Simulating rural medicine logistics: hub allocation, multi-modal routing (truck / motorbike / drone), in-transit tracking, and stockout prevention across a schematic rural terrain
Rural medicine supply begins at a regional distribution hub — typically serving a district or province — that receives bulk consignments from a national medical store and must decide how much of each product to push to dozens or hundreds of downstream health facilities. This allocation decision, increasingly algorithmic rather than manual, is the foundation every later logistics stage depends on.
A Logistics Management Information System (LMIS) is the software layer that turns raw consumption and stock data into an allocation decision:
Inputs to the allocation model: • Population catchment per facility — larger catchments receive proportionally larger allocations • Historical consumption trend — a rolling average of recent months, adjusted for seasonality (e.g. malaria commodities spike in rainy season) • Current reported stock-on-hand — facilities closer to stockout get prioritized in the next push • Lead time and buffer stock policy — hub typically maintains a safety stock equivalent to 1–2 months of average consumption
Push vs. pull systems: • Push: the hub decides quantities and ships on a fixed schedule regardless of facility request — simpler, but risks over- or under-supply if consumption data is stale • Pull: facilities submit requisitions based on their own stock counts — more accurate but depends on timely, accurate facility-level reporting, which is often the weakest link in low-resource settings • Most national systems now use a hybrid: pull-based ordering constrained by push-based minimum/maximum stock levels calculated centrally
Why this matters for stockout rates: • Studies across sub-Saharan Africa have found stockout rates for key essential medicines at rural facilities ranging from roughly 20% to 40% in health systems without real-time inventory visibility • UNICEF's supply chain — covering roughly 100 countries — has invested heavily in LMIS digitization specifically to reduce this gap, since allocation accuracy at the hub is the single largest lever on downstream stockout risk • Every subsequent stage in this simulation (routing, tracking, confirmation, feedback) exists to make the numbers feeding this allocation decision more accurate and more current
No single transport mode serves a rural distribution network well. Route planners segment their facility list by road condition, distance, and seasonal accessibility, then assign each delivery to the cheapest mode that can reliably reach it — reserving the most expensive option, drone delivery, for facilities where road transport is slow, unreliable, or seasonally impossible.
Route planning in a rural supply network is fundamentally a triage exercise across three cost/speed/reliability profiles:
Truck (paved and well-maintained unpaved roads): • Lowest cost per unit volume — a single trip can carry enough stock for many facilities along a fixed route • Vulnerable to road condition: a truck route that works in dry season may be impassable once rains begin • Best suited to hub-to-hub or hub-to-large-facility trunk routes
Motorbike courier (rural tracks, moderate distance): • Higher per-delivery cost than truck (more trips, more driver-hours per unit volume) but far more resilient to poor road surface • Standard "last-mile" mode across much of rural sub-Saharan Africa and South Asia for facilities a truck cannot efficiently reach • Still vulnerable to flooding, washed-out bridges, and extended rainy-season closures
Drone (extreme last-mile, rainy season, emergency resupply): • Highest per-delivery cost but essentially weather- and road-independent • Zipline's fixed-wing drone system, operating in Rwanda since 2016 and later Ghana, delivers blood products, vaccines, and essential medicines with a round-trip range of roughly 160 km • Because the drone flies point-to-point rather than following roads, it collapses a delivery that might take hours or days by road into a flight of well under an hour
Decision logic modeled in this stage: • Facilities within reliable truck-route distance and on maintained roads → truck • Facilities on rural tracks not reachable by truck but accessible by motorbike in under half a day → motorbike • Facilities that are remote, seasonally cut off, or require urgent/emergency resupply (blood products, snakebite antivenom, vaccines needing tight cold-chain windows) → drone, regardless of season • As rainy season severity increases in this simulation, more facilities shift from truck/motorbike into the drone-served tier because road transit time and reliability degrade sharply
Once a shipment leaves the hub, the supply chain's biggest historical blind spot has been the transit period itself — a consignment could be delayed, misrouted, or spoiled with no one downstream aware until it failed to arrive. GPS-enabled vehicle tracking and barcode-scanned consignments integrated into the LMIS close this visibility gap in real time.
In-transit tracking is not just a monitoring convenience — it changes what dispatchers can do while a shipment is still moving:
GPS vehicle tracking: • Delivery vehicles (trucks, motorbikes, and drones) report position at regular intervals to a central dispatch system • Dispatchers can detect a stalled or delayed vehicle within minutes rather than discovering a missed delivery window after the fact • Enables dynamic re-routing: if a bridge is reported flooded, remaining stops on that route can be reassigned to a drone or held for the next dry-road window
Barcode/QR consignment tracking: • Each shipped package is scanned at dispatch, at any transfer point, and at delivery — creating a chain-of-custody record • Integrated with the LMIS, this allows the system to distinguish "shipped but not yet arrived" from "lost" from "delivered but not yet confirmed," each of which triggers a different operational response
Cold chain monitoring: • Temperature loggers on vaccine and other cold-chain-dependent shipments report continuously; an excursion outside the safe range (commonly 2–8°C for most vaccines) triggers an immediate alert rather than being discovered only when the vial is used or wasted • This is particularly critical for drone and motorbike deliveries crossing hot, humid terrain where a passive cold box has a limited safe window
Effect on delivery time in this simulation: • As tracking-driven dynamic re-routing comes online (Stage 3), average delivery time in the model falls modestly even before any change in the underlying transport fleet — purely from avoiding wasted trips to blocked routes and catching problems while they are still fixable
A shipment that has arrived but not yet been confirmed in the system is, from the hub's point of view, still a risk. Digital proof-of-delivery — a health worker scanning or digitally acknowledging receipt at the point of arrival — updates the facility's stock-on-hand instantly, replacing the traditional model where paper requisition forms could lag reality by weeks.
