There are six subscription pricing models in common use. Flat rate charges every customer the same price for the same product. Tiered charges different prices for different feature sets or volumes. Usage-based charges by measured consumption. Prepaid term charges for a longer commitment up front at a lower effective monthly rate. Freemium provides a permanently limited version and charges for the full one. Hybrid combines a recurring base fee with a second variable charge.
Choosing between them is a two-part decision that most teams collapse into one. First, you choose a value metric, which is the unit your price scales against. Then you choose a model, which is how you charge against that unit. Getting the second right while getting the first wrong produces a pricing page that looks reasonable and a business that cannot expand revenue inside its own customer base.
Part One: Choose the value metric before the model
The value metric is the thing you multiply price by. Seats, boxes, gigabytes, transactions, orders, active users, API calls, locations, subscribers under management.
A value metric that works meets four tests.
It correlates with the value the customer receives. If a customer’s outcome doubles and their bill does not move, you have a metric that measures your cost rather than their benefit.
The customer can predict it before the invoice arrives. Buyers tolerate a bill that grows. They do not tolerate a bill they could not have forecast. This is the single most common failure in usage-based pricing, and it produces cancellations from customers who are otherwise satisfied.
You can measure it accurately and defend the measurement. If your metric is disputable, every renewal becomes a reconciliation meeting. Metering disputes are a support cost and a churn cause.
It grows naturally as the account succeeds. This is what produces expansion revenue without a sales conversation. A value metric that never grows means every dollar of growth has to be acquired.
Test your candidate metric against your existing base before you commit. Pull twelve months of account level data, plot your current revenue per account against each candidate metric, and look at the correlation. If your best candidate metric has weak correlation with what customers actually pay you today, you are about to reprice your entire base, not adjust a page.
The Six Models
Flat Rate
Definition: One price, one product, one billing interval.
Mechanics. Single SKU, single price, no metering, no entitlement logic. The billing configuration is a plan object and a recurrence interval. Proration is only needed for mid cycle cancellations.
Where it fits. Replenishment products with tight unit cost variance. Coffee, supplements, consumables, pet food. Also fits early-stage software with one job to do.
Where it breaks. Flat rate has no expansion mechanism. Every dollar of net new revenue must come from acquisition, which means your growth rate is capped by your acquisition spend, and your net revenue retention is structurally below 100% by the amount of your gross churn. A flat rate business with 4% monthly gross churn has a hard ceiling on net revenue retention of 96%, no matter how good the product is.
Diagnostic. Segment your base by usage or consumption volume. If your top decile of customers consumes five or more times what your bottom decile does at the same price, flat rate is transferring margin from your best relationships to your worst.
Tiered
Definition: Two or more published price points, each unlocking a different feature set, volume allowance, or service level.
Mechanics. Requires entitlement management, meaning your application has to check what plan a customer is on before granting access to a capability. It also requires upgrade and downgrade paths with proration logic, and a decision about whether downgrades take effect immediately or at the next renewal. Immediate downgrades with credit issuance create refund exposure. Renewal dated downgrades are the standard.
Tier architecture. Build three tiers unless you have a specific reason not to. Three is the point where the compromise effect operates, where buyers avoid the extremes and select the middle, and where the highest tier does useful anchoring work whether or not it sells. A fourth tier increases decision time and decision abandonment more than it increases revenue.
Fences. The thing that separates tiers is called a fence, and there are five kinds. Feature fences gate capability. Volume fences gate quantity. Seat fences gate users. Service fences gate support level or response time. Time fences gate contract length. Feature fences are the most common and the most fragile, because customers negotiate against them. Volume and seat fences are the most durable, because they are objective and they grow on their own.
Where it breaks. Tiers ossify. Once a base is distributed across three price points, repricing means grandfathering a legacy cohort indefinitely or running a migration that will spike churn. Every legacy plan you keep alive is permanent engineering and support overhead. Decide your grandfathering policy before you launch tiers, not when you need to change them.
Usage-based
Definition: The customer pays for measured consumption, per unit, per action, or per transaction.
Mechanics. This is the most demanding model to operate. You need an event ingestion pipeline, an aggregation window, an idempotency strategy, so duplicate events do not double bill, a rating engine that applies the price to aggregated usage, and a way to present the usage detail on the invoice. You also need to decide how you handle late arriving events, which happens more than teams expect, and whether you rebill or absorb.
Pricing structures within usage-based. Straight per unit charges the same rate for every unit. Tiered volume applies different rates to different bands, with each band priced at its own rate. Volume pricing applies one rate retroactively to all units based on the total reached. These produce materially different bills at the same nominal price and buyers rarely understand the difference, so state which one you use in plain language.
Where it breaks. Two places. Revenue predictability drops, which affects your own forecasting and your buyer’s budget approval. And your churn signal degrades badly, because a customer can go to near zero usage for months without ever cancelling. A usage-based business that reports subscription churn is reporting a number that does not describe its health. Track active usage cohorts instead, defined as accounts with usage above a threshold you set.
Guardrails. Add a floor, a cap, or both. A committed minimum gives you forecastable revenue and gives the buyer a discount for the commitment. A spend cap with a notification prevents the invoice that generates a chargeback and a public complaint.
