NHS England stroke networks
Verified evidenceExceptions onlyAI can surface suspected stroke-transfer cases before sequential specialist communication finishes.
Customer service · England, United Kingdom
Collections: Human still decides
Executive brief
The value came from collapsing sequential interpretation and transfer handoffs, not from replacing the stroke specialist.
AI value · Door-in-door-out time at an acute stroke center
Verified79 minutes after deployment, a 62-minute reduction
Retrospective before/after observational study with small cohorts; other temporal effects cannot be fully excluded.
Before
Acquire CT imaging and wait for local interpretation and sequential specialist consultation. → Review images and decide whether to accept the patient for thrombectomy.
After
Analyzes CT images and rapidly surfaces features consistent with large-vessel occlusion. → Review shared images and AI output in parallel, decide treatment and transfer, and initiate transport.
Human boundary
AI supports scan interpretation; clinicians decide diagnosis, referral acceptance, thrombolysis, thrombectomy, and transfer.
Why it matters
AI can surface suspected stroke-transfer cases before sequential specialist communication finishes.
Before
How the work ran before the change.
Acute stroke center clinician
Acquire CT imaging and wait for local interpretation and sequential specialist consultation.
ControlLocal escalation pathway
Specialist team
Review images and decide whether to accept the patient for thrombectomy.
ControlClinical decision
What changed
AI can surface suspected stroke-transfer cases before sequential specialist communication finishes.
Decision rightAI handles the default; humans own exceptions
After
How the same work runs now.
e-Stroke AI
Analyzes CT images and rapidly surfaces features consistent with large-vessel occlusion.
ControlDecision-support only
Clinical teams
Review shared images and AI output in parallel, decide treatment and transfer, and initiate transport.
ControlClinicians retain diagnosis and treatment rights
Exception path
Clinicians override or disregard AI output when imaging, symptoms, or eligibility criteria conflict; standard stroke pathways remain available.
Work removed
Decision authority
AI supports scan interpretation; clinicians decide diagnosis, referral acceptance, thrombolysis, thrombectomy, and transfer.
Before
Acute stroke center clinician
Acquire CT imaging and wait for local interpretation and sequential specialist consultation.
Control: Local escalation pathway
Specialist team
Review images and decide whether to accept the patient for thrombectomy.
Control: Clinical decision
After
e-Stroke AI
Analyzes CT images and rapidly surfaces features consistent with large-vessel occlusion.
Control: Decision-support only
Clinical teams
Review shared images and AI output in parallel, decide treatment and transfer, and initiate transport.
Control: Clinicians retain diagnosis and treatment rights
Work that left the path
Human role before
Local clinicians and specialists interpreted and relayed scans through a sequential referral process.
Human role after
Local and specialist clinicians act on shared, AI-prioritized imaging in parallel while retaining treatment and transfer authority.
AI roleImaging decision support that detects and quantifies time-critical stroke features and accelerates sharing across the network.
141 minutes before e-Stroke deployment→79 minutes after deployment, a 62-minute reduction
14 months before versus 12 months after 2020-03-01 deployment · One U.K. acute stroke center; 47 referred patients across pre/post cohorts
Retrospective before/after observational study with small cohorts; other temporal effects cannot be fully excluded.
Anti-pattern
Publishing the tripling in functional independence as causal without carrying the observational design and small sample.
Questions
Portability conditions
Reputation risk
low
Evidence and authority
Current · updated
1 peer reviewed, 1 primary; publication outcomes are verified.
Bundle 1.0.0 · reviewed 2026-08-23 · stable ID b967943b4e81c041
Ocado Group / Ocado Retail · Exceptions only
Robotic arms can pick more grocery items on-grid while people handle exceptions and fulfillment operations.
Vodafone Germany · Exceptions only
A service bot does not need to follow a fixed decision tree and escalate nearly every complex request; generative AI can resolve more requests while people handle exceptions and review response quality.
Waymo · Exceptions only
Within a defined domain, automated driving can control the vehicle while remote assistance handles exceptions.
Georgia State University · Exceptions only
Enrollment support does not require counselors to initiate and answer every routine student interaction; a task-aware assistant can provide personalized outreach while routing exceptions to staff.