Skip to main content

Failed Queries

When Qluent can't answer a question, it creates a failed query that your team can review and learn from. Your team can reply directly in the thread (human-in-the-loop) or use the failure to improve your configuration for next time.

Why it matters

Failed queries reveal gaps:

What triggers a failed query

TypeDescription
No DataSQL ran but returned zero rows
Execution ErrorDatabase returned an error
Generation ErrorCouldn't generate valid SQL

Managing failed queries

Each failed query shows:

  • Status - unconfirmed, unresolved, resolved, ignored
  • Priority - 1-5 (1 = highest)
  • Original question - What the user asked
  • Generated SQL - What Qluent tried to run
  • Channel - Where it came from (web, Slack, Teams)

Resolving failed queries

Fix the configuration

Look at what went wrong and choose the right fix:

ProblemFix
Qluent used the wrong table/columnImprove descriptions to clarify what each contains
Qluent didn't understand terminologyAdd context to instructions
Calculation was wrongCreate a metric with the correct formula
Query pattern was wrongAdd a validated query showing the pattern
Value wasn't matchedEnable semantic search on the column

Respond to the user (human-in-the-loop)

When Qluent can't answer, a human can step in. Reply directly in the same thread where the question was asked—provide the correct answer, share the SQL, or ask clarifying questions.

This keeps the conversation going without the user having to start over.

Replying to questions from the web app is not currently supported. Human-in-the-loop works in Slack and Microsoft Teams.

Mark as ignored

Some failures aren't actionable: question is out of scope, data doesn't exist, or user misunderstanding.

Tips

  • Review regularly—don't let failed queries pile up
  • Look for patterns—multiple similar failures = configuration gap
  • After fixing, mark as resolved
  • Focus on high-priority items first