Prepaid term
Definition: The customer commits to a fixed term, typically three, six, or twelve months, and pays up front at a lower effective monthly rate.
Mechanics. Cash is collected in full at the start. For accounting, that cash is deferred revenue on the balance sheet, recognized ratably across the term. Under ASC 606 this is a straightforward performance obligation satisfied over time, but it means your cash position and your recognized revenue diverge sharply, and reporting one when you mean the other is a credibility problem in any diligence conversation. [Not legal or accounting advice, confirm treatment with your accountant.]
The discount math. Here is the calculation almost nobody runs. A prepaid annual plan is worth offering when the discount you give is smaller than the revenue you would have lost to churn over the same twelve months.
If monthly retention is r, expected revenue from a monthly subscriber over twelve months is P multiplied by the sum of the retention series, which equals P times (1 minus r to the twelfth) divided by (1 minus r). Prepaid annual revenue is twelve times P times (1 minus the discount).
Setting those equal gives the maximum discount you can offer before the prepaid plan destroys value:
Read that table carefully. A business with 2% monthly churn offering the standard “two months free” annual discount, which is 16.7%, is giving away more than it recovers. A business with 8% monthly churn offering the same discount is capturing significant value. The identical promotional offer is a mistake at one business and a bargain at the other, and the only variable is churn.
Two things this table does not price in, both of which argue for a slightly larger discount than the breakeven: the cash you collect eleven months early has a time value, and the eleven billing events you eliminate are eleven chances for a payment to fail.
Adoption. Recurly’s 2026 State of Subscriptions, drawn from 2,200 subscription businesses and 76 million subscribers, found that 78% of merchants now offer both monthly and annual billing. Offering both is the norm rather than the differentiator. The decision that still separates operators is the size of the discount, and the table above is how you set it.
Where it breaks. Prepaid delays your churn signal by the length of the term. An annual cohort tells you nothing honest about product health until month thirteen. Teams that scale annual prepaid aggressively often discover a retention problem four quarters after it started.
Freemium
Definition: A permanently available limited version, with payment required to remove the limit.
Mechanics. The critical design decision is not which features to withhold. It is which limit a growing customer inevitably reaches. That limit should be a natural consequence of success with the product, not an arbitrary gate. Storage, seats, volume, and history depth all work. Withholding an obviously core feature does not, because it reads as a demo rather than a product.
The economics. Freemium is only viable when your marginal cost to serve an unpaid user is close to zero and your conversion trigger is reliable. Model it explicitly: free user acquisition cost, marginal cost to serve, conversion rate to paid, and time to conversion. If marginal cost to serve multiplied by average months as a free user exceeds the expected value of a free user, the model loses money at scale, and scale is exactly what freemium produces.
Where it breaks. Freemium anchors your market at a price of zero, which raises the cost of every subsequent pricing decision. It also imports a support burden from a population that will never pay. And it is very difficult to reverse. Removing a free tier that customers have organized their workflow around produces a churn and reputation event.
For C suite readers. Note that freemium is a distribution strategy that happens to have a pricing implication, not a pricing strategy. If you cannot articulate the distribution advantage it buys you, choose a trial instead.
A third option worth testing. Recurly’s 2026 research found that short duration paid passes, meaning day, week, or weekend access, converted 13% of buyers into recurring plans. If your free trial conversion is weak, a small paid commitment qualifies intent in a way a free trial cannot, and it collects a payment credential that has already cleared before the first full cycle.
Hybrid
Definition: A recurring base fee combined with a second variable charge, usually usage based or per transaction.
Mechanics. Most mature subscription businesses land here. A membership fee plus member pricing on purchases. A platform fee plus a per transaction rate. A tiered plan with usage overages above an included allowance. The included allowance model is the most common and the most buyer friendly, because it preserves bill predictability inside the allowance while capturing upside above it.
Where it fits. Businesses with both a fixed cost to serve and a variable cost to deliver. Also, the natural destination for a box brand moving toward a membership model, where the fee buys access and pricing rather than product.
Where it breaks. Complexity compounds. Every pricing dimension is another thing the buyer must understand before purchasing, another line the billing system must compute correctly, and another place a dispute can originate. The practical test: if your pricing page requires a paragraph of explanation before the numbers make sense, the model is too complex for the buying process you have.
The Comparison
Part Two: How to actually set the number
Choosing a model tells you the shape of the price. It does not tell you the price. Three research methods, in ascending order of cost and reliability.
Van Westendorp price sensitivity meter
Four questions, asked of prospects or customers who understand the product:
- At what price would this be so expensive you would not consider it?
- At what price would this be expensive but you would still consider it?
- At what price would this be a good value?
- At what price would this be so cheap you would question the quality?
Plot the cumulative distributions and read the intersections. The crossing of “too expensive” and “too cheap” gives the optimal price point. The crossing of “expensive” and “cheap” gives the indifference price point. The range between the other two intersections is the acceptable price range.
What it is good for: establishing a plausible range fast, with a small sample. Roughly 150 to 300 usable responses gives a readable curve.
