How software products make money now: seats, credits and outcomes

How software products make money now: seats, credits and outcomes

In September 2026 our news desk recorded 28 changes by software vendors that had both a notice date and an effective date. The median notice was 31 days. Six gave a week or less, three of them none at all. Eight gave more than 90 days. Four of the 28 switched something on by default for existing customers, so that the customer had to act to opt out.

The changes themselves show where software pricing is going. Atlassian is raising Cloud list prices by roughly 3% to 10% on 13 October, and from 3 December will start billing for usage beyond plan allowances on three meters: Rovo AI credits, automation steps, and resolutions completed by AI agents. From 2 November, new Microsoft 365 Copilot Business licences bought through Microsoft's reseller programme come with pay-as-you-go usage billing switched on by default, with a preset monthly limit. Trae, ByteDance's AI coding editor, is keeping its Pro+ and Ultra prices the same for existing monthly subscribers while cutting their monthly usage allowances, from $90 to $60 and from $400 to $200. Microsoft's reseller promotions on Microsoft 365 E5, E7 and Copilot end on 30 September. We reported each of these, with sources, in our news section: Atlassian, Microsoft's CSP changes and Trae.

Our catalogue shows how fast this makes published prices stale. 5,092 software listings on TrustList still show prices from the original 2020 import, with no date on them. When we checked a sample of 30 against the vendors' own pricing pages, only 6 still matched. Prices move faster than directories, ours included, and we are dating or removing those figures as we rebuild the listings.

This article sets out how software is priced in 2026, what the vendors' own filings say about which models carry the revenue, and then gives two practical tools: a decision method for a small software company choosing how to charge, and a checklist for a buyer reading a usage-priced contract.

Four pricing models, and most products now use two of them

Almost every business software product charges in one of four ways, and in 2026 the typical product combines at least two.

Per seat. A fixed price per named user per month or year. It is predictable for both sides, easy to budget and easy to compare. Its weakness is that it charges the same whether a user works in the product all day or logs in once a quarter, and it has no natural way to charge for work done by software rather than by people.

Usage or credits. The customer pays for what is consumed: API calls, compute minutes, tokens, automation runs, or an abstract unit such as a credit that different actions consume at different rates. Usage pricing matches cost to value more closely, but it makes the bill harder to predict, and credits add a conversion step that makes comparison between vendors difficult.

Outcome. The customer pays when a defined result happens, such as a support conversation resolved without a human. This is the model AI agents have made practical. It is attractive when the outcome is easy to define and verify, and contentious when it is not.

Flat rate or tiered. One price for a plan, often with limits on users, records or storage. Common for small-business tools and for free tiers.

Here is what the vendors' own pricing pages showed when we opened them on 25 September 2026.

Product	Base price	How AI or heavy use is charged
GitHub Copilot Business	$19 per user a month	Includes $19 of AI Credits per user; each credit is $0.01; usage measured in tokens at each model's published API rates
GitLab Premium	$29 per user a month, billed annually	GitLab Credits at $1 each on demand; 12 credits per user a month included as a promotion
Salesforce Agentforce	$125 per user a month add-on, or $2 per conversation, or Flex Credits	Flex Credits at $500 per 100,000; a standard action uses 20 credits, which works out at $0.10 an action
Intercom	$29, $85 or $132 per seat a month, billed annually	Fin AI Agent at $0.99 per outcome, charged once per conversation
Atlassian Cloud	Per user, by plan	From 3 December: Rovo credits, automation steps and AI agent resolutions metered beyond plan allowances

Every one of these is a hybrid. The seat still pays for access. A second meter pays for the work that AI does, or for heavy use. GitHub's announcement of its move to usage-based billing, published on 27 April 2026 and effective from 1 June 2026, shows the design choice clearly: code completions and next-edit suggestions stay included and do not consume credits, while chat, agents and code review draw on a monthly credit pool. The cheap, frequent action stays in the seat. The expensive action gets a meter.

Atlassian's third meter, resolutions completed by AI agents, is an outcome price inside a seat-priced product. Its announcement on 1 September 2026 describes these as outcome-based resolutions where an AI agent autonomously completes a request. Intercom's $0.99 per outcome is the same idea sold on its own.

Why vendors are adding meters to seats

The reason is cost. A seat has always cost a software company very little to serve: storage, some compute, support. An AI feature costs real money every time someone uses it, because each request runs a model. A heavy user of an AI assistant can cost a vendor many times what a light user costs, while paying the same seat price. Under a pure seat model, the vendor either sets the seat price high enough to cover heavy users, which overcharges everyone else, or accepts a loss on the heaviest users.

Trae's change shows the pressure from the other side. Trae already charged by token use against a monthly dollar allowance. Its change leaves the price alone and gives existing subscribers less usage for it. When the unit cost of the service is visible, the allowance is the lever that moves.

It is also why so many of these changes switch billing on by default. A meter that the customer has to turn on is one that most customers never turn on. Microsoft's decision to enable Copilot Business usage billing by default for new CSP licences from November, with a preset monthly limit, and Atlassian's decision to bill overages with alerts before limits are reached, both put the meter in the path of normal use. Buyers should read that as a signal that metered spend is expected to grow.

