As of September 2026, the honest answer for most B2B companies is to go hybrid rather than pure. A base fee plus a metered component is now the most common model in the market, and it solves the margin problem without handing your revenue forecast over to your customers' behaviour.
This question has moved from a pricing debate to an operational one, because a growing share of software now has a cost of goods sold that is itself metered. If your product calls a model on every request, your costs move with usage whether your prices do or not.
Here is what the current survey data actually shows, the arithmetic that is forcing the change, and the cases where we would tell you not to touch your pricing at all.
Hybrid has become the plurality. Kyle Poyar's 2026 State of B2B Monetization survey, run from April to May 2026 with 230 B2B software and AI companies, found hybrid pricing to be the most popular model at 37 percent.
The direction is the interesting part. The same survey reports that 25 percent of respondents said they were hybrid 12 months ago, and that the number has jumped to 37 percent. That is a fast move for something as sticky as a pricing model.
It also splits by stage. Early stage companies under 5 million dollars ARR favoured flat fee pricing at 37 percent, while large companies above 150 million dollars ARR sat at 29 percent on per seat. Small companies keep it simple, big companies have installed bases they cannot easily move, and the change is happening in the middle.
Because a metered cost sits underneath a growing share of products. Look at what model inference costs and the problem is obvious. Anthropic's published pricing lists Opus 5 at 5 dollars per million input tokens and 25 dollars per million output tokens, Sonnet 5 at 2 dollars and 10 dollars, and Haiku 4.5 at 1 dollar and 5 dollars.
Those are per token prices, which means the cost of serving a customer scales directly with how much they use the feature. A flat monthly subscription against a variable per unit cost is a margin bet, and it is a bet you lose specifically on your most engaged customers.
There are levers on that cost and they matter to the pricing conversation. The same pricing page lists a 50 percent saving with batch processing, and substantially reduced prompt caching rates, with Sonnet 5 caching at 0.20 dollars per million tokens on read against its standard rate. Engineering choices move your margin before pricing does.
Your forecast, and your buyer's ability to get approval. A finance team asked to sign off on a variable line item will either cap it or delay it, and both outcomes slow your deal cycle. A predictable number is a feature of the purchase, not just of the invoice.
It also breaks the relationship between value and payment at both ends. A customer who uses the product heavily in month one and lightly in month two pays you less at exactly the point they are deciding whether to keep it, which is the opposite of what you want.
And it removes the floor. With per seat pricing, a customer who stops using the product still pays until renewal, which buys you time to win them back. With pure usage, revenue disappears the moment attention does, without any of the signals a cancellation would give you.
A platform fee that covers access plus a meter on the thing that scales with your costs. The platform fee gives finance a number to approve and gives you a revenue floor. The meter aligns the rest with what the customer consumes.
The design decision that matters most is what you meter. Meter something the customer controls and understands, and something that rises when they get more value. Metering an internal implementation detail, such as compute seconds, is a recipe for support tickets from people who cannot predict their own bill.
Include a meaningful allowance in the base. A hybrid plan where the meter starts at zero feels like usage pricing with extra steps. A plan where most customers never exceed the included amount feels like a subscription with a safety valve, which is what you want.
It is more common than it was, and it is not free. The same survey found that 29 percent let customers choose between multiple pricing models, up from 21 percent last year. That is a real trend and it usually reflects a company serving two very different segments.
The cost lands on your team rather than the customer. Two pricing models means two forecast models, two sets of sales collateral, two billing paths and two versions of every comparison conversation. Companies underestimate that overhead consistently.
Our view is that a choice of models makes sense when you have two genuinely distinct buyer types, and is a mistake when it is being used to avoid a decision. If you cannot say which segment each model is for, you are not offering choice, you are offering confusion. Our piece on defining an ICP is usually the missing step.
More than most GTM plans account for. ChartMogul's retention analysis, covering more than 2,500 SaaS businesses across H1 2021 to 2024, found that only the top quartile with 500 dollars or more in average revenue per account still achieve 100 percent net revenue retention or better, and that companies between 25 and 500 dollars in ARPA find that harder to reach.
Scale compounds it. That analysis found companies with 1,500 subscribers or fewer achieving 100 percent NRR, while companies with 12,000 or more subscribers typically sat at 76 percent, with only 6 percent of them reaching 100 percent or better. Churn tracked with it, at 7 percent for low NRR companies against roughly 3.5 percent for those at or above 100 percent.
That data runs to H1 2024, so treat it as the shape rather than today's number. The shape is that a pricing change which raises your average account size does more for retention than most retention programmes do, which is an argument that rarely gets made in the pricing meeting.
You need to be able to measure the meter accurately, bill it correctly, and show it to the customer in real time. If any of those three is missing, you are not ready, and shipping anyway produces billing disputes that cost more than the pricing change gains.
The usage display is the one teams skip. A customer who cannot see their consumption until the invoice arrives will assume the worst, and one surprise bill will undo a year of goodwill. Build the usage view before you turn the meter on.
You also need a plan for the customers who will pay more. Some accounts always will, and finding out via an invoice is the worst possible way for them to learn it. Grandfather, cap or communicate, but decide in advance.
It has to be readable by someone who does not know your product. TrustRadius identified transparent pricing as buyers' number one wish list item for vendors for four years running, since it started asking in 2023, and a usage model makes transparency harder rather than optional.
The pattern that works is a worked example. Show what a company of a given shape actually pays, with the numbers filled in. A calculator is good, a concrete example is better, because a buyer can check whether it looks like them without doing any work.
What does not work is a price per unit with no context. Nobody knows how many units they will use, so the number is unusable. Our notes on pricing page design cover how we lay this out.
When you are trying to fix a growth problem. A pricing change will not make people want a product they do not want. It will occupy your best people for a quarter, disrupt your sales motion, and leave you with the same underlying problem plus a migration.
Also when your costs are genuinely fixed. If your product is software with no metered dependency, the margin argument does not apply to you, and you would be adopting complexity for fashion. Per seat is not obsolete, it is just no longer automatic.
And when your buyers are procurement heavy enterprises with annual budget cycles. A variable line item is genuinely harder to buy in that world, and you may find you have optimised for a model your market cannot approve. Our piece on whether product led growth is still the default covers the same tension in motion design.
Model it before you decide. Take your last twelve months of usage data, apply the proposed pricing, and see what each existing customer would have paid. That single spreadsheet answers most of the debate, and it usually surprises everyone in the room.
Then pick the smallest version that tests the mechanism. A meter on one feature, for new customers only, for two quarters. You learn whether your billing works, whether customers understand it, and whether it moves the numbers, without betting the base.
If you would like a second opinion on how the change looks on the page and in the funnel, we are glad to help. You can reach our team at phoenix.studio, and we will tell you honestly if we think your pricing is not the thing holding you back.
Tell us where you want to go. We'll tell you how we'd get you there.