What it is not good for: predicting volume. Van Westendorp measures stated perception, not purchase behavior, and stated willingness to pay runs consistently above revealed willingness to pay.
Gabor Granger
Present a single price and ask purchase likelihood. Then raise or lower the price and ask again. Repeat across a ladder of prices. The output is a demand curve and, multiplied through, a revenue curve with a visible maximum.
What it is good for: finding the revenue maximizing point on a single SKU.
What it is not good for: anything involving multiple tiers or feature bundles, because it prices a fixed product rather than a configurable one.
Conjoint and MaxDiff
Present respondents with competing bundles of features at different prices and force a choice. The model backs out the implied value of each individual attribute, including price, from the pattern of choices.
What it is good for: designing tiers. This is the only method on this list that tells you which features belong at which price point rather than just what the price should be.
What it is not good for: small budgets and fast timelines. Conjoint needs a larger sample, careful design, and analysis capability. Roughly 300 or more respondents per segment is a working floor.
Testing price changes without breaking anything
You cannot run a classic split test on price the way you run one on a headline. Three reasons: showing different prices to different users simultaneously creates fairness and disclosure exposure that varies by jurisdiction, prospects compare notes and screenshot pricing pages, and the effect you care about takes months to appear because it lives in retention, not in conversion.
Four workable approaches:
New customers only.
Apply the new structure to new signups and leave the existing base untouched. Compare the new cohort against the immediately preceding cohort on conversion rate, first cycle retention, second cycle retention, and ARPU. This is the default and it is usually right.
Geographic split.
Apply the change in one market. Cleanest isolation, but only valid if the markets are genuinely comparable on mix and channel.
Sequential testing.
Run the old price for a defined window, then the new price for an equal window, and compare. Vulnerable to seasonality and to any concurrent marketing change, so hold everything else fixed and document what you held fixed.
Cohort holdout on migration.
When repricing an existing base, hold out a randomly selected control group from the migration. This is the only way to measure what the migration actually cost you in churn, and almost nobody does it.
Whichever you choose, decide the measurement window before you start. Conversion rate moves in days. ARPU moves in weeks. Retention moves in months, and retention is where the real answer is. Calling a pricing test at day fourteen because conversion held is how businesses discover in month five that they traded margin for a worse retention curve.
Migration mechanics for an existing base
If you are changing price on customers who already pay you, the sequence matters more than the number.
Notify well ahead of the change.
Check the notice period required by your terms and by the consumer protection rules in your operating markets before you set a date. In several jurisdictions, a price increase on a recurring charge requires advance notice and in some cases a fresh affirmative consent.
Give the increase a reason and name what is new.
A price change presented alongside specific added value converts differently from a price change presented alone.
Offer a lock in.
Letting existing customers prepay a year at the old rate converts a churn risk into cash and gives your most price sensitive customers an exit that is not cancellation.
Grandfather deliberately, with an end date.
Indefinite grandfathering is a permanent tax on your engineering and support teams. If you grandfather, set the term.
Hold out a control group.
As above. Without it you will never know the counterfactual and you will argue about it for a year.
Frequently asked questions
What is the most common subscription pricing model?
Tiered pricing is the most widely used model across both subscription ecommerce and subscription software, because it lets one product serve customers of very different sizes without requiring separate products. Businesses that launch on flat rate commonly move to tiered once they can identify the dimension along which customer value varies.
What is a value metric in subscription pricing?
A value metric is the unit that price scales against, such as seats, boxes, transactions, or gigabytes. A good value metric correlates with the value the customer receives, is predictable to the customer before the invoice arrives, is measurable and defensible, and grows naturally as the account succeeds.
How much of a discount should an annual prepaid plan offer?
The maximum defensible discount is the amount of revenue you would otherwise lose to churn across the same twelve months. At 3% monthly churn the breakeven discount is roughly 15%. At 5% monthly churn it is roughly 23%. Businesses with low churn that offer the standard two months free discount of 16.7% are usually giving away more than they recover.
How do you test a subscription price change?
Apply the new structure to new customers only and compare against the immediately preceding cohort on conversion rate, ARPU, and first and second cycle retention. Set the measurement window before the test starts, and measure for months rather than days, because the meaningful effect of a price change appears in retention rather than in conversion.
Should I grandfather existing customers when I raise prices?
Grandfather deliberately and with a defined end date. Indefinite legacy plans become permanent engineering and support overhead, and every additional legacy plan raises the cost of the next pricing change. Offering existing customers the chance to prepay a year at the old rate is usually a better instrument than an open ended grandfather.
What is the difference between tiered volume pricing and volume pricing?
Tiered volume applies a different rate to each band of units consumed, so early units stay at the higher rate. Volume pricing applies a single rate retroactively to all units based on the total reached. The same nominal rates produce materially different invoices, so state which structure you use in plain language on the pricing page.
Conclusion
Pricing is the only decision that touches acquisition, retention, and margin at the same time. Every other lever moves one of the three.
The operators working through this in real time are the ones worth learning from. That is the conversation we host at SubSummit. Learn more about SubSummit 2027, hosted in Nashville.