The Short Answer
AI agents could plausibly cut platelet waste by coordinating transfers, but the proven gain so far comes from a sharing policy, and AI results are mostly simulated. A 2020 paper put US outdating near 16%. The hard part is shared inventory data and permissions. With ibl.ai you own all the code and the data, behind your firewall.
A recent social post described a hospital developer building an agent for platelet sharing. We could not find a named system, so nothing here rests on it. The idea can be tested against published evidence, and that evidence is useful.
Why do hospitals waste platelets?
Hospitals waste platelets because the product expires in days and demand is hard to predict.
Storage in the US was limited to five days for many years because of bacterial risk, and a 2020 paper put the national outdate rate at about 16%. Seven-day storage is now possible with cleared safety measures.
That 16% is the only national figure we found, and it is six years old. Treat it as a rough size, not a current rate. A UK study describes a shelf life of 3 to 5 days, and reports that 8 to 10% of issued platelets came back unused from wards, many too near expiry to reuse.
Does sharing platelets between hospitals reduce waste?
Yes, and the strongest evidence involves no AI at all.
A study in Operations Research developed a transshipment policy for moving platelets between two hospitals. Implemented in the Emory University Hospital System, it reduced outdates by about 20%, and summaries of the paper report no increase in shortages.
That result sets the bar. A written rule, followed reliably, moved the number. Before a hospital buys an agent, it should ask whether it has already tried the rule.
What has AI added to platelet inventory management?
AI has added forecasting and smarter decisions, mostly in research settings. A 2025 review in Acta Haematologica calls these advances "largely exploratory" and says models need rigorous validation.
| Approach | Result | Evidence type |
|---|---|---|
| Sharing policy between two hospitals (Emory) | About 20% fewer outdates | Implemented |
| Issuing the youngest unit when a return is predicted (UK) | About 14% less wastage | Simulation at one hospital |
| Reinforcement learning across a hospital network | Lower wastage and cost than base-stock ordering | Modelled study |
| Facility-level optimization, across reviewed studies | Outdating roughly halved, for forecasting at facility level | Systematic review, mostly retrospective simulation |
The network study is closest to the sharing idea. It models near-expiry units moving from smaller local hospitals to larger ones. It is still a model, and the review's figures come from varied settings, so don't add these gains together.
What would an AI agent do in hospital blood logistics?
An agent would handle the coordination around a decision that a person still makes. That work already exists.
Through Versiti's Platelet Partners in Michigan, hospitals post same-day-expiring units for hospitals in other systems to claim. Shipping is arranged between the two hospitals, because Versiti does not provide transportation.
A sensible first agent would do four things, none of them clinical:
- Watch expiry dates against a forecast and flag units at risk.
- Find nearby partners with a likely need, and draft the offer.
- Chase the reply, and draft the transport request.
- Log every step for the blood bank's records.
The blood bank's staff approve each transfer. Keep the agent out of transfusion decisions. As our look at hospital agent testing found, tasks that change a record are where agents do worse, so a person stays on every write.
Why is hospital logistics a better first AI project than diagnosis?
Hospital logistics is a good first project because the result is countable and a mistake is easier to catch. Units expired is a number a hospital already tracks. A diagnosis agent needs clinical validation, and its errors can reach a patient.
This isn't a claim that clinical AI is wasted. It is a question of order. Logistics also teaches an organization what it needs next: clean data, permissions, audit and a human in the loop.
What stops hospitals from sharing platelet inventory with AI?
Data access and agreements stop them, more than models do. Inventory lives in separate systems, and hospitals in different groups have little reason to expose it. As we have argued, healthcare AI tends to fail at the information layer.
The technical need is an ontology: a shared map that says a unit at one hospital is the same kind of thing as a unit at another, with an expiry, a type and an owner. Connectors then read each system in place, scoped by role, so nobody hands over a full copy of their data.
How does ibl.ai support hospital operations agents?
On ibl.ai, agents reach hospital systems through connectors, with Roles limiting access and Transcripts keeping the record of each conversation. Skills are plain Markdown, so a blood bank lead can write the procedure.
With ibl.ai you own all the code and the data, behind your own firewall. The platform is model-agnostic across any LLM, has no per-seat pricing, and you can deploy anywhere. See our guide to self-hosted AI for hospitals.
ibl.ai is family-owned and operated from New York, NY.
Sources: the Emory 20% figure is from the journal's abstract. Other figures are from the linked papers and reporting. None measures an AI agent in live blood logistics.