resources / security and engineering

human "echo" detection: how a bot knows when to stay quiet.

heysteff is an AI platform for customer support and sales across WhatsApp, Instagram, Messenger, Gmail and Shopify; Steff is the AI agent that runs it. One of the most awkward moments of any badly implemented bot is this: a human agent on the team replies to a customer from the native Instagram app, and seconds later the bot sends another answer on top — contradictory or redundant. heysteff is designed so that never happens.

◆ key idea

When a human replies from outside heysteff, the system detects it and pauses the bot in that conversation instantly — without anyone having to press a button.

The real problem: two answers to the same message

Many businesses don't handle support exclusively from a unified platform like heysteff: sometimes someone on the team — the owner, a salesperson, support — replies directly from the Instagram or Messenger app on their phone, without going through the dashboard. That's normal, especially in small businesses where several people touch the same channel.

The risk appears when the bot doesn't find out a human already answered: the customer gets two different messages, sometimes with contradictory information, and the conversation looks uncoordinated — exactly the opposite of what you'd expect from well-implemented AI support. It's the kind of mistake that makes a customer stop trusting the entire channel, not just the bot.

How echo detection works

heysteff monitors each conversation's outgoing messages, including the ones that originate outside the dashboard — for example, a reply sent from the native Instagram app by a team member. When it detects that "echo" of a human reply that didn't come from the platform, it reads it as a clear signal: someone is already handling this conversation in person. The automatic reaction is to pause the bot in that specific conversation, so it doesn't answer again on top of what the person already said.

There's nothing to configure for this to happen — it's the default behavior. The team can keep working from wherever is most comfortable, without manually coordinating with anyone or worrying about "letting the system know".

Beyond echo: 1-click pause, at four levels

Echo detection is automatic, but explicit manual control also exists when you need it. Anyone on the team can pause the bot in one click at four different levels depending on the situation:

  • Per conversation: pause only that specific chat with a customer, without touching the rest.
  • Per channel: pause the bot across all of Instagram, for example, while something specific to that channel gets sorted out.
  • Per workspace: pause the bot across the whole business — useful during a delicate campaign or a one-off issue that needs full human review.
  • Global: account-wide, for exceptional cases.

This granularity is central to heysteff's copilot mode: AI isn't all-or-nothing, it's a capability the human team controls with precision — conversation by conversation if needed.

Escalation with context, not from scratch

When the bot escalates a conversation to a human — because of business rules, keywords, or because it detects the customer is frustrated — it doesn't do it silently. It hands over the full context of everything discussed up to that point, so the person taking over doesn't have to ask the customer to repeat everything from the beginning. That smooth handoff is, together with echo detection, part of what lets a business trust the bot to keep chatting without constant supervision.

Why this is a trust requirement, not a technical detail

No business adopts a customer service bot without fearing it will "put its foot in it" in front of a real customer. Echo detection and the different pause levels aren't just one more feature on a list: they're the minimum condition for a team to trust that the bot and the people can coexist in the same channel without stepping on each other. Without that, any customer service automation ends up creating more friction than it resolves.

Related

◆ next step

See in a demo how the bot pauses itself the moment your team takes over a conversation.

Get a demo See WhatsApp AI