PLG pricing: The model, the metric, and the number

Pricing arguments are three separate decisions wearing one name.

The model is how somebody starts paying: free forever, a trial, a downgrade, an invoice. The metric is what the price attaches to: seats, usage, tiers. The number is the amount.

Teams argue longest about the number, which is the easiest to change and matters least. The model and the metric are usually inherited from whatever the first version shipped with, are far harder to change later, and determine almost everything about how the business behaves.

This page is about the model and the metric, and about how the tiers fit together. The neighboring chapters own the pieces: where the free line falls, what a good metered unit looks like, and when to make the ask.

What pricing model should a product-led company use?

Four shapes, and the choice follows from properties of the product, not from preference.

The fastest selector is how long your product takes to prove itself. If the aha moment arrives in minutes, a trial has time to work and a reverse trial has time to create a habit worth losing. If it takes weeks, a trial expires before the user has anything to judge, and a permanently free tier is the only structure that gives them long enough.

Two further questions settle the rest faster than a workshop.

What does one free user cost you? If the answer is meaningfully more than zero (compute, storage, a human, a per-call fee you pay downstream), freemium is a subsidy with no end date, and a trial or a metered floor is the honest structure.

Can an individual get real value alone? If not, a free tier gives them nothing to experience, and the model has to start with something that works for a team.

One sub-decision inside any trial deserves its own answer: whether to ask for a card up front. The trade is generally understood to run one way: asking for a card reduces how many people start and raises the share of starters who convert. Published figures for the size of that trade are the same uncomparable trial-conversion numbers this page declines below, so treat the direction as the usable part. Rob Walling, who founded the email tool Drip and now funds bootstrapped software companies, puts it at roughly ten times the trials without a card. His figure is an experienced practitioner’s, not a measurement.

What is worth having is what he does with it. He defaults to asking for the card anyway, on the grounds that a card is a qualifying event and a small company cannot support ten times the trials.¹ That is the opposite of the usual product-led instinct and the reasoning is about support capacity rather than conversion, which is the part worth stealing whichever way you decide. It suits situations where each signup is expensive to serve, or where somebody will follow up personally on every one. Skipping it suits volume, self-serve motions and products confident of their activation. Either is defensible; what is not is choosing without knowing which of those you are.

Tiers: three, and each with a job

Most software converges on three paid tiers plus whatever comes before them. Three is enough to serve distinct buyers and few enough to compare on one screen, and each tier can be given a job:

The entry tier exists to be started on without a conversation. Its job is removing the reason to hesitate, which means the price has to sit below the level at which somebody needs permission.

The middle tier is where most revenue should come from, and it should be the obvious choice for the customer you actually want. If your middle tier is not your most common purchase, the tiers are mis-specified.

The top tier serves the buyer with requirements, not appetite: security review, control, compliance, support commitments. It is bought by people evaluating risk rather than features, and it is where expansion into an organization lands.

The common failures are symmetrical. Too many tiers turns the pricing page into a comparison exercise nobody finishes. Too few leaves the top of the market unserved, so your largest opportunities either negotiate or leave.

Packaging is the harder half. What goes in each tier is a stronger lever than what each costs. A serviceable default, which most software converges on because the dimensions genuinely separate different buyers:

Security, control and compliance are what the largest tier is actually selling, and they are bought by people who will never open the product. Putting a genuinely useful feature up there instead is a common mistake: it annoys the middle tier without helping the buyer who needed an audit log.

The decision has a trap in it: the features you gate for revenue are frequently the same ones you hide from new users, so nobody discovers the thing they were meant to pay for. That collision is worth checking explicitly, and progressive disclosure covers how.

Changing the price on people who already pay

Every pricing decision above is easy compared with the one that arrives eighteen months later, and this is the part almost nothing written about pricing covers.

You have customers on an old plan. The new pricing is better for the business and worse for some of them. What you do next is mostly a question about trust, not revenue.

