The term is used loosely enough to mean "marketing, but on WhatsApp". That definition excludes nothing and is therefore useless.
A definition that excludes something#
Conversational marketing is marketing where the customer's reply changes what happens next.
If a reply does not alter the sequence — if the same three messages go out on the same schedule whether or not anyone responded — that is broadcast marketing delivered over a messaging channel. There is nothing wrong with it. It is just not conversational, and calling it that will lead you to measure it wrong.
What follows from the definition#
A reply path is mandatory. Sending from a number that cannot receive disqualifies the programme immediately. This sounds obvious and is violated constantly.
Somebody has to be there. Either a person or an automated agent must handle what comes back, within a time the customer finds reasonable. A conversation you do not answer is worse than a broadcast you never sent, because you asked.
Branching is the product. If the journey has no branches, the "conversation" is decoration.
Why it works when it works#
Three real mechanisms, none of them mysterious:
Same three messages, same schedule, whether or not anyone replied. Nothing wrong with it — just measure it as a campaign.
A reply changes what happens next. Branching is the product, not a feature of it.
A conversation you invited and cannot staff. Worse than a broadcast, because you asked.
An agent that traps the customer in a menu. Worse than no reply, because it consumed their attempt.
- The channel is one people read. Personal messaging apps have attention that email lost.
- Replying is cheaper than clicking through. The threshold for "yes, tell me more" collapses when it costs one word instead of a page load and a form.
- Permission is renewed by participation. Someone who replied has just refreshed your basis for contacting them, on almost every channel and under most consent frameworks.
Where it goes wrong#
Volume. The channel's advantage is that it feels personal. Send at broadcast frequency and you spend that advantage permanently — and unlike email, people block.
Answering with a bot that cannot answer. An automated agent that handles the top intents well is genuinely good. One that traps a customer in a menu is worse than no reply, because it consumed their attempt.
Measuring it as a campaign. Opens and clicks miss the point entirely. Measure reply rate, resolution, and outcome per thousand sent.
And one obligation#
From 2 August 2026 the EU AI Act requires that people be told when they are interacting with an AI system, clearly and at first interaction. If the automated half of your conversational programme talks to EU customers, the disclosure is not optional and cannot be buried in terms.
The operating model matters more than the tooling#
The failure mode is almost never the software. It is starting a conversation the organisation cannot finish.
Before launching anything, answer three staffing questions. Who answers, and during which hours? What is the promised response time, and is it written where the customer can see it? What happens outside those hours — an honest "we will reply tomorrow morning" is fine; silence is not.
A conversational programme with no answer to these is a broadcast programme that has accidentally given people a way to reply.
Frequency is the whole discipline#
The channel's advantage is that it feels personal. That advantage is spent, not renewed, and the exchange rate is brutal: unlike email, where over-sending erodes engagement slowly, over-sending on a messaging channel gets you blocked, and a block is permanent and invisible.
Practical guardrails: a global frequency ceiling above all journeys, a minimum gap between promotional messages to the same person, and a rule that a service message never carries a promotion. That last one is worth being absolute about — mixing them is how a channel people trust becomes a channel people mute.
Measure the conversation, not the campaign#
Campaign metrics do not describe this. The ones that do:
- Reply rate, the only number that tells you it felt like a message.
- Resolution rate for anything that started as a question.
- Time to first human response, because that is what people remember.
- Opt-out per send, as a trend. The leading indicator of over-use.
- Outcome per thousand sent, against a holdout.
What to take away#
- If a reply changes nothing that follows, it is a broadcast on a messaging channel.
- Answer the staffing questions before the tooling questions.
- Frequency is the discipline; the channel punishes over-use permanently.
- Disclose automation at first interaction — required in the EU since 2 August 2026, and worth doing everywhere.
Sources
Every claim worth checking, with somewhere to check it.