Pricing and packaging is the architecture that captures the value your positioning earned: the value metric you bill against, the good-better-best tiers fenced around willingness-to-pay, the price points, and a migration plan that moves your existing book without breaking it. A SaaS pricing strategy is broken when discounting is a reflex instead of a decision, and when you bill for the wrong unit (seats or tokens) while the customer measures success in something else. Fix the value metric first, because every tier and price point sits downstream of it, and the wrong axis cannot be saved by a better page.
What is the constraint, really?
Most teams treat pricing as a number picked in a meeting and a page someone redesigns once a year. The real constraint is the machinery underneath: positioning sets the ceiling on what you can charge, and pricing is the lever that captures it. The load-bearing decision is the value metric, the unit you charge against. Patrick Campbell frames price as an exchange rate on the value you provide, which makes the value metric the denominator. A team billing per seat while an AI agent does the work of twenty seats has built a meter that punishes its own best work.
How do you tell pricing is YOUR binding constraint?
Three signals say the problem is structural rather than a sales-execution gripe:
- The meter ticks up when the customer does something they want to minimize. You bill per seat while they consolidate seats, or per ticket while their goal is fewer tickets. The bill grows when the customer fails, so renewals fight you.
- Gross margin bleeds on usage you do not recover. For an AI-infused product, inference can run roughly 35 to 40 percent of cost at scale, so a flat per-seat price on unbounded usage is a margin time bomb.
- Realized price is soft, or a tier sits empty. You give back roughly a sixth of list before signature every time, or a default middle tier nobody chooses tells you the fences are wrong.
If two or more are true, pricing is your constraint, not your messaging or your funnel. If you are not sure which is binding, run the diagnostic before touching a price.
The method: how to design a SaaS pricing strategy that holds
The short version of a ten-step architecture audit. Five moves carry the result.
1. Pick the value metric on a four-test scorecard. A metric has to be value-aligned (it rises when the customer succeeds), scalable (no artificial ceiling), simple enough that a buyer self-estimates the bill, and measurable enough that you meter it and they verify it. Drop one test and it breaks. Run the adversarial-meter check: if the meter only ticks up when the customer does something they want to minimize, re-meter on what they maximize.
2. Choose the model, and run the credit gate. Charge for access (subscription), consumption (usage), or result (outcome). Most mature companies land on hybrid: a platform fee plus a usage or outcome layer, now the default for AI vendors as adoption climbed from roughly a quarter to 40 percent inside a year. If you are tempted by credits, stop. They expose your cost structure, become incomprehensible at scale, and deliver the pain of paying all at once at renewal. Reserve them for infrastructure that other software companies resell.
3. Set price points off economic value, not cost. Cost is a floor you bleed below, never the basis. Build an Economic Value Estimation per segment: the buyer's real next-best alternative (often "do nothing" or "build it") plus the differentiation value you add, minus their switching cost. That sets the ceiling, and your price is a deliberate share below it. Write the ceiling as a function of the competitor's price, since a static estimate holds only until the rival re-prices.
4. Validate willingness-to-pay as a band, never a number. Stated willingness-to-pay is evidence, not truth: people overstate it by roughly a third in surveys, so haircut every number. Then cross the band against revealed behavior: where deals die on price in your win/loss notes, and how deep reps discount before they cave. Where the economic-value model and stated willingness-to-pay disagree, do not average them. The gap is the finding.
5. Build good-better-best, fenced on the value metric first. A tier is a willingness-to-pay segment wearing a feature costume, not a feature list. Classify every feature by its job (a Leader drives the buy, a Filler sweetens without selling, a Killer de-values the bundle and stays out), and fence on the value metric so that when an account outgrows a tier on the meter, expansion is automatic. That deliberate success gap is the engine of net revenue retention. Plan the migration first, because price changes break on existing customers, not new ones.
A worked example with real numbers
Take a workflow platform, call it Meridian, selling an AI agent that reads and structures regulated financial filings, sold today as a flat 24,000 dollars a year per seat. Positioning handed over a defensible ceiling closer to 40,000, on a reframe from "OCR tool" to "extraction-and-control system."
