Follow-up that is not spam: the rules behind an AI agent's follow-up message

Twelve messages a year to a list of 2,000 contacts is 24,000 messages. Suppose each of those blasts makes only 2% of recipients block the number or stop reading. After a year you are left with 0.98 to the power of 12, roughly 78% of the list: about 430 people you paid to acquire who will never hear from you again. Against that stands a single message sent 48 hours after someone asked about a price and stopped replying, sent only to people who actually asked, and cancelled if they replied in the meantime. The first is called "marketing" and feels like activity. The second is called "follow-up", and most business owners are afraid of it because they picture the first. This article is about the rules that separate the two, and about what happens when you let an AI agent send follow-up messages.

The difference between a blast and a follow-up is not the text

The exact same sentence, "wanted to check whether this is still relevant", can be good service or it can be spam. What decides is three questions: who it goes to, when, and why. A blast goes to everyone, on a date that suits the business, because a month has passed. A follow-up goes to one person, at a moment that follows from their behaviour, because there is an open conversation that has not ended.

When a customer complains about being "pestered", it is almost always a failure in one of those three questions. A message that arrived after they had already booked. A second message the same day from a different rule. A message about a service they had already bought. None of these is a wording problem, so rewriting the text does not fix them. What fixes them is a mechanism that can answer all three questions before every send.

Two triggers: silence or a schedule

Every follow-up rule in WhaleBiz starts with a decision about what fires it. There are two options, and they meet completely different needs.

A silence trigger. The rule fires when the conversation has stalled: the agent asked a question or sent an offer, and the customer has not replied for a period you set. This is the classic case of a lead who asked for a price and vanished, or someone who started booking and never picked a time. The clock runs from the last message in the conversation, so if the customer writes anything, it resets.

A schedule trigger. The rule fires on a date or at a fixed interval from an event: a day before an appointment, a week after a treatment, six months after a purchase. There is no stuck conversation here, just a point in time when a message is simply more useful than at any other moment. Reviving a dormant customer database is one special case of this trigger, but most of its use is far more everyday.

The second decision gets less thought: is the rule aimed at a conversation or at a customer. A rule aimed at a conversation lives and dies with that specific conversation, and disappears when it closes. A rule aimed at a customer follows the person even if they have opened several different conversations over time. An appointment reminder belongs to the customer. "Still thinking about the offer?" belongs to the conversation.

Who it goes to: the filter that runs before every message

A trigger says when. A filter says who. Without a filter, a "three days of silence" rule will also catch the customer who closed the deal on the phone and never bothered to reply on WhatsApp, and the person who wrote "not relevant, thanks".

The filter is built from fields that already exist on the customer card in the CRM: status, lead source, the service they asked about, tags and owner. A typical rule looks roughly like this: two days of silence, only enquiries from the last week, only with status "offer sent", not anyone marked "not interested". Every such condition removes people from the list, and that is exactly its job. A good rule sends to very few people.

This is also where the agent itself comes in. While it runs the conversation it fills in those fields: it marks the service, updates the status, writes a summary. So the filter works on current information rather than on whatever someone remembered to type. A customer who wrote "I already booked with your other branch" gets a status that removes them from the rule, without anyone touching the card.

Three brakes: gap, cap and cancellation

These three mechanisms are the difference between a system you can leave alone and a system you have to babysit.

• A minimum gap between touches. Even if two different rules match the same customer on the same day, they will not get two messages. A minimum time is set between one outbound message and the next, and the second rule waits or gives up.
• A cap per person. A maximum number of follow-ups one customer can ever receive from the same rule. Once they reach it, the rule simply stops touching them, even if they still meet every condition. This is the brake that protects you from the most expensive mistake: a rule with an over-broad condition that runs for months.
• Cancellation when the customer replies. Any reply from the customer cancels the messages waiting for them under that rule. The check runs at the moment of sending, so someone who wrote a minute ago will not get "just checking".

The third is the most important of all, and it is also the one email and blast tools lack. A blast tool sends from a list saved yesterday. A system that sits on the conversations themselves knows what happened a minute ago.

The maths: one well-timed touch vs twelve untimed ones

To see why these rules are worth money, compare two approaches on the same list. The figures in the table are an arithmetic example, not a measured result, and you should replace the assumptions with your own numbers.

The last row matters most on WhatsApp. Meta rates the quality of a business number by how recipients react, and a number that piles up blocks and reports can be limited in how many conversations it may open. A broad blast damages not only the list but also the channel you use to answer new customers.

