value-pricing

Value-based pricing study for a product or service, from its website and knowledge base to a ratified price list. Use when someone wants to price or re-price an offering, design tiers or packages, estimate what each feature is worth to customers, or find willingness to pay. Triggers on "price my product", "what should we charge", "pricing tiers", "packaging", "willingness to pay", "value of each feature", "/value-pricing".

Value-based pricing

Price an offering from three sides and let the gaps between them do the arguing:

Side	Question	Method	Gives
Market	What do rivals charge for this feature set?	hedonic regression over competitor plan cards	the reference price and each feature's market premium
Cost	What does one more customer cost to serve?	cost-driver model + people time	the floor, and the volume needed to carry fixed cost
Value	What is it worth to the buyer?	economic value to the customer (EVC) Monte Carlo + a pre-registered conjoint on a persona panel	the ceiling, and each feature's added value

A council of persona agents (buyers per segment, plus pricing experts) argues the proposals; the synthesis scores them; the cost interview turns them into packages that make money; a real-buyer validation plan says what would prove them wrong. The owner rules at every fork.

Every number is a hypothesis until real buyers confirm it. Simulated buyers get direction right and level wrong; say so on every page that quotes them.

Ground rules (every phase)

• Current prices are not inputs. Never feed today's price list to any phase; it anchors the corridor on a number nobody ratified. Compare against it at the end.
• Unknown is null plus a stated gap, never a plausible number. Every assumption is one sentence in the phase's summary.json.assumptions, so the owner can disagree with it.
• The owner rules. Judgement calls (segment axis, crawl permission, archetypes, anchors, which proposal) go to the user as a question with a recommendation, then into DECISIONS.md (YYYY-MM-DD · phase · decision · why · who ruled).
• Read-only against anything in production. Raw crawls and transcripts stay outside git.
• Money in one currency per study, per month, ex VAT; intervals are 90%.
• Before any expensive step (a crawl of hundreds of pages, LLM coding of a corpus, a panel of hundreds of calls) post a token and time estimate to the user and wait for a go.

Workspace

Create pricing-study/ in the user's working directory (or where they say) from templates/:

text

pricing-study/
  study.yaml          product, currency, segments, archetypes, scenarios, conjoint, lenses
  offering.yaml       the capability vector, every feature with evidence (quote + source)
  taxonomy.yaml       the ONE feature vocabulary, phrased the way the market advertises
  value_drivers.yaml  feature → value line → formula (the valuation rubric)
  costs.yaml          the cost structure, from the cost interview
  DECISIONS.md        append-only rulings
  corpus/             crawled pages and ingested notes (gitignore it)
  market/ cost/ value/ council/ panel/ synthesis/    one folder per phase; each writes results/summary.json

Scripts live in this skill's scripts/ and take --study pricing-study. Set up the environment once: uv venv .venv-pricing && uv pip install --python .venv-pricing/bin/python -r <skill>/scripts/requirements.txt (add playwright and run playwright install chromium for JS-heavy sites). Every script has --help. Run scripts/selftest.py once on a new machine: it checks every script against planted ground truth. Keep the study folder on persistent disk, not /tmp: a machine or VM restart clears /tmp.

Parallel work: market, cost and value are independent. When they run as separate sessions or subagents, each writes only inside its own folder, and changes to the shared files (taxonomy.yaml, study.yaml) are proposals logged in DECISIONS.md.

The sequence

Each phase ends on a completion criterion. Read the phase's reference file when you start it.

1. Intake — crawl, ingest, build the offering and the taxonomy. → references/1-intake.md Done when every feature in offering.yaml carries a quote and a source, taxonomy.yaml has an id for every offering feature and every competitor feature seen so far, and the owner has ruled the segment axis and the archetypes.
2. Market — competitor corpus, plan cards, feature coding, hedonic model, meter elasticities. → references/2-market.md Done when market/results/summary.json holds the implied price per segment (OLS, lasso and coverage neighbours), a premium per feature with its interval, κ per feature, and the meter table.
3. Value — value drivers, scenarios, Monte Carlo, per-feature added value, sensitivity. → references/3-value.md Done when every feature has a share of scenarios it valued, a median €, a Shapley share and a leave-one-out loss, and Sobol indices name the parameters that move the ceiling.
4. Council — persona agents consult on value, objections and fences; then the pre-registered panel. → references/4-council.md and references/5-panel.md Done when council/minutes.md records each member's verdict on each draft structure, and the panel's pre-registration is committed before its first response file exists.
5. Cost — interview the owner on the cost structure, then the floor. → references/6-cost.md Done when costs.yaml splits every line into variable / fixed / people / channel with a source or a stated gap, and people time per customer has a number or an explicit range.
6. Synthesis and packages — corridor, feature-value table, ≥3 proposals scored on contribution, extras and multi-unit curves, coherence across channels. → references/7-synthesis.md Done when each proposal states prices per tier, expected mix, contribution after people time under every level lens, anchor and decoy logic, the main risk, and what would prove it wrong.
7. Validation and hand-over — real-buyer survey with decision rules fixed before data; the owner picks. → references/8-validation.md Done when the owner's pick and every open ruling are in DECISIONS.md and the report is delivered.

Phases 2 and 3 can run in parallel; 5 can run any time before 6. The owner asked for a fast pass? Run 1 → 3 → 4 (council only) → 5 → 6 and label the result "market side unmeasured".

Reference

• references/lessons.md — what went wrong or surprised in the studies this method came from. Read it before phase 2 and again before phase 6.
• references/output-contract.md — the summary.json shape per phase.
• references/literature.md — the basis for each method. Verify a citation before it goes into a deck.
• The council personas ship as this plugin's agents: buyer-persona, pricing-strategist, cfo-unit-economics, behavioral-economist, competitive-analyst, gtm-sales-lead, devils-advocate.