Setup & Customization Checklist

1. Data sources

• {{CRM}} — confirm closed-won deals carry a close date and amount, and that upsells appear as additional deals (or map to however your CRM records plan changes).
• {{BILLING_SOURCE}} — the skill must be able to read the live plan definition (base fee, included allocation, per-unit rate) and the raw charge history. If your plan metadata lives inside the billing object (e.g. price metadata), point the skill there explicitly — never at a static mapping table.
• {{USAGE_SOURCE}} — monthly consumption by account, active users with roles, and which capabilities are configured. If overage purchases are merged into a single balance field, note where the plan allocation lives so 2.2 doesn't double-count.
• {{USAGE_UNIT}} — one word, used consistently in every table.

2. Tunable defaults

All thresholds ship as defaults — tune them to your pricing, then leave them alone so analyses stay comparable across accounts.

• {{CONSERVATIVE_FACTOR}} (0.90) and {{GROWTH_FACTOR}} (1.50) — multipliers on the trailing 3-month average for the two proposals. Widen the gap if your customers' usage is spiky; narrow it if steady.
• {{COMMIT_DISCOUNT}} (10%) — the per-unit discount the growth option earns. Must match what you can actually honor.
• {{OVERAGE_MULTIPLIER}} (2.0×) — your on-demand rate vs. committed rate. This ratio is the engine of most volume plays; if yours is lower, the "repeated overage" pattern fires less often and that's correct.
• {{HIGH_UTILIZATION_THRESHOLD}} (90%) — what counts as "at the ceiling."
• {{INACTIVITY_WINDOW}} (30 days) — the churn-risk gate. Shorten for daily-use products, lengthen for monthly-cadence ones.
• Rounding boundary for proposal volumes — "nearest natural tier" should map to your plan tiers (nearest 1,000 units, nearest seat band, etc.).

3. Policy decisions

• {{EXPANSION_OWNER}} — one named person. They approve every play, receive every no-reply escalation, and can override proposal volumes by fiat.
• Never auto-send — the strongest invariant. The skill drafts copy and queues sequences; a human sends everything. If you want pure analysis with no drafts, delete section 3.4 and stop at the timing call.
• Decide your commitment-averse framing: if you sell "bulk purchase / draw-down" instead of "annual contract," keep the language rule in 2.6; otherwise delete it and say "annual" plainly.
• Relationship-led plays require confirming the champion + stakeholder list with {{EXPANSION_OWNER}} before any sequence exists — keep this gate; multi-threading the wrong exec burns real relationships.
• Map the role-to-gap guidance in 3.2 to your product: write down, for each buyer role you sell to, the one adoption gap that most concerns them. The skill fills in specifics per account, but the role framing is yours.

4. Behavioral invariants (do not remove)

• Base fee and consumption fee reported separately — the whole invoice is never "platform fee."
• Included allocation always read from the plan definition, never from a balance field.
• Proposals sized from the three most recently completed months — the current partial month never enters the average.
• The inactivity gate ends the analysis at Part 1 — no expansion play on a silent account.
• Every analysis lands on a matched pattern or an explicit new-pattern proposal — never a shrug, never a forced match.
• No placeholder numbers in copy intended to send.
• One active expansion play per account at a time — tag on launch, clear on reply or close.
• No-reply escalations go to {{EXPANSION_OWNER}} internally — never another touch to the customer.