Home/Nursing homes/Fall detection tender checklist
This AI fall detection tender checklist makes every claim testable.
An AI fall detection tender works best when each requirement comes with a test the home can run. This checklist from Havenwatch AI covers detection, accuracy, privacy, alerts, security and support.
AI fall detection tender checklist
- DetectionFalls, bed exits and residents at exits, in rooms and common areasLive demo with staged events
- AccuracyA target defined on agreed eventsCount hits, misses and false alerts on site
- ResponseAlert within seconds and escalation if unansweredTime from event to the phone
- PrivacySilhouette view in rooms, footage kept in the homeSee what care staff see
- IntegrationNurse call and care records by APIShow an alert on a ward handset
- SecurityScoped certificate, or a VA/PT within a yearCheck it covers the product
- ComfortNo added noise or light at nightCheck a room at night
- SupportTraining, repairs and an uptime targetAsk for the support plan
AI fall detection tender Singapore: what Ren Ci asked for
Ren Ci Hospital's 2025 notice RC25MM02 asked for a non-contact sensor system for its nursing homes. They are in Ang Mo Kio, Bukit Batok and Woodlands. A ceiling camera in privacy view is one way to answer a non-intrusive brief.
- 273 non-contact sensor units across 3 nursing homes
- Bed exits predicted or detected, with staff notified
- Low noise and light, with resident privacy preserved
- A live demo with at least 90% accuracy at interview
- Cyber security and data protection evidence, such as CSA Cyber Trustmark or ISO/IEC 27001
Time each step of an alert in the demo
A tender demo can time each step, from the event to the alert, the escalation and help arriving.
- 02:14Care staff on duty alerted
- 02:15Charge nurse, room still empty
- 02:16Help reached the resident, seen on camera
Define accuracy before the demo
An accuracy figure means little until the tender says what it counts. Agree the events, the camera views and what counts as a false alert before any demo.
Related: Fall detection camera, Bed exit alert, CCTV and privacy.
Events listed
Falls, sitting on the floor on purpose, bed exits and staff kneeling beside a resident.
Views agreed
Rooms at night in privacy view, and corridors and lounges by day.
Counted on site
Hits, misses and false alerts per night, on the home's own cameras.
Pilot on one floor
Four weeks, with measures agreed with the director of nursing first.
Questions about fall detection tenders
What should an AI fall detection tender include?
It should state the events to detect and an accuracy target with its definition. Add alert times, privacy rules, integrations, security evidence and support. Each item needs a test the home can run.
How do we test accuracy in a vendor demo?
List the events first, then stage them on the home's own cameras. Count hits, misses and false alerts. Test rooms at night as well as corridors by day, because lighting changes results.
Is an AI fall detection system a medical device?
That depends on its intended use, so check with HSA for the system in question. Havenwatch is a staff alert and workflow tool that makes no diagnosis. Care decisions stay with the nurses.
Can a camera count as non-contact or non-intrusive?
A ceiling-corner camera touches neither the bed nor the resident. It adds no noise or light at night. In privacy view it shows only a silhouette, so ask each vendor to show what care staff see.
How long should a pilot run before a full rollout?
Four weeks on one floor usually gives enough nights to count alerts, false alerts and response times. Agree the measures with the director of nursing before it starts.
Test the checklist on one floor before the tender.
Four weeks of real nights give counts of alerts, false alerts and response times.