---
title: "Can AI Agents Cut Platelet Waste? Hospital Logistics Evidence"
slug: "hospital-ai-logistics-platelet-sharing-waste"
author: "ibl.ai Engineering"
date: "2026-11-02 12:00:00"
category: "Premium"
topics: "healthcare ai, hospital logistics, platelet inventory, blood bank, ai agents, hospital operations, supply chain, data ontology, self-hosted ai"
summary: "Platelets last days, and a 2020 paper put US outdating near 16%. Sharing policy cut outdates about 20% at Emory, and simulation points to more. The evidence favours hospital logistics as an agent job, and the hard part is shared data."
banner: ""
thumbnail: ""
linkedin: |
  A platelet unit lasts days. One hospital can throw units out while another runs short. That is a logistics failure, and logistics may be where hospital AI pays off soonest.

  The evidence is better than the hype, and more modest.

  A 2020 paper put the US outdate rate at about 16%, while noting its cause: a very short shelf life.

  At Emory, a sharing policy between hospitals, with no AI in it, cut outdates by about 20% without adding shortages.

  In the UK, a simulation of a smarter issuing rule cut wastage by about 14%.

  Notice which result is implemented. The plain rule. The AI gains are mostly simulation, and deployed agents are still mostly a promise.

  In Michigan, hospitals can already post expiring units for others to claim, and they arrange the shipping themselves. That is the job an agent would do: match, draft, chase, and hand a person the decision.

  The barrier isn't the model. It is getting inventory data from separate systems, and permission to share it.

  #iblai #HealthcareAI #HospitalOperations #AIAgents #SupplyChain
---

## 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](https://pubmed.ncbi.nlm.nih.gov/32581035/) 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](https://integratedcarejournal.com/how-operational-research-and-ai-are-transforming-hospital-blood-banks/), 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](https://pubsonline.informs.org/doi/fpi/10.1287/opre.2022.2410) 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](https://pmc.ncbi.nlm.nih.gov/articles/PMC12176357/) calls these advances "largely exploratory" and says models need rigorous validation.

<table>
<thead><tr><th>Approach</th><th>Result</th><th>Evidence type</th></tr></thead>
<tbody>
<tr><td>Sharing policy between two hospitals (Emory)</td><td>About 20% fewer outdates</td><td>Implemented</td></tr>
<tr><td>Issuing the youngest unit when a return is predicted (UK)</td><td>About 14% less wastage</td><td>Simulation at one hospital</td></tr>
<tr><td>Reinforcement learning across a hospital network</td><td>Lower wastage and cost than base-stock ordering</td><td>Modelled study</td></tr>
<tr><td>Facility-level optimization, across reviewed studies</td><td>Outdating roughly halved, for forecasting at facility level</td><td><a href="https://pubmed.ncbi.nlm.nih.gov/42541243/">Systematic review</a>, mostly retrospective simulation</td></tr>
</tbody>
</table>

The [network study](https://www.sciencedirect.com/science/article/pii/S1366554525006416) 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](https://versiti.org/products-services/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](/blog/hospital-ai-agents-single-turn-tests-grade-the-actions) 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](/blog/healthcare-ai-fails-at-the-information-layer), healthcare AI tends to fail at the information layer.

The technical need is an [ontology](/product/agentic-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](/docs/os/organization-settings/roles) limiting access and [Transcripts](/docs/os/analytics/transcripts) keeping the record of each conversation. [Skills](/docs/os/agent-settings/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](/blog/self-hosted-ai-for-hospitals-and-health-systems).

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.*

## Why does owning the AI stack matter?

**ibl.ai is the agentic AI platform where you own all the code and the data. You self-host the entire stack inside your own perimeter, run it model-agnostic across any LLM and switch anytime, and pay by usage with no per-seat pricing — so you can deploy anywhere: your cloud, on-premise, GovCloud, or fully air-gapped.**

- **You own all the code and the data.** Full source code under a perpetual license, running on your infrastructure. Not API access to someone else's platform — the stack itself is yours.
- **Model-agnostic.** Run any LLM — Claude, GPT, Gemini, Llama, Command, or your own fine-tune — and switch providers without rewriting the platform.
- **No per-seat pricing.** Usage-based billing against a budget cap you set. Cost tracks what your organization actually uses, not how many people you employ.
- **Deploy anywhere.** Your cloud, your VPC, on-premise, GovCloud, or a fully air-gapped network with no outbound connectivity.

1.6M+ users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University.

ibl.ai is family-owned and operated from New York, NY — a U.S.-headquartered, domestically-owned long-term partner, not a vendor that sells licenses and moves on.