Every allocation decision in Stage 1 depends on the hub's belief about current facility stock levels — and that belief is only as good as the most recent confirmed delivery data:
Legacy paper-based confirmation: • Health worker records receipt on a paper stock card • Data is aggregated manually up a district-to-national reporting chain, often monthly • By the time the hub sees an accurate stock picture, it may already be weeks out of date — during which real consumption has continued and true stock levels have drifted from the record
Digital proof-of-delivery: • Health worker scans a barcode or QR code on arrival, or manually confirms receipt via a mobile app or SMS-based system • Stock-on-hand at the facility updates in the LMIS within minutes of physical arrival • Discrepancies (e.g. quantity received differs from quantity shipped) are flagged immediately rather than discovered at the next physical stock count
Downstream effects modeled here: • More accurate, more current stock data directly improves the allocation algorithm in Stage 1 for the next shipment cycle — the feedback loop tightens • Facilities with reliable digital confirmation tend to see fewer both-direction errors: fewer false stockouts (where the hub over-ships because it believes stock is lower than it is) and fewer real stockouts (where the hub under-ships because it believes stock is higher than it is) • This stage is often the cheapest intervention in a last-mile logistics upgrade — it requires no new vehicles or fleet, only a simple mobile confirmation workflow at the facility — yet it disproportionately improves the accuracy of every other stage
The most advanced last-mile systems do not wait for a facility to report an empty shelf. Instead, consumption data flowing back from delivery confirmation and periodic stock counts feeds a reorder-point model that triggers replenishment automatically once projected stock is expected to fall below a safety threshold before the next scheduled delivery.
A predictive stockout-prevention system runs on a fairly simple underlying formula, made powerful by having timely, accurate inputs:
Core reorder-point calculation: • Reorder point = (average daily consumption × lead time in days) + safety stock buffer • When a facility's current stock-on-hand (from the latest confirmed delivery and any interim consumption reports) falls to or below this reorder point, the system automatically flags it for the next allocation cycle — no manual requisition required • Safety stock buffer is typically set to cover 1–2 months of average consumption, sized larger for facilities with high demand variability or longer/less-reliable delivery lead times (e.g. rainy-season-affected rural sites)
Why this beats reactive resupply: • Reactive systems only respond after a facility reports zero stock — by which point patients have already been turned away or given substitute treatment • Predictive systems anticipate the shortfall before it happens, giving the hub enough lead time to route a delivery, potentially via a faster (if costlier) mode like drone if the standard route cannot make the window • Facilities operating under a mature predictive feedback loop see meaningfully lower stockout incidence than the 20–40% baseline rates observed in unmanaged rural supply chains
Interaction with transport mode choice: • A facility flagged as approaching its reorder point close to a scheduled truck/motorbike delivery can simply wait for the routine trip • A facility flagged with insufficient lead time before the next routine delivery — especially during rainy season when road delivery is slow or blocked — becomes a candidate for an urgent drone resupply, directly connecting the predictive model to the mode-choice logic from Stage 2
Drone delivery is more expensive per trip than road transport, but that comparison alone misses the point: the real question is cost-effectiveness per outcome achieved, not cost per delivery in isolation. Zipline's operations in Rwanda (since 2016) and Ghana provide the most extensively documented real-world dataset on how drone delivery compares to road transport for time-critical rural medical supply.
Zipline began operating in Rwanda in 2016, delivering blood products and later vaccines and other essential medicines by autonomous fixed-wing drone from centralized distribution centers to rural health facilities; the model was subsequently extended to Ghana with a larger multi-site distribution network.
What the time comparison shows: • Before drone delivery, urgent items like blood products for postpartum hemorrhage could take hours to days to reach a remote rural facility by road, particularly during rainy season or across difficult terrain • Zipline's drone delivery reduced this to roughly 30 minutes from order to arrival for point-to-point flights within its ~160 km round-trip range • For genuinely time-critical items — blood for emergency transfusion, snakebite antivenom, vaccines needing to beat a cold-chain clock — this time reduction has direct clinical significance, not just logistical convenience
What the cost comparison shows: • Per-delivery cost for drone service is commonly cited in the range of roughly $20–40, reflecting aircraft operation, launch/recovery infrastructure, and the distribution center network behind it • Road delivery cost varies enormously by context: cheap and fast on a maintained road in dry season, but effectively unbounded (in delay cost, wasted product, and clinical risk) when roads are washed out or a facility is simply too remote for a scheduled route • The honest cost-effectiveness comparison is therefore not "drone dollars vs. road dollars" but "drone cost vs. the fully-loaded cost of road delivery including stockout risk, product spoilage, and delayed emergency care" — and that comparison favors drone specifically for the highest-urgency, most road-inaccessible cases, while routine bulk resupply remains cheaper by road
Where the model in this simulator lands: • As rainy season severity increases and drone fleet share is expanded (via the sliders), average delivery time falls sharply and stockout risk for the most remote facilities drops — while overall cost per delivery rises, reflecting the real tradeoff logistics planners navigate: use the cheap mode where it works, and reserve drone capacity for where road transport genuinely cannot compete
Key Insight: Zipline's Rwanda and Ghana data reframes the drone-vs-road question from "which is cheaper" to "which is cheaper for this specific delivery" — road remains the economical default for routine, accessible resupply, while drone economics only look favorable once you price in the clinical and logistical cost of delay for time-critical, hard-to-reach deliveries, exactly the segment it was deployed to serve.