IoT Alarms to WhatsApp: Designing Alerts People Still Read After a Month
November 11, 2026 · 5 min read · Etzal Group Team
A cold storage operator connects temperature sensors to a WhatsApp group. In the first week the group receives 300 messages: every door opening, every defrost cycle, every brief spike. By the third week every member has muted it. In the fifth week a compressor fails at 02:00, the alert arrives exactly as designed, and nobody sees it until morning. Sending IoT alarms to WhatsApp is a ten-minute integration. Making them useful is a design job.
The rule that governs everything
An alarm is a message that requires a person to do something. If no action is expected, it is not an alarm. It may be a log entry, a line on a dashboard or an item in a daily summary, and it does not belong in someone's pocket at night.
Apply this test to every alert you plan: who should act, what should they do, and how soon? If there is no answer, do not send it.
Step 1: thresholds that reflect harm
Set limits where harm begins, not where normal ends. A chiller room normally at minus 20 does not need an alert at minus 19. Ask at what temperature and after how long the product is at risk.
Use two levels where it helps:
- Warning. Outside normal, worth a look in working hours.
- Critical. Harm is close, act now.
Step 2: delay and hysteresis
These two settings remove most nuisance alarms.
Delay. The condition must persist for a period before the alarm is raised. A door opening lifts the temperature for five minutes. A fifteen-minute delay ignores that and still catches a failed compressor.
Hysteresis. The alarm clears only when the value has returned well inside the limit, not the moment it crosses back. Without it, a value hovering at the threshold produces alarm, clear, alarm, clear, dozens of times.
Example: alarm when above minus 15 for 15 minutes, clear when below minus 17 for 5 minutes.
Step 3: suppress what is expected
- Planned states. Defrost cycles, maintenance windows and shutdowns should silence the related alarms automatically or by a switch with an expiry.
- Consequences. When the power fails, twenty sensors go out of range. Send one message, "Power lost at Site B", not twenty.
- Repeats. One message when the alarm starts, a reminder at a sensible interval if it continues, one when it clears.
Step 4: route to the right person
A single group for everything is the usual mistake.
- By site and system. The refrigeration technician receives refrigeration alarms for their sites.
- By severity. Warnings to a team channel in working hours. Criticals to the person on call.
- By time. An on-call roster decides who receives night alarms.
Fewer recipients per alarm means more ownership. When ten people receive a message, each assumes another will act.
Step 5: acknowledgement and escalation
A critical alarm needs a response.
- The alert goes to the person on call.
- They acknowledge, by a reply or a link.
- If there is no acknowledgement in ten minutes, it goes to the second person.
- If still none, an automated phone call and the supervisor.
Acknowledgement tells the system a person has it. Escalation covers the phone that is silent, flat or out of coverage.
Step 6: write the message for a person at 02:00
Compare two messages.
Poor: "ALARM TT-104 HI 2026-11-11T02:14:07 VAL -13.8 SP -15.0"
Better:
"CRITICAL: Cold room 2, Warehouse B Temperature minus 13.8, limit minus 15, for 15 minutes and rising Likely: compressor trip or door left open Do: check door, then compressor panel On call: Budi Reply OK to acknowledge"
A good message contains:
- Severity, as the first word.
- Where, in the name people use, not a tag number.
- What, with the value, the limit and the duration.
- Trend if known: rising, stable, falling.
- What to do first.
- How to acknowledge.
Keep it short enough to read on a lock screen.
Step 7: the daily summary
Everything that is not an alarm goes here. One message each morning:
- Alarms in the last 24 hours and how long each took to resolve.
- Warnings that did not escalate.
- Sensors offline or with low battery.
- Values drifting toward a limit.
This gives people the awareness they wanted from the noisy group, in one message they will read.
Sensor and connection health
A silent sensor looks the same as a healthy one. Monitor the monitoring:
- Alert when a device has not reported for longer than expected.
- Alert on low battery well before it fails.
- Alert when the gateway or connection is down.
Send these as warnings in working hours, unless the sensor guards something critical.
The monthly review
Thirty minutes with the alarm history:
- Top ten alarms by count. These few usually make up most of the volume. Fix the cause or change the setting.
- Alarms with no action taken. They should be warnings or summary items.
- Events that should have alarmed and did not.
- Time to acknowledge and time to resolve for criticals.
Alarm systems decay. Equipment changes, seasons shift, and settings that suited March are noisy in August. The review is what keeps the channel trusted.
About the WhatsApp channel itself
- Use the official business platform or a provider built on it for automated messages. Unofficial automation on a personal number risks the number being blocked, which is the worst time to lose your alarm channel.
- Messages a business initiates outside an active conversation generally use pre-approved templates. Design the alert wording with that in mind.
- Keep a second channel for criticals: SMS, a phone call or a siren.
We design monitoring and alerting as part of our IoT and Control Systems work, and the alarm philosophy is written before the first sensor is installed. If your team has muted its alarm group, talk to us.
Summary
Send an alarm only when someone must act. Set thresholds where harm begins, with delay and hysteresis. Suppress expected and consequential alarms. Route by site, severity and roster, require acknowledgement and escalate. Write messages a tired person can act on. Put everything else in a daily summary, watch sensor health, and review the alarm history every month.
Frequently asked questions
Why do people stop reading alarm messages?
Because most of them need no action. When a group receives dozens of messages a day about brief or minor deviations, people learn that an alert usually means nothing, and they mute it. The serious alarm then arrives unread. Reducing the count is the main design task.
How many alarms per day is reasonable for one person?
As a working guide, a handful that each need action. Process industry guidance on alarm management points to roughly one alarm every ten minutes as the most an operator can handle at a console, and a person receiving alerts on a phone while doing other work can handle far fewer. Aim for single digits per day per recipient.
Is WhatsApp reliable enough for critical alarms?
It is a good notification channel and should not be the only safeguard. Messages depend on mobile data, the phone and the platform. For anything affecting safety, keep local alarms such as sirens and interlocks in the control system, and use WhatsApp to inform people, with escalation by phone call if nobody acknowledges.
Related service
WhatsApp AI Chatbot
We build, train, and run a WhatsApp AI agent that answers customers 24/7, qualifies leads, and follows up.
