purchased, how they should reach their destination and whether assistance should be delivered as food, cash or vouchers.
Traditionally, these decisions can be considered separately. But changing one affects the others. A cheaper food may cost more to transport. Buying locally may reduce delivery time but depends on local availability and prices. Changing the contents of a food basket may reduce costs, but nutritional requirements still need to be met.
For WFP, the question was therefore not simply where costs can be reduced. It was how available funding can provide nutritionally adequate food assistance to as many people as possible while accounting for the realities of the supply chain.
Why is this a prescriptive analytics problem?
Prescriptive analytics focuses on what decision should be taken given an objective and a set of constraints.
For WFP, there are millions of possible combinations of foods, suppliers, transport routes and forms of assistance. Comparing them manually becomes difficult, particularly when planners also need to account for nutrition, funding, lead times, procurement requirements and local markets.
Therefore, this project made use of Optimus, a decision-support system that evaluates these decisions together. Its optimisation model can simultaneously determine the composition of a food basket - the combination and quantity of foods provided to meet a person or household's nutritional needs - where its ingredients should be purchased and how they should be transported.
Rather than starting from a fixed basket and trying to transport it as cheaply as possible, Optimus can start from nutritional requirements and identify different combinations of foods that meet them. It then considers the supply chain needed to deliver each option.
This allows planners to ask practical questions such as: if funding falls, what can change while maintaining nutrition? If more people need assistance, what combination of food and supply routes makes that possible?
How did WFP and its research and industry partners develop the approach?
The model emerged through several years of collaboration between WFP practitioners, researchers and industry partners.
It brought together the World Food Programme, Tilburg University, the University of Amsterdam, Northeastern University and UPS, with earlier prototypes also drawing on research collaborations with the Georgia Institute of Technology.
The collaboration developed iteratively rather than beginning with a finished model. Early prototypes were tested against WFP's operational data and refined around the decisions country teams actually needed to make.
From 2015, WFP took responsibility for maintaining and developing the model with internal specialists and external partners. UPS contributed expertise and funding to the wider rollout and change-management process, while other industry partners later worked with WFP on automating the fragmented data flows needed to run tools such as Optimus at scale.
This mattered because the technical model alone was not enough. It also had to fit the way nutrition, procurement, logistics and resource-management teams make decisions together.
What changed in Iraq?
In 2015, WFP's Iraq operation was looking for ways to redesign its food basket under different levels of available funding.
Teams using Optimus worked with local colleagues to capture procurement requirements, logistics constraints, different forms of assistance and the preferences of people receiving food assistance. Hundreds of alternatives were then analysed and reviewed with the local team before a new basket was introduced.
The result reduced the monthly cost of the basket by 12%, while providing almost the same nutritional content.
At the time, the programme was providing food assistance to around 800,000 people each month. The change reduced costs by approximately $1.3 million a month. The optimised basket was distributed throughout 2016 and 2017, generating more than $25 million in total savings.
The significance becomes clearer when translated back into programme capacity. At the same funding level, the monthly saving was calculated to be enough to provide the optimised food basket to around 109,000 additional people. Alternatively, if funding fell by as much as $1.3 million in a month, WFP could still provide the same number of people with a nutritionally adequate basket.
The $25 million figure therefore represents more than an operational cost reduction. It increased the amount of food assistance that could be provided from the same funding.
Food Supplies from World Food Programme's Gallery
How did the approach allow WFP to pilot it elsewhere?
WFP applied the same model to different decisions because constraints vary from one operation to another.
In Yemen, WFP needed to examine how to expand monthly food assistance from three million to six million people while operating with limited resources and continuing conflict.
Optimisation allowed teams to compare the number of people reached, the nutritional composition of rations and the funding required. The project gave planners a way to understand what would need to change under different resource levels before scaling the operation.
During the El Niño crisis in Southern Africa, the model was used differently. Poor harvests created the risk that local food prices and availability would change, so WFP used optimisation to compare sourcing and delivery strategies under alternative scenarios.
These applications show why one optimisation model does not imply one fixed answer. The same framework can support different decisions depending on local needs, available funding, food markets and supply constraints.
How did Optimus move from individual analyses into wider use?
The early analyses required substantial manual work. Data had to be gathered, cleaned and combined before an optimisation could be run, which limited how widely the approach could be used.
WFP therefore began developing an automated, web-based version of Optimus. The interface was designed so staff did not need a background in data science to use the underlying optimisation model.
Data integration became another part of that change. Information needed by Optimus was spread across more than a dozen WFP systems. Through the DOTS data platform developed with Palantir, more than 260 linked data tables could be synchronised and updated automatically, reducing the manual preparation previously required for analysis.
Change management was also treated as part of implementation. With support from the UPS Foundation, WFP began building a network of staff who could use optimisation within regional operations. By early 2021, more than 40 Optimus power users were working across WFP regional bureaux in areas including logistics, procurement, nutrition and resource management.
The change was therefore not simply from one food basket to another. It was from optimisation being an occasional specialist analysis towards becoming a method that operational teams could use repeatedly.
What did the wider analytics programme mean for WFP?
Optimus became one part of a broader analytics system at WFP. The organisation combined it with the Supply Chain Management Dashboard, which brings together operational information and identifies potential supply problems, and DOTS, which connects and updates data from different systems. Together, these tools support a sequence from understanding what is happening, to anticipating problems, to comparing possible responses.
Across this wider analytics programme, WFP reported more than $150 million in savings, which the organisation estimated was equivalent at the time to the cost of supporting around two million food-insecure people for a year. This figure covers multiple analytics initiatives and should therefore not be attributed to Optimus alone.
Other applications included more than $100 million in reported savings in South Sudan, where analytics supported changes in transport and pre-positioning and reduced reliance on expensive air deliveries.
These results show how the value of analytics changes when it moves beyond a single model: better data, optimisation and organisational decision processes can reinforce one another.
What can this mean for future food-assistance planning?
One continuing question is uncertainty. Food prices can change rapidly during conflict, climate shocks or economic disruption. Later research built on the Optimus approach to examine whether procurement decisions should be designed to withstand uncertain prices or adjusted as new information becomes available.
The results suggest that when organisations must commit far in advance, planning explicitly for uncertainty can reduce the risk of expensive corrections later. But where plans can be revised each month, regularly updating a simpler model with the latest prices may perform just as well or better.
For food-assistance operations, the trajectory is therefore from optimising one part of the supply chain towards connecting nutrition, procurement, transport, funding and uncertainty in the same decision process. All while keeping programme teams and local operational knowledge central to how those decisions are made.