What the filings say: seats still carry the revenue

It is easy to read the pricing announcements and conclude that seat pricing is finished. The vendors' own results say otherwise, at least for now.

Salesforce's results for the second quarter of its 2027 financial year, published on 26 August 2026, reported subscription and support revenue of $10.8 billion for the quarter, up 12%. Its Agentforce annual recurring revenue exceeded $1.5 billion, up more than 240% year on year, and Agentforce and Data 360 together reached nearly $3.9 billion of annual recurring revenue. Salesforce also reported 3.2 billion "agentic work units" delivered in the quarter, up 97% on the previous quarter. The AI line is growing very fast. It is also still small next to the core business: $1.5 billion of annual recurring revenue against a subscription business running at more than $40 billion a year. And Salesforce says the fastest-growing bookings were Agentforce One Edition and Agentforce for Apps, which it describes as premium SKUs anchored in Sales and Service. In other words, much of the AI revenue is being sold as a more expensive edition of a seat-based product.

Atlassian's shareholder letter for the fourth quarter of its 2026 financial year, published on 6 August 2026, reported revenue of $1.8 billion, up 28%, cloud revenue of $1.2 billion, up 31%, and subscription annual recurring revenue of $6.6 billion, up 23%. The letter attributes growth to customers adding new users and teams, and to customers upgrading to higher-value editions. It also notes that customers on its Teamwork Collection bundle use more than twice as many AI credits per user as customers on standalone products. The credit meter is being measured and reported to investors, but the growth story is still told in users and editions.

The honest summary is that the pricing model is shifting at the margin, where AI costs are highest, while the seat remains the core of how established vendors are paid. For buyers, that means a seat bill plus a usage bill, not a usage bill instead of a seat bill.

Free tiers, open source and "free forever"

At the other end, free software is as common as ever, and serves a clear commercial purpose.

HubSpot's pricing page offers free tools for up to two users and 1,000 contacts, with no credit card required and no time limit. GitLab's Free plan allows five users per top-level group, 400 compute minutes a month and 10 GiB of storage. These are acquisition channels: a free tier costs the vendor little to serve, puts the product inside the customer's business, and turns into paid seats when the team grows past the limit.

AI changes the arithmetic of free tiers, because AI features are expensive to serve. GitLab's page says AI features on the Free plan are available through the purchase of its credits add-on. That pattern, free core product and paid AI, is likely to spread, because a free tier with unlimited AI would lose money on every active user.

Open-source software sits in a different place. The software itself is free to use under its licence, and the companies behind many open-source products make money from hosting, support, enterprise features or services. For a buyer, "free" open source is often the cheapest licence and the most expensive in staff time. For a vendor, open source is a distribution strategy that works best when the paid offer solves a problem the free one does not: running it reliably, securing it, or meeting an auditor's requirements.

"Free forever" plans deserve a specific caution. They are free until the vendor changes the plan, and vendors do change them. The Trae case shows an allowance cut for paying customers at an unchanged price; free-plan limits can move the same way. Treat "free forever" as the current policy, not a promise.

Marketplaces as a sales and billing channel

Cloud marketplaces are increasingly how business software is bought, because the purchase goes on an invoice the customer already pays. Microsoft's documentation, updated on 22 July 2026, says it charges a 3% standard store service fee on offers sold through Microsoft Marketplace. For software-as-a-service offers it supports flat-rate, per-user and usage-based plans, and a metering service that lets a vendor bill up to 30 custom meter dimensions per offer, such as tickets or emails processed, on top of a contract's included entitlement.

Two details in that documentation matter to both sides. First, offers billed by consumption after use are not eligible for refunds. Second, a vendor is responsible for reporting usage accurately, and Microsoft bills the customer on what the vendor reports. A buyer paying a metered marketplace offer is trusting the vendor's meter.

For a small vendor, a marketplace listing can shorten procurement with large customers who have committed cloud spend or prefer a single bill. The fee is modest. The cost is the integration work and the fact that the marketplace, not you, now sits between you and the invoice.

A monetisation decision method for a small software company

If you run a small software company and are deciding how to charge, work through these steps in order. Write the answers down.

1. Find your cost driver. List what costs you money per customer each month: hosting, model inference, third-party APIs, support time. If one of these grows with use and can vary tenfold between customers, you need a usage component. If costs are flat per customer, you probably do not.
2. Find your value metric. Ask five customers what they would measure to decide whether your product was worth it. Users? Projects? Documents signed? Tickets resolved? The best price metric is one the customer already counts, that grows as they get more value, and that you can measure without argument.
3. Match the two. If the cost driver and the value metric are the same thing, price on it. If they differ, charge a base price on the value metric and put a usage allowance on the cost driver, as GitHub and GitLab do.
4. Keep the frequent action in the base price. Customers resent being charged for the thing they do most. Put the everyday action inside the seat or plan and meter the expensive, occasional one.
5. Consider outcome pricing only if the outcome is unambiguous. A resolved conversation can be defined. "Improved productivity" cannot. If you and a customer could argue about whether an outcome happened, do not price on it.
6. Design the free tier around a limit that tracks growth. Users, projects or records work. Time limits and feature crippling work less well. Keep expensive AI features out of free tiers unless you have budgeted for them.
7. Publish your price and date it. Our catalogue shows how quickly prices become stale in other people's listings. Put a date on your pricing page, and give generous notice of changes. Our news desk's median of 31 days is a floor, not a target. Existing customers remember a short-notice change long after they forget the price.
8. Give customers a spending control. Any metered component needs a cap, an alert before the cap, and a usage report. Vendors that provide these, as Atlassian and GitHub now say they do, remove the main objection to usage pricing.
9. Review the model every six months against actual cost and usage per customer. If a small group of customers costs you more than they pay, change the allowance or the meter, with notice, before you change the headline price.

