Pricing Strategist
Structured answer to the pricing-design question: which model, which range, which packaging.
The skill recommends a model and a price range. A human picks the number, owns the trade-off, and runs the GTM.
When to Use
- Launching a SaaS / API / AI product and choosing a first pricing model
- Revisiting pricing after 18+ months of GTM data (model shift, not just a price bump)
- Designing or redesigning tier packaging (Good/Better/Best, Bronze/Silver/Gold)
- You have Van Westendorp survey responses and want the acceptable price range
- An exec or board asks "what should we charge?" and you need the structured answer
- You suspect packaging anti-patterns: decoy tier, feature dump, no upgrade trigger
Do Not Use For
| Need | Use instead |
|---|---|
| Per-deal discount approval | deal-desk workflow |
| Brand, category creation, CMO positioning | CMO advisory |
| Whole-company revenue strategy | CRO advisory |
| Technical-sale enablement / POC scoping | sales-engineer workflow |
Workflow
Step 1 — Assess customer context (~20 min)
Capture, in a JSON brief:
- industry and buyer (SMB / mid-market / enterprise)
- average deal size and current customer count
- value drivers — what the customer measurably gains
- consumption pattern: seat, usage, value, or hybrid
- adoption curve (viral / sales-led / bottom-up-then-top-down)
- competitor models and their published price points
Missing fields are not fatal — the picker degrades gracefully and marks lower confidence.
Step 2 — Pick the pricing model
python3 scripts/pricing_model_picker.py --input brief.json --profile saas --output markdown
Ranks five models (seat, usage, value-based, freemium, hybrid) by fit score 0–100, each with a rationale and a trade-off. Read the top two, not just the top one — the runner-up is usually the answer for the next 24 months.
Step 3 — Validate willingness to pay
python3 scripts/wtp_analyzer.py --input survey.json --output markdown
Van Westendorp Price Sensitivity Meter over four questions per respondent:
- At what price would this be so cheap you'd doubt the quality? (
too_cheap) - At what price is it a bargain — clearly good value? (
bargain) - At what price does it start to feel expensive but still worth considering? (
expensive) - At what price is it too expensive — you would not buy? (
too_expensive)
Output: the four intersection points (OPP, IDP, PMC, PME) and the Range of Acceptable Prices (PMC → PME).
Minimum N = 30. Below that the curves are noise; the script warns and still runs so you can sanity-check direction.
Step 4 — Design packaging
python3 scripts/packaging_designer.py --input features.json --profile saas --output markdown
Assigns features to Good / Better / Best by value-to-segment and cost-to-serve, then flags anti-patterns:
- Feature dump — Best tier carrying more than ~60% of total feature value
- Decoy tier — a tier no segment would rationally choose
- No upgrade trigger — adjacent tiers separated only by limits nobody hits
- Price ladder gap — step ratios that break the 2–3× rule of thumb
- Orphan feature — feature with no segment that values it above its cost
Step 5 — Decide
Take model + range + packaging to the pricing committee. The skill does not commit the number and should not be asked to.
Scripts
All three are stdlib-only Python 3.8+, no install step, no network calls.
| Script | Purpose | Input |
|---|---|---|
scripts/pricing_model_picker.py |
5-model fit scorer | brief.json |
scripts/wtp_analyzer.py |
Van Westendorp PSM | survey.json |
scripts/packaging_designer.py |
Good/Better/Best designer + anti-pattern detection | features.json |
Common flags: --input PATH, --output markdown|json, --profile saas|api|marketplace. Run any script with --help for the full schema, and see references/pricing_canon.md for worked input examples.
Minimal inputs
brief.json
{
"product": "AI support triage",
"buyer": "mid-market",
"avg_deal_size": 18000,
"customer_count": 140,
"consumption_pattern": "hybrid",
"value_measurable": true,
"value_metric": "tickets deflected per month",
"adoption": "bottom-up-then-top-down",
"competitor_models": ["seat", "usage"]
}
survey.json
{
"currency": "USD",
"respondents": [
{"too_cheap": 50, "bargain": 150, "expensive": 400, "too_expensive": 900}
]
}
features.json
{
"segments": ["smb", "mid", "enterprise"],
"features": [
{"name": "Core dashboard", "cost": 1, "value": {"smb": 8, "mid": 6, "enterprise": 4}, "table_stakes": true}
]
}
Reading the Output
- Model score is a fit score, not a probability. 70+ means "defensible", not "certain".
- OPP is not the price. It is the point where equal shares call you too cheap and too expensive. Real prices cluster between IDP and OPP.
- PMC → PME is the negotiation envelope, not the list price. List at or just under IDP; discount down toward PMC.
- Tier count — three is the default because it maps to a reference effect. Two is fine for API pricing; five is almost always a mistake at the SMB end.
Anti-Patterns
- Recommending a specific number — this skill emits a range
- Running PSM with N < 30 and treating the result as evidence
- Treating the OPP as "the price" rather than the centre of a range
- Choosing value-based pricing without a value metric you can actually meter
- Designing tiers before the model is chosen — packaging is downstream of the model
- Packaging feature dumps into Best to justify the price
- Copying a competitor's published price as a proxy for willingness to pay
Verification
Before presenting results, confirm:
- The brief's
consumption_patternmatches the winning model's assumptions - Survey N ≥ 30 and the four thresholds are monotonically ordered per respondent (
too_cheap < bargain < expensive < too_expensive) — the analyzer reports violations - Every tier has at least one feature a named segment is willing to pay for
- The recommended range has a stated currency and billing period
References
references/pricing_canon.md— model selection heuristics, PSM mechanics, tier design rules, and source reading list
License
MIT — see LICENSE.