Pricing

Pricing

You decide two things here: the number (the list price) and the shape (packaging, tiers, model, and the floor underneath each price). You produce a price card — a table where every tier has a price, a cost, a value metric, and a margin that someone else can recompute and disagree with. That artifact is the deliverable; opinions without a recomputable margin are not.

You do not charge the money (that is ../invoicing/SKILL.md), build the billing objects (that is ../stripe/SKILL.md), record revenue in the books, or validate the number against acquisition cost (that is ../unit-economics/SKILL.md). Stay on the decision: what should this cost, and how is it boxed.

The spine — run it in this order

The order matters because each step constrains the next. Skip a step and you set a price you cannot defend.

1. Floor (cost). Compute fully-loaded unit cost (COGS + must-cover). Below this you lose money on every sale — why first: it is the one number that is not an opinion.
2. Ceiling (willingness-to-pay). What the buyer will pay before walking — why next: it bounds everything above the floor; price between floor and ceiling.
3. Value metric. The thing the customer buys more of as they grow (seats, usage, outcomes) — why before tiers: it is the spine the tiers hang on.
4. Model. Flat / per-seat / usage / outcome / hybrid — why before tiers: the model decides what a tier even varies along.
5. Tiers. Good/Better/Best or usage bands, ~3 of them — why here: now you have a metric and a model to differentiate on.
6. Discount floor. Max discount = how far list can fall before hitting the cost floor — why after price: a discount rule needs a list price to discount from.
7. Localize (optional). PPP-adjusted regional prices — why last: localize a price that already works at home.

If you cannot answer step 1, stop and get the cost. Everything downstream is guesswork without it.

Margin math you must not get wrong

Markup and margin use different denominators. Conflating them is the single most common pricing error.

markup %  = (price - cost) / cost      <- denominator is COST
margin %  = (price - cost) / price     <- denominator is PRICE

Reference points to sanity-check any card (verify.sh uses these):

A 50% markup is only a 33% margin. A 50% margin needs a 100% markup. They diverge fast.

Bad:  cost $70, want "30% margin", apply 30% markup -> price $91
      actual margin = (91-70)/91 = 23%. You undershot by 7 points.
Good: cost $70, want 30% margin -> price = cost / (1 - 0.30)
      = 70 / 0.70 = $100. Margin = (100-70)/100 = 30%. Correct.

The price-from-margin formula: price = cost / (1 - target_margin). Memorize it; never apply a markup percentage when someone said "margin".

Pick the model

Decide the model before drawing tiers — it determines what the tiers vary along.

The market is moving off pure per-seat. ~85% of SaaS used some usage-based pricing by 2024 (up from ~30% in 2019), and Gartner expects ~40% of enterprise SaaS spend on usage/agent/outcome models by 2030. If you are defaulting to per-seat for an AI product, justify it.

AI-COGS caveat — the margin assumption changes. Classic SaaS runs 80-90% gross margin. AI products run 50-60% because every query has real compute cost. A "we'll just take 80% margin" plan is wrong for AI: the cost floor (step 1) becomes load-bearing, and you reprice when compute cost moves. See references/pricing-models.md for the outcome credit + overage formula (platform fee ≈ 2× delivery cost, e.g. $12K/yr incl. 100 resolutions then $5K per additional 100) and seat-compression math.

Package the tiers

Good/Better/Best is the dominant SaaS shape; the average SaaS company offers ~3.5 tiers. Three is the safe default — enough to anchor, not enough to paralyze.

• Distinct on all three: each tier must differ in features, limits, and price. If two tiers share limits, you have one tier with a typo.
• Value metric is the axis. Tiers move along the value metric you picked in step 3 (more seats, more usage, more outcomes), not a random feature grab-bag.
• Anchor with the top tier. The Best tier makes Better look reasonable; it is doing a job even if few buy it.
• Name the limit, not just the feature. "10 projects" beats "projects" — a limit is what a buyer hits and upgrades on.

Good     Better        Best
$0/$X    $Y            $Z (anchor)
3 seats  25 seats      unlimited seats
1K calls 50K calls     500K calls + overage
email    + Slack       + SSO, SLA, CSM

Set the discount floor

A discount needs a floor, not a vibe. Reps will give away the business if the only rule is "use judgment".

• Floor = COGS + must-cover cost for that tier. Below it the deal loses money.
• Max discount % = (list − floor) / list. This is the largest discount a rep can give without burning margin.
• No sign-off, no floor-break. Discounting below the floor requires explicit approval, recorded. The floor is the deal-desk's hard stop.

Tier "Better": list $1,000, cost floor $400
max discount = (1000 - 400) / 1000 = 60%
A rep offering 35% off ($650) is fine; 65% off ($350) breaks the floor -> sign-off.