Mexico's new health-sector digital transformation council puts AI and operations research behind a concrete public-service challenge: getting medicines to the places where they are needed. Its proposed approach shows why better forecasting must be matched by reliable information, clear responsibilities and evidence that delivery is improving.
On 23 October 2026, Mexico announced an advisory council for the digital transformation of its health sector. The official Spanish-language announcement identifies three priorities: requesting supplies in real time, managing warehouses and inventories, and organising last-mile delivery. It links these activities to demand prediction and logistics optimisation.
The stated timetable is ambitious: initial results within a month and implementation envisaged from January 2027. These are government objectives, not evidence of a completed system. The announcement does not establish that shortages have fallen or that distribution costs have already improved.
The case is worth examining because it connects AI to a service people can recognise. The relevant question is whether medicines arrive reliably, rather than how many algorithms are introduced. The analysis that follows considers the organisational conditions such an initiative would need; its practical examples are illustrative, not reported findings from the Mexican programme.
Why purchasing and availability are different problems
In the presidential press-conference transcript, the government describes delays between purchasing medicines and getting them through warehouses, hospital pharmacies and onward to patients. It presents traceability across that chain as an objective, while acknowledging that the complete system does not yet exist.
That distinction changes how a digital project should be evaluated. Purchasing enough stock and knowing where it is are related but separate capabilities. A dashboard may show a positive inventory while a particular facility cannot access the items it needs. Resolving that mismatch requires both information and the ability to act on it.
For example, imagine a district warehouse holding a medicine that a nearby facility has requested urgently. A useful system would make the request visible, establish whether the recorded stock is actually available and route the case to someone authorised to arrange delivery. A prediction alone cannot complete those steps.
Build a shared picture before optimising the network
The first governance challenge is deciding what information institutions can trust. A stock record might describe physical items on a shelf, quantities allocated to another facility or a shipment that has not yet arrived. Treating those categories as interchangeable could produce a convincing but misleading picture of availability.
A practical design would distinguish these states and assign responsibility for updating them. It would also record when information was last checked. Staff should be able to identify uncertainty rather than being forced to enter an apparently precise quantity that they cannot verify.
This is where administrative coordination becomes as important as software. Common definitions, clear handover records and a process for resolving discrepancies would help institutions work from comparable information. Those arrangements are proposed design considerations, not features whose implementation has been demonstrated by the announcement.
Forecasting should inform decisions without hiding uncertainty
Demand forecasting could help planners anticipate needs, but its usefulness would depend on what historical records represent. In a hypothetical facility, low dispensing figures might indicate low demand, or they might reflect a period when the medicine was unavailable. A system that cannot distinguish these circumstances could perpetuate an existing gap.
Planners would therefore need ways to compare forecasts with local knowledge and explain substantial differences. They should be able to see when the underlying evidence is incomplete, and to record why they adjusted a recommendation. That record can support later review without pretending that every deviation is a mistake.
It is equally useful to separate technologies by function. A logistics initiative may combine ordinary inventory software, optimisation methods and predictive models. Calling the overall programme AI does not establish that every component uses machine learning or requires a generative assistant. The appropriate method should follow the operational problem.
Make decision rights visible across institutional boundaries
When a recommendation crosses from one organisation to another, responsibility can become blurred. Who approves a redistribution? Who confirms that goods were received? Who investigates when a shipment recorded as complete has not reached its destination?
Answering those questions in advance would make a shared platform more useful. Each critical handover needs an identifiable owner and an escalation route. An unresolved request should reach a person able to intervene, rather than circulate between teams that can each see the problem but lack authority to resolve it.
Public-sector capability is relevant here. The government says it is bringing together researchers from public institutions and a younger technical team associated with its digital transformation agency. That creates an opportunity to connect scientific expertise with operational knowledge. Whether the connection becomes durable will depend on implementation, staffing and how knowledge is retained.
Evaluate reliability, fairness and the work of frontline staff
A programme focused only on average delivery speed could miss important differences between facilities. A national improvement might coexist with persistent delays in places that are harder to serve. Evaluation should therefore examine the distribution of performance, alongside the overall result.
Possible measures include the proportion of requests fulfilled, the duration of unresolved requests, inventory discrepancies and the additional checking work required from staff. These are suggested indicators, not published results or official targets for this initiative. Establishing a baseline would make later claims of improvement easier to assess.
Comparative research reinforces the need for this broader view. The OECD study of 200 government AI use cases across 11 functions places public-service delivery alongside questions of risk, safeguards and institutional readiness. The lesson for a logistics project is to evaluate the service arrangement around the technology as carefully as the model itself.
A staged rollout would make the January ambition testable
An initial milestone should reveal what has actually been built: a design, a prototype, a limited operational trial or a service running at scale. Those stages carry different levels of evidence. Presenting them clearly would help the public distinguish progress from a promise.
A limited trial could test the quality of records, the handling of exceptional cases and the ability to continue operating if a platform becomes unavailable. Staff feedback would be particularly valuable where entering data competes with immediate service demands. A process that is too cumbersome may undermine the information on which it depends.
Mexico has set out a concrete ambition for digital public administration. Its wider significance lies in the connection between analytical capacity and the everyday ability to deliver. Success would mean that institutions can see a problem, assign responsibility and correct it, with evidence that people receive a more reliable service.
Illustrative image: Hospital, Mexico City, photographed on 19 November 2014. Photo: Ricardodaniel.maldonado, via Wikimedia Commons. Licensed under CC BY-SA 4.0.