DEV Community

Cover image for From SLA clauses to alerts: watching a WhatsApp group portfolio during the hours you sold
9bot
9bot

Posted on Edited on Originally published at 9bot.com.br

From SLA clauses to alerts: watching a WhatsApp group portfolio during the hours you sold

An agency or an internal operations team that promises to run WhatsApp groups already has an SLA, even if it is only a paragraph in a proposal. Coverage hours, how fast a person must pick up a question the bot cannot resolve, how fast banned content must come down, who approves a campaign: those sentences are the service. They become real only when someone is told that a clause was missed, while there is still time to repair it.

This article turns those clauses into a short alert list. It is not a vendor scoreboard. It is a way to notice a broken room during the window you sold, rather than in a monthly meeting. The setup work that sits behind the alerts is described in the 9bot business automation guide. 9bot is sold as a paid subscription; you get seven days to try it. The company has no official tie to Meta or to WhatsApp.

Separate what the bot covers from what the contract covers

Automation can approve joins with filters and a daily cap, post a standard welcome, apply a link filter and banned words with a warning ladder, answer repeated questions through commands such as .menu, send recurring notices across many rooms, and export a PDF of activity. That is the rule layer. It runs whether or not an analyst is staring at the phone.

The contract layer is everything a person still owes: hours during which a human will pick up an exception, the maximum time to remove content the filter did not catch, how quickly a person must handle a question no command can close, and the name of the person who signs off on a campaign. If you write "the bot handles it" in those rows, you have not sold a service. You have sold a hope.

Four alerts that map to clauses

Coverage gap. A message that needs a person arrives outside the stated hours, or sits inside the hours past the reply target. This alert cannot come from 9bot alone, because 9bot is not your ticket queue. Pair the group with the queue you already use, and page the on-call operator only inside the window you sold.

Filter failure or filter off. Unwanted links or repeated spam visible in a room, or a health check that finds the filter disabled in one client group. Treat a disabled filter as a production incident, even if no bad link has appeared yet. The next one will choose that room.

Schedule miss. A notice that should have gone out in the last cycle did not, or went out in only some of the rooms it was meant for. Recurring schedules across a portfolio exist so that a campaign is not "whoever remembered." A miss is a broken delivery, not a soft delay.

Report late. The agreed monthly pack (who was active, who stayed silent seven days, arrivals, departures, net movement, and the busiest hours) did not leave as a PDF on the agreed date. For an agency, that pack is part of the product the client bought.

Optional fifth: outbound command errors. If a .menu alternative or an order-status command calls a CRM or ERP, a week of failed outbound requests is an integration incident. 9bot can emit those requests as outbound HTTP (GET, POST or PUT) to a system you authenticate. Watch the result, not only the command text.

Your marketing team shouldn't be posting one message at a time

How to wire the alerts without inventing a platform

You do not need a new monitoring stack. You need a named owner, a threshold, and a place the alert lands (a ticket, an internal room, an email). A weekly pass can confirm that each client room still has two company admins, that the expected template is in force, and that last cycle's schedules fired. The monthly PDF answers the reporting clause. The queue answers the human-reply clause.

Write the threshold in the same words as the contract. "Banned content down in 30 minutes during coverage hours" is an alert. "We stay on top of things" is not.

What not to alert on

Do not page anyone because a group was quiet. Inactivity for seven days is a metric for the client report and for a re-engagement decision, not an incident. Do not page because a single join was refused by a filter; that is the filter doing the job. Do not page the creator of the client company about a schedule miss; page the operator who owns that client.

Agencies that run several clients in one 9bot panel should scope alerts by client, using per-user access so that an analyst is not drowning in another team's rooms. Shared passwords make every alert worse, because you cannot tell who changed the template.

A 15-minute weekly drill

Pick one client. Confirm the four clauses still have owners. Fire a test look at the last schedule cycle and at the last PDF. Open one room and confirm the filter and the warning ladder are on. If anything fails, write a one-line cause and a repair, then decide whether the same failure could be sitting in the other clients. The SLA is a promise about time. Alerts are how that promise is watched.

Top comments (0)