Your WhatsApp Bot Can Die While Your Phone Rings
The clearest number I have from the past three weeks is 10 minutes.
That is how long it took to bring a business’s WhatsApp automation back to life, once somebody finally told the owner it was down. The line had been dead since at least 23:05 the previous night. The notification went out at 08:00. By the next check, 10 minutes later, it was working again.
Almost nine hours of silence. Ten minutes of work. The distance between those two numbers is, I think, the least examined thing in the way Israeli small businesses buy automation — and it appears in almost no contract I have ever seen.
Your phone is not proof that your bot is alive
Ask an Israeli business owner whether their WhatsApp line is working and they will glance at their phone. Messages are arriving. Case closed.
It isn’t, because those are two different connections.
Most automation sold to Israeli SMBs — appointment reminders, order intake, after-hours answering — does not run on the owner’s handset. It runs on a gateway attached to the business number as a companion device, the same mechanism as WhatsApp Web, or on Meta’s official Business Platform. The handset and the gateway fail independently, and usually do. The owner’s phone keeps buzzing with messages people send directly. The automated half — the part that answers at 23:40, the part they are paying for — can be gone for days without changing a single thing the owner can see.
WhatsApp is quite open about the fragility. Its own Help Center page on linked devices states: “Linked devices work without your phone online, but will log out if your phone is unused for over 14 days,” and adds that WhatsApp will “automatically disconnect linked devices after 30 days of inactivity.” Change the phone, restore from a backup, spend a fortnight abroad on the other SIM, reinstall the app after an update — the business number carries on perfectly while the automation quietly detaches.
Nobody gets an SMS about this. No invoice line stops. There is no error. There is just a service that used to answer and now doesn’t.
What three weeks of actually looking turned up
Since the evening of 8 September I have been checking six Israeli business WhatsApp lines every 10 minutes and writing down what I find. As of this morning that is 1,953 sweeps and 11,728 individual line checks across 18 days. All six lines are connected as I write this.
In those 18 days there were five disconnection events, spread across four of the six lines. That count surprised me, and its shape surprised me more.
Three of the five cleared by themselves inside a single 10-minute window. The line dropped, the gateway reconnected, nothing needed doing, and no human would ever have known. One was a line that was already dead when the watching started — more on that shortly. And exactly one became the almost-nine-hour outage I opened with.
This is the part worth sitting with. A disconnection that heals itself in 10 minutes looks, from the outside, exactly like one that lasts nine hours. Both look like nothing. Both look like a quiet Monday. The failure makes no sound, so the only thing determining whether it costs you money is whether anything is watching.
The four days before anything was watching
I did not build the monitoring because I am wise. I built it out of embarrassment.
Our records put the failure at 20:47 on Friday, 4 September. The line stayed dead through Shabbat, through Sunday, through Monday, and was found on Tuesday evening — four days almost to the hour. For four days a business was paying for an automation that was not there, and it surfaced only because I happened to open the gateway for an unrelated reason.
What made that possible was monitoring that was technically correct and practically useless. It confirmed that the server was running and the service was up. Both were true the entire time. Neither had anything to do with whether that business’s line was connected. “The system is up” and “your line is answering” are different claims, and the first is the one vendors tend to measure, because it is the easy one.
Israel’s WhatsApp automation market is largely boutique studios — Achiya Automation among them — building custom bots on top of a small handful of gateways. This is not a story about one careless supplier. The same blind spot is sitting in a great many deployments right now, quite possibly including yours.
Nine hours, by design
Back to the long outage, because it contains a decision worth arguing about.
The monitor caught the line at 23:05 on a Sunday night. It then wrote, every 10 minutes for nearly nine hours, a line that reads: quiet hours, customer message deferred to next run. At 08:00 on Monday the notification finally went out. Ten minutes later the line was back.
That delay was deliberate. I set the notification window to 08:00–21:00, with silence from Friday afternoon through Saturday night, because I did not want to be the vendor who wakes a business owner at two in the morning over something they will fix with their coffee. I still think that is right for most businesses. I am much less sure it is right for all of them — a pharmacy and a wedding venue do not have the same Saturday night — and I am certain the owner should be choosing the window rather than discovering it afterwards.
There is a second honest caveat in that timeline. I know when the line was seen to be down. I do not know when it actually died. A monitoring log tells you when somebody looked, not when the thing broke. Any vendor who quotes you a precise outage duration without a continuous record is telling you a story.
What to ask before you sign
Four questions get a buyer most of the way:
What exactly do you check, and how often? “We monitor the server” is not an answer. The answer you want names the connection state of your line and gives an interval in minutes.
Who finds out, how, and within what window? Ask whether the alert reaches you or only the vendor, on which channel, and whether there are quiet hours. If there are, ask when they are and say whether they suit your week.
How do I get back online without waiting for you? The best answer is that you can reconnect the line yourself from your phone. A line that can only be revived by a developer who might be on a flight is a line with a person-shaped single point of failure.
Show me last month’s log. Not a dashboard, not a promise — the actual record of checks and events. If it does not exist, nothing was being watched.
And a test you can run yourself in 30 seconds, needing nothing from anybody: borrow a phone that has never messaged your business, send the trigger word, and time the reply. Your own phone cannot perform this test, because your own phone sits inside the system you are trying to check. Friday morning, once a week, before the weekend closes in — it is the cheapest habit an owner can build.
The uncomfortable summary of 18 days of data is that lines drop far more often than anyone assumes, that most drops are harmless, and that the harmless ones and the expensive ones look identical until someone looks. Ten minutes of work. Nine hours of not knowing. That ratio is a choice, and most Israeli business owners are currently making it without realising there was one.
So: if your WhatsApp line went dead at 20:47 this Friday, what — specifically — would tell you before Sunday morning, and whose phone would it ring on?
Achiya Cohen builds custom WhatsApp and phone automation for Israeli businesses at Achiya Automation.
