Support SLA & Cancellation Signal Queue

Privacy-first read-only human review queue for SaaS support teams. Bring authorized pre-anonymized Zendesk or equivalent support ticket exports in CSV/JSON. This Actor flags response deadlines approaching or overdue under customer-supplied SLA rules and explicit cancellation or non-renewal expressions from permitted customer text. It never predicts churn, sends replies, updates tickets, or accesses Zendesk live.

First-run demo and automated health check

The Apify Store input form prefills demoMode=true (no real file or credentials). If and only if no CSV/JSON source is supplied, this setting generates a single completely fictitious DEMO-TICKET-001 record with a clearly fictional 120-minute SLA and explicit canned cancellation text. Demo timestamps are calculated relative to run time so that Store QA can obtain a non-empty Dataset in under five minutes on any day. The returned RUN-SUMMARY explicitly marks inputMode: SYNTHETIC_DEMO and isSyntheticDemo: true. No fabricated demonstration record is ever mixed with a real CSV/JSON source; uploaded exports always take precedence even if the checkbox remains checked. The standard paid queue-item event also applies to the demo result. Turn off demoMode in production if desired.

Ordinary API/CLI calls without a ticket source and without explicitly setting demoMode to true still return the actionable EXACTLY_ONE_TICKET_SOURCE_REQUIRED error. This prevents silently substituting demo data for forgotten customer exports.

Inputs

Choose exactly one: ticketsFile (Apify upload of CSV/JSON, max 4 MiB), tickets (JSON array), ticketsCsv, or ticketsJson. Max 5,000 input rows and max 1,000 queue results. File URLs must be HTTPS api.apify.com key-value-store record uploads; no arbitrary remote requests, redirects or credentials.

Ticket column aliases: ticket_id / id / ticketId; required created_at and status; optional updated_at, tags, last_agent_reply_at, last_customer_message_at, reopened_at; optional last_reply_at paired with last_reply_actor of agent/customer. For tickets with no reply history, supply awaiting_agent_response:true explicitly. Optional customer_text, customer_message, or latest_customer_message may contain customer-only content you have authority to process. Do not supply full threads, names, emails or agent-authored messages.

If reply history or an explicit pending-reply indicator is absent, no SLA violation is inferred. If last_reply_at exists without an actor role, the SLA is ambiguous. A later reopened_at restarts a reply clock after an earlier agent answer. Closed/solved tickets are excluded.

SLA rules

Each slaRules object has id, targetMinutes, optional warningMinutes, priority (lower wins), tagsAny, tagsAll and activeStatuses (default new/open/reopened; pending and hold excluded). Without a matching user-provided rule, no SLA breach is claimed.

Optionally supply timezone (IANA, default UTC) and businessHours with days (0=Sunday...6=Saturday), start and end (HH:MM) and holidays (YYYY-MM-DD). Without businessHours, 24x7 calendar time applies. This models time-to-next-agent-response only, NOT full provider SLA semantics. Multiple shifts, overnight schedules, complex pause/resume policies, and >2-year spans are not supported.

Every timestamp including optional asOf requires ISO 8601 with a Z or numeric UTC offset. Naive local timestamps are not accepted. Business-hour computation accounts for IANA DST transitions and specified holidays.

Output

Default Dataset includes ONLY: ticketId, reviewRank, reviewPriority, slaStatus, responseWaitState, deadline, remainingMinutes (negative=overdue), overdueMinutes, matchedRuleId, matchedTags, cancellationSignal, signalEvidence, evidenceSource, evaluationTime and reviewAction (always HUMAN_REVIEW_ONLY). RUN-SUMMARY contains only aggregate counters.

Ordering: overdue with a cancellation signal first; other overdue second; explicit signal third; nearing an SLA deadline fourth. A tag alone does not prove cancellation intent. Cancellation evidence is a fixed dictionary phrase, not raw text or a private message. Selected English and Turkish explicit phrases supported; false positives and negatives remain possible. A cancellation signal does not mean confirmed churn or a completed cancellation.

Dataset is restricted to non-identifying ticket IDs, allowed tag names, rule IDs, timestamps and canned evidence, with no customer names, emails or ticket bodies. Other input columns are ignored. The original upload and run INPUT remain in Apify storage according to the user's own retention settings; redact the source file and use short retention before uploading. Keep ticket IDs and tags non-identifying.

Examples and testing

Use examples/input.json (synthetic records), examples/csv-input.json or examples/tickets.csv. For local development: npm ci; npm test; copy a sample input to storage/key_value_stores/default/INPUT.json; npm start. Inspect Dataset and RUN-SUMMARY for privacy.

Pricing

Verified published pricing as of 2026-10-12 is $0.005 per delivered review queue item; no separate start event is configured. The Store pricing page is authoritative. Unflagged input rows carry no result event charge. No LLM, proxy, Zendesk API, automatic replies or background ticket updates. User-set spending limits and cloud compute charges are governed by Apify.

Zendesk API policy (not integrated in v1)

A future read-only integration would require tenant authorization and validation of OAuth tickets:read scope, pagination/cursor handling, incremental export limits and Retry-After/429. The v1 Actor never requests Zendesk keys.

• OAuth scopes: https://developer.zendesk.com/api-reference/ticketing/oauth/grant_type_tokens/
• Incremental exports: https://developer.zendesk.com/api-reference/ticketing/ticket-management/incremental_exports/
• Rate limits: https://developer.zendesk.com/api-reference/introduction/rate-limits/
• Apify monetization: https://docs.apify.com/actors/publishing/monetize/pay-per-event

Version 1.0.0, October 12 2026.