Skip to content
AskElephant Knowledge Base home
AskElephant Knowledge Base home

Conditional Error

Introduction

If a conditional node is blocking your workflow unexpectedly, or taking a branch you didn't intend, the cause is usually one of three things.

1. Missing Variable References in the Conditional Prompt

Conditional nodes evaluate a prompt against upstream node output. If the conditional prompt doesn't reference any variables, it has nothing to evaluate.

Fix:

  1. Open the conditional node.

  2. Confirm the prompt text references at least one upstream variable (for example, {{meeting_classification}}).

  3. If you want the conditional to depend on an AI classification, add that classification node before the conditional and reference its output.

2. Reversed Logic

Double-check that your conditions map to the outcomes you actually want. A common mistake is writing "No second call = send success message" when it should be the opposite.

Say the rule out loud: "If X is true, the workflow should do Y." If that sentence doesn't match what your conditional is configured to do, flip the logic.

3. Ambiguous Criteria

If the conditional fires but doesn't route the way you expect, the AI prompt probably needs more specific keywords.

  • Review the prompt text in your conditional step.

  • Identify keywords that show up in your test data but aren't in the prompt.

  • Add those keywords under your Yes and No sections so the model knows what to look for.

  • Clarify ambiguous definitions — for example, clearly distinguish "first sales call" from "first CS call" if your workflow depends on the difference.

When in doubt, ambiguous cases default to blocking the workflow. Err on the side of more specific prompts.

Check for a Duplicate Workflow

If the conditional looks correct but the workflow still behaves strangely, check whether a second workflow is running against the same trigger and sending a conflicting output. See Duplicate workflow outputs.

Still Stuck?

If the conditional logic looks right but the branching is wrong, share the workflow with support along with a specific run ID from Workflow History. See Workflows for the full troubleshooting guide.