A buyer's checklist for reading a usage-priced contract

Before you sign or renew anything with a meter, a credit or an outcome price, get written answers to these questions.

The unit

• What exactly is counted? Tokens, actions, conversations, steps, credits?
• If it is a credit, what does each action consume, and can the vendor change that rate during the term?
• Which common actions are free and stay free?

The allowance

• How much usage is included per user or per plan, and is it pooled across the organisation or per person?
• Does unused allowance roll over, or expire each month?
• Is any of the allowance promotional, and when does the promotion end? GitHub's and GitLab's pages both show promotional allowances.

The overage

• What does usage beyond the allowance cost, per unit, in your currency?
• Is overage billing on or off by default? Who in your organisation can change it?
• Is there a monthly limit, who sets it, and what happens when it is reached: does the service stop, slow down, or continue and bill?

Visibility and control

• Can you see usage by user, team and feature, in close to real time?
• Are there alerts before a limit is reached, and who receives them?
• Can you set limits per user or per team to stop one automation or agent consuming the whole pool?

Change and notice

• How much notice must the vendor give before changing prices, allowances or credit rates? Get the number of days in the contract.
• Can the vendor switch a new metered feature on by default? Can you require it to stay off until you enable it?
• If the vendor changes the model mid-term, can you leave without penalty?

Disputes and refunds

• How is usage measured, and can you audit it?
• Are consumption charges refundable if a meter was wrong? On Microsoft Marketplace, consumption-billed offers are not.
• For outcome pricing: who decides that the outcome happened, and can you challenge it?

Your own homework

• Measure your current usage before the meter starts. Atlassian customers have until 3 December to count their automation steps and agent resolutions.
• Model three scenarios: current use, double, and a runaway automation or agent.
• Assign one person to watch the usage report every month.

A seat price tells you what software costs. A usage price tells you what it costs if you use it the way you did last month. The checklist is how you find out the difference before the invoice does.

For how buyers weigh price against everything else when choosing new software, see what buyers look for in new software. For the new AI models that sit behind many of these meters, our AI models board lists recent releases.

Sources

• TrustList news desk, September 2026: 28 vendor changes with notice and effective dates; notice periods; default-on changes. TrustList, counted September 2026.
• TrustList catalogue check: 5,092 software listings with undated 2020 prices; 6 of 30 sampled still matching vendors' pages. TrustList, September 2026.
• Atlassian changes its prices twice this autumn: list prices on 13 October, metered AI from 3 December. TrustList, 21 September 2026.
• Microsoft ends Microsoft 365 E5, E7 and Copilot CSP promotions on 30 September: promotions ending, and usage billing on by default for new Copilot Business CSP licences from 2 November. TrustList, 23 September 2026.
• Trae cuts usage allowances for existing Pro+ and Ultra subscribers: allowance cuts at unchanged prices. TrustList, 22 September 2026.
• Atlassian's usage-based pricing: AI value with predictability and control: the three meters, 3 December billing date, what stays free. Atlassian, 1 September 2026.
• GitHub Copilot is moving to usage-based billing: 1 June 2026 change, Business and Enterprise prices and credits, what does not consume credits. GitHub, 27 April 2026.
• GitHub Copilot plans and pricing: plan prices and the $0.01 credit value. GitHub (read 25 September 2026).
• GitLab pricing: Free plan limits, Premium price, GitLab Credits. GitLab (read 25 September 2026).
• Agentforce pricing: Flex Credits, per-conversation and per-user prices. Salesforce (read 25 September 2026).
• Intercom pricing: seat prices and Fin at $0.99 per outcome. Intercom (read 25 September 2026).
• HubSpot CRM pricing: free tools for up to two users and 1,000 contacts. HubSpot (read 25 September 2026).
• Salesforce delivers record second quarter fiscal 2027 results: subscription and support revenue, Agentforce ARR, agentic work units, premium edition bookings. Salesforce, 26 August 2026.
• Atlassian Q4 FY26 shareholder letter: revenue, cloud revenue, subscription ARR, AI credits per user, edition upgrades. Atlassian (Form 8-K exhibit), 6 August 2026.
• Microsoft Marketplace transact capabilities: 3% store service fee, pricing models, metering dimensions, refunds on consumption offers. Microsoft Learn, 22 July 2026.