Grandfathering is the safe default and it is not free. Leaving existing customers on their original terms indefinitely costs you the revenue and, more expensively, leaves you maintaining several pricing models at once: every future change has to be designed against all of them, and support has to know which world each customer lives in. It is the right call more often than not, and it should be a decision rather than an accident. Walling’s rule of thumb makes it one: if moving your existing customers would not raise monthly recurring revenue by at least 10%, the support load and the goodwill are not worth it, so grandfather them deliberately and move on.¹

If you do migrate people, the sequence matters more than the increase. Tell them well before it happens, not at renewal. Say plainly what changes and why. Give a window at the old price so nobody feels ambushed, and Walling suggests two to four months.¹ Too short reads as a trap; too long means they have forgotten by the time it lands and take it for a second increase. Make leaving easy during that window, because a customer who feels trapped is one who stops expanding and answers reference calls honestly. An unannounced increase discovered on an invoice will cost more in goodwill than the increase earns.

Price new customers first. Almost every price change should be tested on people who have no expectation to violate. If the new number holds for them, you have evidence, not a theory, before you go anywhere near the installed base.

One related decision worth making deliberately: annual billing. Paying up front improves cash flow and removes eleven monthly chances to cancel, because a customer who has already paid for the year does not re-evaluate you monthly. The usual discount for it is real money, so the question is whether a year of committed revenue and one fewer churn decision is worth more than the discount. If monthly churn is a problem for you, it is.

The number matters least, but not zero

Two things about price levels are worth more than any benchmark.

Cross the permission threshold deliberately. There is a price above which a person cannot buy without asking somebody, and it varies by market, so there is no universal figure. Below it, purchase is an individual decision made in minutes. Above it, you have entered procurement and need everything procurement wants. Knowing roughly where that line sits for your buyer is worth more than optimizing the figure underneath it.

Underpricing is easy to miss, which is why it persists. It shows up as healthy conversion and weak revenue per account, and that combination reads as success on the metrics most teams watch. Raising prices is unpopular and testable: new customers only, one segment, conversion and revenue read together, not separately. Eran Galperin raised prices at Gymdesk, his gym-management software, in 2021, in some cases by more than half, held existing customers at their old rate for a few months and then moved them to a discounted plan. Of more than six hundred customers, a handful left, and monthly recurring revenue rose about a quarter.² The number worth noticing is the handful.

Conversion rates by model circulate widely, and each one depends on how that company defines a signup, whether a card is required, and what counts as a conversion.³ The comparison worth running is your own model against an alternative, on new signups, which is unusually cheap in pricing: a reverse trial can be tested against freemium and answered in a quarter.

Expansion beats conversion

The number that decides whether a pricing structure is good is not how many free users convert. It is what an account is worth over time.

Slack’s 2019 S-1 reported net dollar retention of 143%: the cohort of customers it had a year earlier was paying 43% more, with no new logos counted.⁴ That is what a structure aligned to customer growth produces: the price rises as the customer gets more value, without a renegotiation and without a salesperson.

So the test for any pricing decision is whether it lets a successful customer pay you more, automatically, as they succeed. A flat per-account fee cannot pass it. Per-seat pricing passes when adoption spreads inside a company. A metered unit passes continuously, which is why consumption pricing keeps spreading into software that is not infrastructure.

The corollary is that low conversion with strong expansion is a healthier business than the reverse, and companies optimizing the first number frequently damage the second.

What to get right

1. Write down which of the three decisions you are actually making. Most pricing meetings are about the number and should be about the model.
2. Cost a free user, fully loaded. That single figure determines whether freemium is available to you at all.
3. Check your middle tier is your most common purchase. If it is not, the packaging is wrong before the prices are.
4. Cross-check gated features against deferred features. Anything on both lists is revenue you have hidden from yourself.
5. Test a price increase on new customers in one segment, and read conversion and revenue per account together.
6. Measure expansion separately from conversion, because a structure that converts well and never grows an account is worse than the reverse.

All of this decides what you collect from customers who buy without talking to anyone. So what happens to the accounts too large, too regulated or too cautious to buy that way, which are frequently the ones paying for everything else? Reaching them without abandoning the product-led motion means letting usage decide who gets a human: product-led sales is how the product qualifies the lead.

Updated August 2026