1---2name: apexcloud-retention-ops-113description: Build ApexCloud retention-ops JSON deliverables from the public task-env API and churn CSV exports. Use when asked for customer risk queues, QBR packets, receivables and pipeline reviews, retention action boards, or churn validation/ranking from the staged ApexCloud environment.4metadata:5 source: https://skillmd.com/skills/prism-shadow/apexcloud-retention-ops-116---78# ApexCloud Retention Ops910## Start11121. Identify the deliverable family from the prompt: risk queue, QBR packet, receivables review, action board, or churn validation/ranking.132. Read the prompt template first, then mirror its key order exactly.143. Use [references/apexcloud_rules.md](references/apexcloud_rules.md) for field precedence, joins, and rollup rules.154. Use [scripts/apexcloud_ops.py](scripts/apexcloud_ops.py) when a normalized draft will save time.1617## Data precedence1819- Use billing snapshots for ARR when the prompt asks for current ARR, exposure, or ARR at risk.20- Use exact normalized legal-name matching for A/R aging links. Do not fuzzy-match similar subsidiaries or noise rows.21- Exclude spam, duplicate, and cancelled tickets from clean support counts and SLA math.22- Treat SLA as met only when both first response and resolution SLA are true.23- Use the latest non-retracted NPS response inside the window.24- Sum only open opportunities inside the requested window for open pipeline.25- Keep CRM ARR separate from billing ARR unless the prompt explicitly asks for CRM exposure.2627## Output rules2829- Return JSON only unless the prompt explicitly asks for something else.30- Preserve the template shape and key order.31- Round currency to 2 decimals, percentages to 1 decimal, counts as integers, and probabilities to 3 decimals.32- Sort rows the way the prompt asks: score desc, customer name asc, month asc, or rank asc.33- Fill policy-code fields last. Choose the code whose label matches the rule you actually used.3435## Task families3637### QBR packet3839- Build monthly revenue from the account metrics endpoint.40- Recompute support ticket counts and SLA compliance from clean tickets, not from raw monthly ticket totals.41- Derive the ticket trend from the month sequence in the prompt.42- Use the prompt's source enums for revenue, support tickets, SLA compliance, and NPS.4344### Receivables and pipeline review4546- Start from A/R aging rows with overdue in the older buckets.47- Link A/R rows to CRM accounts only by exact normalized legal name.48- Sort overdue followups by `customer_name` ascending.49- Summarize pipeline from opportunities in the requested window, separating open, won, and lost.50- Pull HR and event context from the requested summary endpoints and aggregate across regions when needed.5152### Risk queue and action board5354- Score accounts from renewal timing, overdue balance, NPS, SLA, usage trend, tenure, and open expansion pipeline.55- Use `collections_followup` when overdue receivables dominate.56- Use `technical_recovery` when support or product health dominates.57- Use `renewal_save` when renewal timing is the main risk.58- Use `nurture_monitor` only when the account is low risk.5960### Churn validation and ranking6162- Fit on `train.csv`, validate on `validation.csv`, then rank only the requested candidates.63- Use the heuristic fallback in the helper script if you do not have a stronger model.64- Keep the validation summary simple: row counts, feature count, accuracy, band, and tenure direction.6566## When the environment is awkward6768- If a per-account billing or A/R URL returns `not_found`, fall back to the collection endpoint and filter in memory.69- If the prompt gives a fixed follow-up calendar, map actions to those dates directly.70