Answer Pricing

Answer Pricing Skill

Setup required: Before using this skill, update knowledge/platform_credits.md with your product's pricing model (seat-based, usage-based, or hybrid), your rate constants, and any tier-specific details. The scripts ship with no hardcoded values — you supply them via environment variables or CLI flags.

Use this skill to project a prospect or customer's cost from their prior usage data. It standardizes the projection math across seat-based, usage-based, and hybrid pricing models.

Always read knowledge/platform_credits.md first — it holds the pricing model taxonomy, rate configuration, conversion options, and projection formulas. Cite it in your answer.

Prerequisites

• METABASE_API_KEY in the environment — only needed for Step 0 (pulling a customer's usage from a data warehouse). If you already have usage numbers, skip it.
• Rate config — set the relevant env vars from knowledge/platform_credits.md for your pricing model (e.g. SEAT_PRICE, USAGE_UNIT_PRICE, INCLUDED_UNITS_PER_SEAT). No rates ship in the repo.
• Python 3 (the scripts are stdlib-only; no pip install needed).

Inputs

The user typically provides:

1. Pricing model — seat-based, usage-based, or hybrid. Confirm from the pricing sheet or contract if not stated. (Required.)
2. Current usage or spend and the window it covers (e.g. "10,000 API calls over 7 days", "$3,000 inference spend last month"). If given a team/account ID instead, pull it with Step 0. (Required.)
3. Rate constants — set via env vars (see knowledge/platform_credits.md). No defaults ship in the repo. (Required.)
4. Conversion assumption — if you need to convert from one metric to the billable unit (e.g. inference spend → compute hours), state the ratio and its source explicitly. (Required if applicable.)
5. Seats — POC seat count + target seat count, for scaling projections. (Optional.)
6. Whether they self-host / BYOC — if so, billed compute = $0; optionally quantify the avoided infrastructure cost.

If an optional input is missing, use the documented approach and state the assumption. Rate inputs have no built-in defaults — set them via env vars or CLI flags before running.

Workflow

Step 0 (optional): Pull usage by account/team ID

If you only have a customer name or account ID, fetch their usage from the data warehouse first. This gives you observed usage numbers — the best anchor for any conversion assumption.

python3 .warp/skills/answer_pricing/scripts/fetch_usage.py --account-id <id> --days <n>

Always scope by account/team ID, not email domain — domain filters miss service accounts and bots that still consume usage. Requires METABASE_API_KEY.

Step 1: Confirm the pricing model and gather rate constants

Read knowledge/platform_credits.md to identify the applicable model (seat-based, usage-based, or hybrid). Set the relevant env vars for the customer's tier — no rates are committed to the repo. If the user references older figures, confirm which rate was in effect so the numbers reconcile.

Step 2: Identify and state any conversion assumptions

If you need to convert from an observed metric to the billable unit (e.g. inference tokens → compute hours, API calls → seats), this is the biggest uncertainty lever (±30–50% typical error). Prefer the customer's own observed ratio from Step 0. Otherwise choose and state an assumption explicitly — pass it via CLI flag or env var. Note the direction of the lever.

Step 3: Run the projector

Don't do the math by hand. Set rate env vars first (or pass flags); the script ships with no hardcoded values:

python3 .warp/skills/answer_pricing/scripts/project_credits.py \
  --pricing-model <seat|usage|hybrid> \
  --usage <value> --window-days <days> \
  [--seat-price <price>] [--seats <n>] [--target-seats <n>] \
  [--unit-price <price>] [--included-per-seat <n>] [--overage-price <price>] \
  [--conversion-ratio <ratio>] [--json]

It prints a one-line headline and a compact monthly/annual grid across the configured scenarios. Set PRICE_GRID / USAGE_GRID to compare several price/volume points. Use --json for full multi-horizon detail, or --self-test to confirm model arithmetic after any edit.

Step 4: Write the answer

Use the output format below. Lead with the pricing model used, the key numbers, and any savings vs. an alternative (e.g. current spend vs. projected, or self-hosted vs. managed). State the conversion assumption in one line. Keep it short — don't dump every horizon or sanity-check row.

Output Format

Keep the reply short and scannable: one headline (pricing model + key inputs), the projection result, a savings/comparison line if applicable, and one basis line. Default to the monthly horizon (annual in parens). Offer other horizons or price points only if asked.

---BEGIN FORMAT---

[Customer] — Cost Projection ([Pricing Model]) [Usage] / [W] days → monthly run-rate: $[mo]/mo ($[yr]/yr)[ Self-hosted, so compute = $0.]

Scenario comparison:

• [Scenario A, e.g. "10 seats @ $X/seat"] → ~$[mo]/mo
• [Scenario B, e.g. "20 seats @ $X/seat"] → ~$[mo]/mo
• [Alternative, e.g. "usage-based @ $Y/unit"] → ~$[mo]/mo

→ [Scenario A] is [S]% [cheaper/more expensive] than [alternative]. Basis: [conversion ratio and source]; directional (±30–50% without full history).

---END FORMAT---

Guidelines

• Keep the reply presentable. Short headline, neutral scenario options, one comparison line, one basis line. Don't dump every horizon or sanity-check row.
• Show scenarios plainly. When comparing multiple price points or models, present them neutrally without labeling any as preferred.
• Never quote a single point estimate for usage-derived costs — show a range or multiple scenarios so uncertainty is visible.
• Use account/team ID, not email domain, when sourcing usage data (service accounts hide behind non-customer domains).
• Be evidence-driven. Base rates on knowledge/platform_credits.md; keep facts (configured rates) vs. assumptions (conversion ratios) clearly separated.

After Completing

For a one-off answer, reply inline. For a deliverable, save a markdown brief to reports/pricing/pricing_<customer>_<YYYY-MM-DD>.md and create a PR. For a formatted deck/xlsx, the skill produces the numbers and brief — deck formatting is out of scope.