Evidence
We will show results as we earn them. Internal test scores are not the same as a real home. We will not compete by printing a bigger percentage than other vendors.
Current stage: in development. First real-home pilot being prepared. No facility deployments. No published field sensitivity, false-alert rate, or uptime.
What can be said when
This hierarchy replaces mixing future goals with claimed outcomes.
Architecture facts — once implemented and technically verified
These are design facts, not clinical outcomes:
- AI runs locally
- no live camera feed
- no audio recorded
- normal imagery is not transmitted
- rolling buffer duration
- event-image policy
- camera-vendor cloud not required
- dedicated network / cellular architecture
- power-loss notification behavior
- heartbeat behavior
After controlled benchmarking — labeled “internal test”
- sensitivity on a named test set
- false-positive performance on a named test set
- model latency
- compute capacity
Always: dataset size, fall vs non-fall counts, whether rooms were seen in training, synthetic vs real, camera viewpoints, confidence intervals where they make sense.
If an internal held-out set looks extremely strong, we will still describe it as internal held-out test performance, not as real-world clinical performance.
After home pilots
- false alerts per home per month
- detection in real dwellings
- night / IR behavior
- camera coverage
- alert delivery time
- system uptime
After larger facility evidence
- time-to-assistance
- reduction in long lies
- staff workload
- transfers, injuries, or fall-rate changes
Those last items need real deployments. We will not claim them from a lab set.
What we will measure (and why)
| Metric | Why it matters |
|---|---|
| Fall sensitivity | Share of real or staged falls that are detected |
| Missed falls | Safety-critical failures, counted explicitly |
| False alerts per resident-month | What families and staff actually experience |
| Median alert latency | How quickly the event is surfaced |
| 95th-percentile latency | Worst normal-case delay |
| Slow-descent sensitivity | Hard, realistic fall category |
| Night / IR performance | Homes are dark |
| Occluded-fall performance | Furniture is unavoidable |
| Sitting / lying-on-floor false alerts | Common real-world confusion |
| Camera uptime | Coverage reliability |
| AI-node uptime | System reliability |
| External alert delivery | End-to-end, not just the model |
What we will not headline
- “100% accurate” or “zero falls missed,” even if an internal set is clean. Real homes are harder, and those slogans are already used elsewhere.
- Fall-rate reduction or injury reduction until Ilonai has its own evidence.
- “We prevent falls.” The product as designed is detection + verified escalation.
- Search-era leftovers: ROI percentages, invented pilot conversion rates, or certification dates that were never met.
Context, not fear
A prospective cohort study of people over 90 found that many falls happened while people were alone, and prolonged time on the floor was associated with serious injury, hospitalisation, and later long-term care (Fleming & Brayne, BMJ 2008). An observational study of real-time video fall detection in memory-care facilities looked specifically at time to assistance and time on the ground (PMC8277400).
Those papers motivate knowing that a fall happened, not a claim that Ilonai has reproduced their results.
Swiss figures: BFU, September 2026 — more than 1,700 fall deaths a year in Switzerland, 95% aged 65+.
Reliability tests we intend to run and then publish
Once a working home installation exists:
- cut the WAN / cellular path
- cut mains power
- unplug a camera
- reboot the local AI computer
- restore everything
A dedicated reliability write-up will appear here when those tests have been done — not before.