AI can improve vaccine supply chains by forecasting demand, predicting stock-outs, flagging cold chain temperature failures and optimising deliveries. These are solvable technical problems. But analysis of WHO and UNICEF data shows most national vaccine stock-outs come from funding and procurement delays, not poor forecasts, and no algorithm pays a late invoice.
A district immunisation officer opens her dashboard on a Monday morning. The new forecasting tool, trained on two years of stock data, is flashing amber: at current use, the district will run out of pentavalent vaccine in 19 days. It is right. It was also right last quarter, when the same warning appeared, the officer sent the same requisition, and the national store replied that its own stock was low because the ministry’s payment to UNICEF’s procurement arm had been held up in the finance ministry.
The children missed their doses anyway. The forecast was accurate. It simply had no power over the reason the vaccine was missing.
This is the gap that most AI-for-supply-chain pitches skip. Predicting stock-outs has become a tractable data problem, and AI can genuinely help with it. But most vaccine stock-outs are not forecasting failures; they are financing and procurement failures that show up in the supply chain. Better prediction is worth having. It is not the fix.
How often do vaccines run out?
More often than most people outside immunisation programmes realise. A study published in Vaccine in 2017, led by Patrick Lydon and drawing on data countries report to WHO and UNICEF, examined stock-outs between 2011 and 2015. Every year, on average, one in three WHO member states had at least one national stock-out of at least one vaccine lasting at least a month. In sub-Saharan Africa the figure was 38% of countries. Vaccines containing diphtheria, tetanus and pertussis components, and BCG, accounted for most events.
Stock-outs at national level rarely stay there. The same study found that when a national stock-out occurred, there was an 89% chance of a stock-out further down the chain, at district level or below.
The study’s most important finding was about causes. In 80% of cases, national stock-outs were attributed to reasons within the country rather than global supply shortages. More specifically, 39% of stock-outs were attributed to government funding delays, 23% to delays in procurement processes, and 18% to poor forecasting and stock management at country level.
That breakdown is the core of this argument. Forecasting and stock management, the problems AI is best at, accounted for fewer than one in five stock-outs. Money and paperwork accounted for more than three in five.
What can AI actually do for vaccine supply chains?
It would be wrong to dismiss the technology, because the problems it can address are real.
Demand forecasting. Traditional forecasts often rely on target populations derived from old census data, multiplied by assumed coverage and wastage rates. Machine learning models can use actual consumption, seasonal patterns, campaign schedules, population movement and facility-level trends to produce more accurate, more local forecasts.
Stock-out prediction. Combining stock levels, consumption rates and delivery lead times, models can flag facilities likely to run out, early enough to redistribute stock from facilities that have surplus.
Cold chain monitoring. Remote temperature monitoring devices in refrigerators generate continuous data. Algorithms can detect patterns, such as a fridge that is starting to fail or a power supply that drops every afternoon, and prioritise repairs before vaccines are damaged.
Delivery optimisation. Route planning and delivery scheduling can reduce costs and missed deliveries, especially where transport is scarce.
Wastage reduction. Better matching of multi-dose vial sizes, session sizes and demand can reduce the vaccine thrown away at the end of the day.
Each of these improves the 18% slice and makes the rest of the chain more efficient. None of them touches the 39% or the 23%.
That is not a reason to skip them. A programme that cannot see its own stock cannot argue for money, cannot redistribute between districts and cannot tell a finance ministry exactly what a delay will cost. Good data is the precondition for every fix that follows. The mistake is treating the precondition as the cure, and buying a forecasting platform in the belief that the stock-outs will stop by themselves once someone can see them coming.
Why can’t better forecasting prevent most stock-outs?
Because the causes sit upstream of the data.
Funding delays. Many countries co-finance vaccines with Gavi or buy them outright, often through UNICEF’s Supply Division. Payment requires budget allocation, release of funds by the finance ministry, and transfer. Each step can stall for reasons unrelated to health: revenue shortfalls, currency problems, political disputes or simple bureaucratic queues. An accurate forecast arriving in the health ministry does not move money in the finance ministry.
Procurement delays. Tender processes, customs clearance, registration and paperwork can add months. A forecast that says “order now” is only useful if ordering is fast.
Last-mile funding. Even when vaccines reach the district store, getting them to health facilities depends on fuel, vehicles and per diems for staff. These operating budgets are often the first cut when money is tight.
The global squeeze. Gavi’s replenishment for 2026 to 2030, in June 2025, fell short of its $11.9 billion target, raising around $9 billion after the United States said it would not contribute. Tighter donor funding means more pressure on countries’ own budgets, which is where the biggest share of stock-outs already originates.
In each case, the forecast can make the problem visible earlier. It cannot solve it. As our piece on how the next hospital crisis will arrive through your supply chain argued, supply failures tend to be financial and contractual long before they are logistical.
A worked example: the forecast that couldn’t pay
This is a hypothetical example with illustrative numbers, not data from a real country.
