Skip to content

Default Filters Library

Filters help control cost, relevance, and quality by limiting which conversations an AI Task runs on.

This chapter provides a library of recommended default filters for common AI Tasks. Treat these as starting points—tenants may override filters for their needs.

General guidance

  • Start with conservative filters for expensive tasks.
  • Validate filter impact by measuring:
  • eligible conversations per day
  • skipped conversations due to filters
  • Avoid filters that silently exclude too much (e.g., duration > 2 minutes might exclude many relevant calls).

Voice call filters (speech)

Filter: Minimum duration

Use for: CSAT, sentiment, topics, reason/outcome
Recommended: duration > 15 seconds (or 30 seconds if many short/empty calls)

Why: Very short calls often lack enough evidence for scoring.

Filter: Inbound calls only

Use for: support-oriented CX metrics (CSAT, churn risk)
Why: Outbound calls may represent sales follow-ups or proactive outreach.

Filter: Exclude internal/test calls

Use for: all analytics
How: exclude calls tagged as test, or from internal queues/numbers (depends on available metadata)

Text channel filters (chat/email/tickets)

Filter: Minimum content length

Use for: summarization, sentiment, topics
Recommended: minimum total characters or minimum number of messages (if supported)

Filter: Ticket status

Use for: resolution/outcome metrics
Recommended: analyze only resolved/closed tickets if you want final outcomes.

Task-specific filter recommendations

Summarization

  • include most conversations (low filtering)
  • consider excluding extremely short conversations

CSAT / NPS-like scores

  • minimum duration/content length
  • optional: only when a resolution is present (if metadata supports)

Topic analysis

  • minimum duration/content length
  • optional: exclude voicemail/hold segments if identifiable

Competitors mentioned

  • focus on sales queues/campaigns if metadata exists

Measuring filter effectiveness

Recommended metrics: - eligible % of conversations - output coverage % (how many get a result) - error rate on eligible conversations - cost per eligible conversation

Available filter attributes

Common filterable attributes in MiaRec include:

  • Duration – filter by call length
  • Direction – inbound / outbound
  • Channel – call / chat / email (as supported)
  • Queue/team – filter to specific routing groups

Provide a standard set of filters for every global task and document them. Track filter skip rates to detect misconfiguration early.