Skip to main content

Overview

Every AI-powered customer interaction has edges — moments where the AI doesn’t have the answer, the customer’s request falls outside configured capability, or the situation requires a human. Fallback flows define what happens at those edges. Without configured fallback flows, edge cases are handled inconsistently. With them, every out-of-scope situation has a defined, controlled response — escalation to an agent, a redirect, a specific message, or a background notification to your team.

How it works

Fallback flows trigger when the AI determines it cannot handle a request within its configured knowledge and actions. When a fallback triggers, Parlel evaluates the configured priority stack and executes the highest-priority applicable flow.

Trigger conditions

  • Knowledge retrieval confidence below threshold
  • Customer request outside the scope of any configured action
  • Journey stage reached without a valid path forward
  • Explicit customer request for human assistance

Response types

Background actions

In addition to the customer-facing response, fallback flows can trigger background actions that run silently:

Priority stack

You can configure multiple fallback flows. Parlel evaluates them in priority order and executes the first one that matches the trigger condition. This lets you handle different out-of-scope situations differently.

Relationships

Knowledge Base

Low confidence in knowledge retrieval is a primary fallback trigger. Well-structured knowledge reduces fallback frequency.

Journeys

Journey configuration defines what’s in scope at each stage. Requests outside that scope trigger fallbacks.

Transactions

Every fallback trigger is logged as a transaction — giving visibility into where and why fallbacks fire.

Integrations

Connect agent and webhook fallback types require a configured integration.

Configuration

Creating a fallback flow

Go to Fallback Flows in the portal → New fallback flow. Define:
  • Trigger condition — when this flow activates
  • Priority — where this flow sits in the priority stack
  • Customer response — connect agent, redirect, or message
  • Background actions — optional email, webhook, or action

Monitoring fallbacks

Go to Transactions in the portal and filter by fallback events. High fallback frequency on a specific topic signals that you should add content to your knowledge base or configure a new action.

Reference