Your chatbot is still offering last month's promotion. How do you stop it without rebuilding the flow?

Most chatbot flows are built once and then left alone. The person who set it up added a line about the Raya discount or the 11.11 bundle at the moment it was current, right where customers would see it: in the greeting, or just before the bot asks for a name. It worked. Then the promotion ended, the team moved on, and the line stayed.

Nothing in a normal flow reminds anyone. The bot does not know a date has passed; it sends the same words on 1 December that it sent on 1 November. The staff rarely see it either, because they read the human side of conversations, not the automated opening. The first person to notice is usually a customer who asks for the discount, is told it is over, and feels misled.

Three habits make this worse:

• The offer lives inside a message that does other work. When the discount is one sentence in the greeting, removing it means editing the greeting, and nobody wants to touch the greeting on a Friday evening.
• Deleting a step feels risky. The step may have attachments, follow-ups or wording that took time to get right. Deleting it means rebuilding it next season, so people leave it "just for now".
• Nobody owns the date. Marketing knows when the promotion ends. The person with access to the chatbot is often someone else.

The scale of messaging makes each stale line expensive. Meta reported in May 2022 that one billion people message with a business each week on WhatsApp, Messenger and Instagram Direct, and in Malaysia 35.4 million people were online at the end of 2025, a 98.0 percent penetration rate. When a business answers on those channels with a bot, the bot's first message is the business's first impression, hundreds of times a day.

You do not need a second flow for every campaign, and you should never need to delete a step to make an offer stop. Four rules keep promotions under control in any chatbot tool:

1. Give every offer its own step. Keep the greeting about greeting. Put "Raya special: 20% off every facial until 30 November" in a separate message step right after it. Now the offer can be switched, dated or moved without touching anything else.
2. Attach the dates to the step, not to your calendar. Set the start and end on the step itself, in your local business time. The flow then decides on every message whether the step runs today. When the date passes, nothing has to be edited and nobody has to remember.
3. Prepare the next offer while the current one runs. Because a step can be switched off, you can build the November offer in October, leave it off, and switch it on the morning it starts. Or give it a start date and let it appear on its own.
4. Test the day before and the day after. A good test tool lets you pretend it is another date. Run the flow as a customer on the last evening of the offer and again the next morning. The offer should appear in the first test and be skipped in the second, with the rest of the conversation unchanged.

The principle is the same as a shop window: the poster comes down on the date printed on it, and the shop keeps trading. If your chatbot can only remove an offer by deleting it, the poster stays up until someone notices.

In Mampu AI, every message, question, knowledge lookup and AI task step in a Flow Agent has a Run this step switch at the top of its panel, and an Only run between period with a From and an Until date and time.

• Switch Run this step off and the flow skips the step: the customer goes from the step before straight to the step after. Nothing is deleted, and you can switch it back on later.
• Fill From, Until or both under Only run between and the step runs only inside those dates. Outside them it is skipped the same way. The times follow the company clock named under the fields, which is the inbox's own timezone when its business hours are on.
• On the canvas, a step that is not running is greyed out with an Off, Starts … or Ended … badge, and a step inside its dates shows Until …, so a promotion with an end date is visible at a glance.
• A customer who is already on the step when it ends finishes that step; nobody new enters it.
• In the Playground, a system line reads Skipped this turn with the step's name and the reason, so a missing message is never mistaken for a broken flow. Set a Scenario with a pretend time to test the day before and the day after.

The switch and the dates take effect on the next customer message. There is no republish and no second flow, and the AI designer leaves them alone when it edits the flow for you.