On the other side, one well-timed touch pays off very quickly. If your average deal is ₪800, and out of a hundred leads who stopped replying after getting a price only five come back and close thanks to one message, that is ₪4,000 a month that was not there, without any of the other ninety-five receiving more than one message. Handling cold leads starts exactly there.

WhatsApp, the 24-hour window and the spam law

Two external frameworks decide what you may send, and both favour rules over blasts.

The first is WhatsApp's. Within 24 hours of the customer's last message the business can write to them freely. Once the window closes, a conversation can only be reopened with a template approved in advance by Meta. So a follow-up after a few hours of silence goes out as free text that refers to the conversation, and a follow-up after three days goes out as a template. The template wording is written during setup, and Meta's approval time is outside our control.

The second is Section 30A of the Israeli Communications Law, known as the spam law. It deals with advertising messages sent without prior consent, and allows compensation of up to ₪1,000 per message without proof of damage. A reply to an enquiry the customer opened is substantively different from a promotions blast to the whole list, but exactly where the line falls is a question for a lawyer, not a blog. What is clear: the narrower your rules, the more of your messages go to people who asked you for something.

What the message itself says

Good rules do not save a bad message. Three principles work in almost every business. First: the message refers to the specific question, not to the business. "You asked about a Tuesday appointment, there are two afternoon slots left" works. "Hi, how are you? We are here for you" does not. Second: one question that is easy to answer in a word. Third: an easy way out. "If this is no longer relevant, just tell me and I will close the enquiry" makes people reply even when the answer is no, and a no is worth far more than silence, because it cleans the CRM and removes the person from every future rule.

When the message goes out inside the 24-hour window, the agent writes it in the context of the conversation, in the language the customer wrote in, and with the correct grammatical gender in Hebrew or Russian. When it goes out as a template, the template is written to sound like a continuation, not like an ad.

Where this lives in WhaleBiz and which plan it is on

The follow-up rules sit inside the CRM of the WhaleBiz platform, next to the conversations themselves, and that is why cancellation on reply works: the system that sends is the same system that sees the reply. You configure the rules yourself in the interface, and what you configure yourself does not count as a change. There is no charge per rule and no counter on the number of runs, so there is no reason to economise by preferring one broad rule over several narrow ones.

The channel is what decides the plan. WhatsApp follow-up starts on Standard, because Start includes only a website widget or Telegram. Follow-up on Instagram and Facebook, and several agents sharing the rules, are Business. You will find the full comparison in the plan table, and how the agent takes a lead from the first question to an appointment or a deal is described on the AI sales agent page.

The bottom line: your customers do not hate follow-up messages. They hate messages unrelated to what happened between you. A system that knows what happened a minute ago, limits itself and stops when it gets a reply sends fewer messages than any blast tool, and closes more.

Frequently Asked Questions

What happens if the customer replies just as a scheduled follow-up is about to go out?

Their reply cancels every follow-up waiting for them under that rule. The check runs at the moment of sending, not when the rule was created, so a customer who wrote a minute ago will not get "just checking whether you saw our message". From there the conversation continues as usual, with the agent or with a person.

How many follow-ups should a lead who went quiet receive?

There is no single number that is right for every business, which is why it is a setting and not a fixed default. What matters is that there is a number: a cap no customer will ever exceed, however many rules are active at the same time. In most businesses we set up, the first sequence is short, and only after a month of data do we decide whether to extend it.

Does a WhatsApp follow-up need an approved template?

Within 24 hours of the customer's last message you can write to them freely. Once that window closes, WhatsApp only lets a business open a conversation with a template approved in advance by Meta. The template wording is written during setup, and the time Meta takes to approve it is outside our control.

Are follow-up messages banned under the Israeli spam law?

Section 30A of the Communications Law deals with advertising messages sent without prior consent, and allows compensation of up to ₪1,000 per message without proof of damage. A reply to an enquiry the customer started is different from a marketing blast to a list, but where the line falls between the two is a legal question. If you plan promotional messages to people who have not contacted you recently, have a lawyer check the wording.

Which plan is this available on?

The follow-up rules themselves are not metered: there is no charge per rule and no counter on the number of runs. What decides is the channel. WhatsApp follow-up starts on the Standard plan, because Start includes only a website widget or Telegram. The full details are in the plan comparison on the pricing page.