Seats are an adversarial meter here. The agent does the work of twenty seats while Meridian bills for one, and inference runs close to 38 percent of cost, so a flat seat price on unbounded extraction bleeds margin as the best customers use it most. The fix is a hybrid: a platform floor for predictability plus a usage layer on documents successfully extracted, the unit the customer already tracks as their own win.
| Decision | Before | After |
|---|---|---|
| Value metric | Per seat | Platform fee + per successful extraction |
| List price (new logos) | $24K flat | ~$31K realized, hybrid |
| Margin on heavy use | Erodes as usage grows | Usage layer recovers inference cost |
| Expansion | Manual upsell conversation | Automatic as extraction volume grows |
| Net revenue retention | Flat | 108% to 121% as the usage layer expands |
On migration, new logos go onto the hybrid sheet now while the existing book keeps its 24,000, with the newly-fenced premium behind the upsell, so upgraders self-migrate on a multi-year price lock instead of churning.
Common mistakes that quietly cost you money
- Feature shock and the undead tier. Cramming every tier with features instead of fencing on the value metric, so buyers cannot tell which tier they are and default to the cheapest, plus a top tier that exists only to flatter the others. Fence on the metric and cut the dead tier.
- Minivation. A price increase so timid it does not move realized price after the deal desk discounts it back. A list increase given away at renewal is a press release, not a pricing change.
- Asserting a willingness-to-pay you never tested. A number from a meeting, not from money or lost deals. If it was not tested against real stakes, it is a hypothesis.
- Migrating in a single switch. The installed book sits at wildly different realized prices, and forced migration spikes churn. Use time-limited grandfathering, and model every account's delta before rollout to prevent bill shock.
How does the fix show up in revenue?
Pricing shows up in three numbers, not one. Realized price, the pocket price after every on-invoice and off-invoice discount, is the only one that matters: a one-point improvement moves operating profit by a multiple set by your margin, at zero change in volume. Net revenue retention rises when the value metric fences expansion automatically. The price-realization rate (realized over list) tells you whether the deal desk ate the increase. Instrument all three by segment, not blended, and freeze the baseline first.
The capture happens where the buyer transacts, which routes into Demand and Conversational Pipeline for the pricing page and the paywall, and the proof routes into Measurement and Attribution for the price waterfall. Pricing does not stand alone. It inherits its ceiling from positioning, since the price is only defensible if the differentiation is real, and it has to agree with your go-to-market motion: a self-serve price on a sales-led product (or the reverse) is a machine fighting itself.
FAQ
What is the difference between pricing and packaging? Pricing is what you charge: the value metric and the price points. Packaging is how you sell it: the good-better-best tiers and the fences that separate buyer segments. They are one decision, because a price means nothing until it is attached to a package a specific buyer reaches for. The value metric is the spine that holds both together.
What is a value metric and how do I pick one? A value metric is the unit you charge against: a seat, a thousand API calls, a successful outcome. Score candidates on four tests: it should rise when the customer succeeds, scale without an artificial ceiling, be simple enough that a buyer predicts their own bill, and be measurable on both sides. The cleanest filter is the adversarial-meter check: if the meter only grows when the customer does something they want to minimize, re-meter on what they maximize.
How do I raise prices without losing existing customers? Migrate cohort by cohort, never in a single switch, because a forced migration across a book at wildly different realized prices spikes churn. Default to time-limited grandfathering: move new logos immediately, hold existing customers on legacy pricing for a bounded window, and put the newly-fenced premium behind an upsell so upgraders self-migrate. Lead with value before the number, anchor to the customer's own old price, and model every account's invoice delta before rollout to prevent bill shock.
Should SaaS pricing be per seat, usage-based, or hybrid? Most mature companies land on hybrid: a platform fee for predictability plus a usage or outcome layer that captures value as the customer scales. Pure per-seat is the most predictable and the weakest on value when AI decouples headcount from output. Pure usage expands hardest but creates the most bill-anxiety. For an AI-infused product, check your cost of goods first, because inference cost makes a flat seat price on unbounded usage a margin risk.
Are usage credits a good pricing model? Rarely, for an application. Credits expose your cost structure, become incomprehensible once each product has its own exchange rate, and deliver the pain of paying all at once at renewal. They are defensible only for genuine infrastructure that other software companies resell. If you do ship credits, make the wallet dollar-denominated, opt-in, and zero-charge when unfunded, and pre-commit to no retroactive rate changes. A redenomination dressed up as a "technical adjustment" is the fastest way to lose your heaviest users.