Alarms and remote monitoring
Alarm escalation by SMS and voice call
Updated
Alarm escalation means the alarm keeps going until a person takes responsibility for it: it notifies the first contact, waits a set time for an acknowledgement, then moves to the next contact and the next, repeating the list if nobody answers. SMS is fast and cheap but has no delivery guarantee, while a voice call can demand a keypress, so a robust chain uses both, from more than one provider.
What does an escalation chain look like?
A chain is a list of steps. Each step names who is notified, how, and how long to wait for an acknowledgement before the next step:
| Step | Who | How | Wait for acknowledgement |
|---|---|---|---|
| 1 | On-duty person this week | SMS and voice call | 5 minutes |
| 2 | On-duty person again | Voice call | 5 minutes |
| 3 | Backup person | SMS and voice call | 5 minutes |
| 4 | Site manager | Voice call | 5 minutes |
| 5 | Everyone on the list | SMS and voice call | Repeat from step 1 until acknowledged |
The wait times are an example. Set them from how fast your process fails: in a dense recirculating fish system, oxygen can reach stressful levels in under 6 minutes after a power failure (SRAC 452), so the whole chain must reach a second person well inside that window. See Preventing a 3 a.m. fish kill.
Real sites follow the same shape. The Freshwater Institute’s RAS alarms call about six mobile numbers; one person holds alarm duty on a weekly rotation, and follow-up calls confirm that person is responding and whether more help is needed (Hatchery International).
What counts as an acknowledgement?
An acknowledgement is a deliberate action by a person that says “I have this”. It stops the escalation, and it is logged with who and when. It is not the same as the alarm clearing.
Valid acknowledgements:
- A keypress during a voice call, after the message has played.
- A reply to an SMS with a code, or a tap in an app, tied to the specific alarm.
- A call back into the system.
Not acknowledgements:
- A call that was answered. Voicemail answers calls too. Classic autodialers solve this by requiring a key: the RACO Verbatim takes acknowledgement “by TouchTone key or by calling back” (RACO).
- An SMS marked delivered. Delivery to a phone in a pocket at 3 a.m. is not a person.
- The condition returning to normal. A level that recovers for a minute may fail again; decide per alarm whether clearing stops the chain, and log it either way.
Two more rules prevent a common failure. First, if the acknowledged condition is still active after a set time (for example 30 minutes), re-alarm, because the person who acknowledged may have fallen back asleep or been unable to fix it. Second, escalate a new, different alarm even while an earlier one is acknowledged.
Retries: how many, how fast?
Classic dial-out boxes retry by design. Ebeling describes a Sensaphone-type unit that dials up to four numbers with a spoken message and keeps dialing, up to 16 attempts, until a staff member acknowledges by phone or code (Global Seafood Alliance). The RACO Verbatim dials a list of up to 16 numbers, “calling until it gets an answer” (RACO).
Guidelines that hold for any system:
- Never stop silently. If the list is exhausted, start again, and raise a local alarm (horn, strobe) on site.
- Space voice retries a minute or two apart so a person reaching for the phone is not cut off by the next call.
- Do not retry SMS rapidly. Repeated identical texts can look like spam to carrier filters; escalate to voice instead.
- Make the message say what and where. The Freshwater Institute’s voice alarms name the alarm type and, in most cases, the location (Hatchery International).
What limits SMS delivery?
SMS feels instant, but it is a best-effort service between carriers, and business messaging in the US is filtered.
- Registration. In the US, application-to-person texts from ordinary 10-digit numbers go through A2P 10DLC registration. Twilio states that registering “results in lower message filtering and higher messaging throughput”, and that unregistered traffic carries extra carrier fees (Twilio). An alarm service that has not registered its brand and campaign risks having texts filtered.
- Throughput. Each number type has a sending rate. An outage that triggers alarms at 50 sites at once can queue messages behind each other.
- No delivery promise. A text can be delayed or dropped between networks, and a delivered text can sit unread on a silenced phone.
- Do Not Disturb. Phones in silent or focus mode suppress texts unless the sender is allowed through. Ask each person on the list to add the alarm number as an emergency contact or exception, and test it.
Voice calls are slower and cost more per alarm, but they ring audibly, and a keypress proves a person heard the message. That is why a chain should not rely on SMS alone.
Redundancy: what happens when a provider fails?
Every link in the chain can fail: the site’s connection, the alarm service, the SMS or voice provider, and the recipient’s mobile network.
- Two ways out of the site. If the controller reports to a cloud service, give it a second path (for example cellular plus a wired line). The Freshwater Institute added an analog backup phone line after its one significant fish loss, which was traced to an alarm failure (Hatchery International).
- Two providers for delivery. A cloud alarm service should be able to fail over from one SMS and voice provider to another. Ask any vendor what happens if their messaging provider has an outage.
- Heartbeat in the other direction. The cloud should alarm when a site stops reporting, because a site that has lost power and connection cannot send its own alarm.
- Battery for the alarm path. Each box in the path from sensor to modem needs backup power. Product examples as of 19 November 2026: the Sensaphone 400 and 800 landline dialers list 24 hours of battery backup (Sensaphone); the cellular Sentinel lists 8 hours for the unit and modem (Sensaphone). See Power-failure alarms that still work when power is out.
- Contacts on different carriers, where possible, so one mobile network outage does not reach everyone.
On-call rotation and the human side
The chain only works if the people on it expect calls. Keep the list short and current, rotate duty weekly with a clear handover, and put the backup person on a different schedule from the primary. Ebeling warns that overly complex alarm systems and frequent false alarms lead part-time staff to disable them (Global Seafood Alliance): every nuisance alarm makes the real one less likely to be answered.
Testing the chain
Test the whole path, not the box. Ebeling describes manually triggering the alarm to test the autodialer and measure staff response time (Global Seafood Alliance). Once a month:
- Trigger a real input (not a software test button) at an unannounced time.
- Let the first contact ignore it, and confirm the chain moves to the second.
- Record the time from trigger to acknowledgement.
- Check the log shows who acknowledged, how and when.
Where halfnine fits
halfnine plans a controller and a cloud alarm service with escalation. Its retry policy, providers and acknowledgement methods are not yet fixed: To be decided: escalation features, SMS and voice providers, failover design. The checks on this page apply to any system you evaluate.
Sources
- SRAC Publication No. 452, Recirculating Aquaculture Tank Production Systems: Management of Recirculating Systems (Masser, Rakocy, Losordo, March 1999 revision), accessed 2026-11-19
- Emergency response for RAS, Hatchery International (John Davidson, 25 April 2022), accessed 2026-11-19
- Process control, monitoring options in intensive recirculation systems, Global Seafood Alliance (James M. Ebeling, 1 June 2003), accessed 2026-11-19
- RACO Verbatim Autodialer product page, accessed 2026-11-19
- Sensaphone 400 and 800 Monitoring Systems product page, accessed 2026-11-19
- Sensaphone Sentinel Monitoring System with Cellular Modem product page, accessed 2026-11-19
- What is A2P 10DLC?, Twilio documentation, accessed 2026-11-19