A country introduces an AI forecasting and stock-visibility system across its 80 districts. In the first year, forecast accuracy at district level improves markedly compared with the old population-based method, and the system correctly predicts most district-level stock-outs at least three weeks ahead.
The number of district stock-outs barely changes.
An evaluation finds out why. Of the predicted stock-outs, a minority were resolved by redistributing vaccine between districts, which the new system made much easier. That is a real gain. But most occurred because the national store itself was low, after the government’s co-financing payment was delayed by four months during a fiscal crisis. The warnings were accurate and ignored, not through negligence but because nobody receiving them could release money.
In year two, the ministry changes one thing. It negotiates with the finance ministry a small revolving fund for vaccine procurement that can be drawn on automatically when national stock falls below a set level, repaid when the regular budget is released. It also agrees a protected transport budget for districts. Stock-outs fall substantially.
The AI system did not become more accurate in year two. It became connected to money that could act on what it knew.
The strongest counterargument: isn’t visibility the first step to accountability?
A fair critic would say this argument undersells what data can do. Before digital systems, stock-outs were often discovered only when a mother arrived at an empty clinic. Accurate, real-time forecasts make stock-outs predictable, which makes them attributable. When a finance ministry knows that a delayed payment will cause a stock-out in 19 days, affecting a known number of children, the delay becomes harder to justify. Visibility creates political pressure. Moreover, the 2017 study relied on country self-reports; better data may show that forecasting failures were more common than people realised, hidden behind other explanations.
This is a fair point, and it is the best case for investing in AI supply chain tools. Visibility can change incentives, and good data can reveal causes that were misattributed.
But visibility creates accountability only if someone is accountable. The critic’s case works when forecasts are shared with finance ministries, parliaments, donors and the public, and when there are consequences for delays. It fails when forecasts stay inside the health ministry’s dashboard. The technology is necessary. The governance arrangements that make it matter are separate, and they are rarely funded alongside it.
What this means for ministries, donors and developers
For health ministries, the most valuable use of AI forecasting may be as evidence in budget negotiations. Present finance ministries with specific, dated predictions: without this payment by this date, these districts will run out and this many children will miss doses. Pair forecasting investments with financing mechanisms, such as revolving funds or pre-approved payment triggers, that can respond automatically. Our piece on how the procurement office is the most powerful quality regulator in your health system applies directly.
For donors, fund the boring parts that forecasts expose: bridge financing, procurement reform, last-mile transport budgets and cold chain maintenance. Consider tying support for digital supply systems to agreements on how their warnings will be acted on. And recognise that with tighter global funding, the risk is concentrated in countries’ own budget processes.
For developers, measure success by stock-outs prevented and children vaccinated, not forecast accuracy. Build tools that show the cause of a predicted stock-out, whether national stock, funding, transport or distribution, so that warnings reach the person who can fix that specific cause. And be honest in sales pitches about which part of the problem your product addresses.
Finally, remember the family at the end of the chain. As our piece on vaccine hesitancy argued, access failures and hesitancy are linked: a parent turned away from an empty clinic may not come back.
Frequently asked questions
What causes vaccine stock-outs?
According to a study of WHO and UNICEF data published in Vaccine in 2017, about 80% of national vaccine stock-outs had causes within the country rather than global supply shortages. The most common were government funding delays (39%), procurement delays (23%), and poor forecasting and stock management (18%). Global supply constraints, such as limited manufacturing capacity for specific vaccines, account for a smaller share.
How does AI help with vaccine cold chains?
AI can analyse data from remote temperature monitoring devices to detect refrigerators that are beginning to fail, identify recurring power problems and prioritise maintenance before vaccines are damaged. It can also help plan where new cold chain equipment is most needed. These tools reduce vaccine wastage but depend on reliable devices, connectivity and a maintenance system able to respond to alerts.
Can machine learning predict vaccine demand?
Yes, and often better than traditional methods based on estimated target populations. Machine learning models can use actual consumption, seasonal patterns, campaign schedules and local trends to forecast demand at facility or district level. Better forecasts reduce both stock-outs caused by under-ordering and wastage caused by over-ordering, but they cannot prevent stock-outs caused by funding or procurement delays.
How is vaccine supply funded in low-income countries?
Many low-income countries receive vaccine support from Gavi, the Vaccine Alliance, and contribute co-financing from their own budgets, with shares rising as their economies grow. Vaccines are often procured through UNICEF. Governments also fund in-country distribution, staff and cold chain operations. Delays in any of these funding streams can cause stock-outs, even when forecasts are accurate.
The takeaway
AI can tell a health ministry, with impressive precision, when the vaccine will run out. It cannot sign a payment order. Until forecasting is wired to money that moves automatically, the most accurate prediction in the world will mostly document the stock-outs it foresaw.
In your experience, what really causes vaccine stock-outs: the forecast, the funding or the road?
