From the Emergency Services edition of November 7, 2026
Wellington ambulance ML recommendations improved simulated average and 90th-percentile response times; live parallel run slated for mid-2027
Victoria University of Wellington / Weewoo / Wellington Free Ambulance · EMS dispatch / ambulance resource allocation · Wellington, New Zealand
- Publisher
- Te Herenga Waka—Victoria University of Wellington
- Original publication
- 2026-11-03
- Source retrieved
- 2026-11-08
What happened
A 3 November 2026 Victoria University of Wellington comment by PhD researcher Jordan MacLachlan describes a machine-learning sidecar, funded by an MBIE Smart Ideas grant and supervised by Professor Yi Mei, that presents dispatch recommendations in a fraction of a second which dispatchers can accept, change, or ignore. Identical situations return identical recommendations so decisions can be audited. In late 2024, Health New Zealand asked publicly traded Norwegian firm Omda to simulate one year of de-identified historic Wellington Free Ambulance emergencies; compared with a dispatcher-style benchmark co-designed with the service, the model improved average and 90th-percentile response times across all urgency categories. That result led Health New Zealand, ACC, and Wellington Free Ambulance to support a phased real-world trial. A six-month passive parallel run on data de-identified in real time is planned to start in mid-2027, comparing every dispatcher decision with the model without affecting live dispatch. The opened page does not publish a seconds-saved figure. MacLachlan’s team has started a company, Weewoo (six people). Context cited: Hato Hone St John and Wellington Free Ambulance attended more than 640,000 incidents in the 12 months to July 2026, up more than 12% on the 12 months to July 2022; the average Wellington paramedic lasts only seven years on the road, with roughly 28% still working a decade after registration versus around 73% of doctors.
Why it matters
Relevant to U.S. EMS medical directors and 911/ECC boards considering AI that recommends which ambulance goes where: this is third-party historic simulation plus a planned passive parallel run, not a live deployment. Transferability is limited by New Zealand fleet rules, urban/rural mix, and a researcher-authored account from the founder of Weewoo. Why now: published 3 November 2026 and confirmed not in the emergency-services archive, and it is the first public description of the Health NZ/Omda year-long replay and the mid-2027 Wellington Free Ambulance parallel evaluation.
Evidence and measured results
University page dated 3 November 2026 (article:published_time 2026-11-03T16:42:41+13:00). First-person account of Omda’s historic-year simulation and partner support for a mid-2027 six-month passive parallel run. No published effect-size in seconds on this page.
Limitations and uncertainty
First-person commentary by the researcher who founded Weewoo; Omda results are summarized without a published seconds-saved figure, confidence interval, or open methods paper on this page; simulation of historic year is not live, messy operations; New Zealand context does not automatically transfer to U.S. PSAPs.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-11-08; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- EMS dispatchers must assign scarce ambulances under uncertainty while demand rises and paramedic tenure is short, without a safe way to test allocation AI on live calls.
- Stakeholders
- EMS medical director
- Dispatch/ECC manager
- Ambulance operations chief
- Workforce/union representatives
- Privacy officer
- State or regional EMS authority
- Discovery questions
- Can a year of de-identified CAD history be released to an independent evaluator?
- What dispatcher-agreed benchmark defines a better assignment?
- Will leadership accept a multi-month passive parallel run before any live control?
- How will crew meal-break and overtime effects be measured alongside response time?
- Value hypothesis
- If a recommend-only model beats a local dispatcher benchmark on historic average and tail response times, a passive parallel run can show whether those gains survive messy live data—without exposing patients to untested automation.
- Potential engagement
- Historic CAD replay design and a scoped observation-only parallel-run protocol with an independent evaluator.
- Unsupported claims
- A specific number of seconds saved on this page
- Live response-time improvement in Wellington or any U.S. system
- That Weewoo is a production-proven product
- Automatic transfer of New Zealand results to U.S. EMS
Pre-sales engineering
Role takeaway
- Fit
- Fits EMS CAD/dispatch platforms that can emit de-identified incident and crew-status events and display a non-binding next-assignment recommendation.
- Architecture and integration
- CAD event stream to scoring service to recommendation pane with accept/change/ignore and immutable logs; parallel-run mode writes scores only, never unit assignment.
- Prerequisites
- De-identified historic CAD extract covering a full year
- Agreed dispatcher benchmark definition
- Real-time de-identification for a parallel run
- Deterministic replay so identical states yield identical recommendations
- Constraints
- No seconds-saved figure published here
- Historic simulation ignores some live data quality issues
- New Zealand operating rules and geography
- Security
- Treat location, clinical, and crew data as restricted EMS records; require de-identification, access control, and prohibition on unauthorized model training.
- Proof of value
- Historic-year replay vs dispatcher benchmark on mean and 90th-percentile response by urgency
- Passive parallel agreement/disagreement report without affecting dispatch
- Audit that identical inputs produce identical recommendations
Delivery
Role takeaway
- Work
- Stand up a no-control evaluation: data-use agreement, historic replay, dispatcher training on overrides, and a timed parallel observation.
- Dependencies
- Ambulance service data owners
- Independent technical evaluator
- Privacy/legal review
- Dispatch supervisors
- Ownership
- EMS operations owns whether recommendations may later affect assignment; the evaluator owns the replay design; privacy owns de-identification.
- Skills and adoption
- Dispatchers need practice ignoring or amending recommendations without extra workload, and supervisors need an audit method for override patterns.
- Governance checkpoints
- Data-use and de-identification approval
- Benchmark sign-off with dispatch leadership
- Go/no-go after historic replay
- Go/no-go after the passive parallel period
- Acceptance criteria
- Documented improvement on the agreed historic metrics
- Parallel run completed without the model changing live dispatch
- Override and audit procedures tested
- Risks
- Researcher/vendor narrative outrunning unpublished methods
- Staff distrust if recommendations conflict with local practice
- Privacy breach if de-identification fails
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Treat dispatch AI as a recommend-only sidecar beside existing CAD/dispatch tools: ingest live incident and crew-status updates, return an auditable recommendation in under a second, and keep human accept/change/ignore as the control path. A passive parallel run that scores the model against de-identified real-time decisions is the integration pattern before any write-back to assignment.
Governance
Who approves, reviews and stays accountable for outcomes?
Health NZ commissioned third-party historic validation before partners would even endorse a non-interfering parallel trial. U.S. agencies should similarly require an independent replay against local historic CAD data, a written benchmark agreed with dispatch leadership, and a no-control observation period before the model can change which unit is sent.
Security and privacy
What data, permissions and controls need testing?
The described evaluations use de-identified historic and (planned) real-time parallel data. EMS location, call, and crew data remain highly sensitive; contracts must specify de-identification, no secondary training use without authorization, and audit logs of every recommendation.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
The author ties dispatch quality to paramedic burnout and short tenure (average seven years on the road in Wellington; roughly 28% still working a decade after registration). A model that also reduces interrupted meal breaks could aid retention, but this page does not measure workforce outcomes, and dispatchers must still understand and override recommendations.
Procurement
What should contracts, pricing and exit terms secure?
Procure evaluation rights—historic replay, independent validator, and a no-control parallel period—before buying live assignment automation. Treat Weewoo as an emerging vendor spun out of the research team, not as an incumbent CAD module.
Operating model
Which teams own the service once it runs?
Keep humans as the assignment authority through at least a passive parallel phase; use the model to surface options and create an audit trail rather than to remove the dispatcher from the loop.
Publication history
- 2026-11-07Emergency Services · Issue 624 resources
Stable resource ID: wgtn-weewoo-omda-wellington-ambulance-dispatch-ml-20261103