How to Buy Software When AI Breaks Pricing

AI has broken the per-seat price, and buyers now hold more leverage than they have in two decades. The catch is that the replacements (credits, consumption meters and outcome fees) can cost more than seats ever did if you sign them the old way. This is my buyer's guide: how I rank pricing models, what fourteen vendors charge today, the traps inside consumption deals, and a five-question test to run before you sign.

GrackerAI is my company. SSOJet and MojoAuth are GrackerAI customers.

The year the seat lost its value

On October 1, 2026, HubSpot changed how it charges new customers in EMEA, according to HubSpot partners. Separate hubs, each with its own tiers, became one edition with two kinds of seats and a monthly pool of credits. Under HubSpot's published credit rates, a marketing email send costs 15 credits, a workflow action costs 3 and a resolved customer agent conversation costs 50. Rates fall at higher volumes.

That is not a pricing tweak. It is a public admission that a login is no longer a good proxy for value.

I made the market case in AI didn't kill SaaS, it killed the seat. The early-2026 selloff erased about $1 trillion in software market value, and the vendors winning investors back are the ones moving off the seat. I made the vendor case for outcome fees in the pricing model that proves you're actually AI-native. This post sits on the other side of the table: what to buy, and what to sign.

The shift shows up in the survey data. Kyle Poyar's 2025 State of B2B Monetization report, based on more than 240 software and AI companies, found seat-based pricing down from 21% to 15% in a year. Hybrid pricing rose from 27% to 41%. Gartner expects at least 40% of enterprise SaaS spend to shift toward usage-, agent- or outcome-based pricing by 2030, as quoted in Deloitte's 2026 TMT Predictions.

Salesforce moved earlier. Agentforce launched at $2 per conversation, then Salesforce added Flex Credits in May 2025: $500 per 100,000 credits, with a standard agent action costing 20 credits, or 10 cents.

For buyers, this is the best negotiating window in two decades. Most procurement teams are still evaluating software the way they did in 2015.

Why the cost floor dropped out

Software has three cost layers, and AI is pressing on all of them at once. This section is my analysis from building and running SaaS for more than fifteen years.

Building got cheaper. When I built LoginRadius and scaled it to over 1 billion users, a meaningful feature meant a team, a quarter and a roadmap fight. Today a small team with coding agents ships in weeks what took us months. The marginal cost of a feature is collapsing, and so is the moat that came from simply having built it.

Deploying got cheaper. Managed cloud, serverless and edge platforms turned infrastructure into a line item that scales with use. A two-person company can run globally distributed software on a bill that would not have covered one server rack in 2012.

Running got cheaper. Support, onboarding, configuration and monitoring were the hidden labor inside every SaaS price. AI agents now handle a growing share of that work. When the vendor's cost to serve drops, a per-seat price that never moves becomes pure margin, and buyers can see it.

The result is a mismatch. Vendor costs fall with usage efficiency, but seat prices are tied to headcount. If AI agents do work that people used to do, a customer may need fewer seats while getting more value. Seat-based vendors either reprice or watch their revenue shrink as their customers get more productive.

That is why every serious vendor is moving. The question for buyers is whether the new model shares the savings with you or just finds a new way to meter you.

Five pricing models, ranked by how honest they are

Not every "usage-based" model is a good deal. I rank them by one test: does the unit you pay for move with the value you get?

Model	What you pay for	Who uses it	My verdict
Outcome-based	A finished result, such as a resolved ticket	Fidiora, Zendesk AI agents, Intercom Fin, Sierra	Best alignment, if the outcome is defined in the contract
Value-unit pricing	The thing that matters to the business: a connection, an active user, a tracked prompt	GrackerAI, SSOJet, WorkOS, MojoAuth, Clerk	Simple and fair; easy to forecast
Flat platform fee	Access, with your own infrastructure costs passed through	Mailazy	Cheapest at scale; you own the variable costs
Credits	A vendor-defined unit spent across many actions	Salesforce Flex Credits, HubSpot Credits, Microsoft Copilot Credits	Flexible, but the vendor sets the exchange rate
Per seat	A human login	Most legacy SaaS	Misaligned the moment AI does the work

Outcome-based pricing is the purest form. Zendesk moved its AI agents to outcome-based pricing in August 2024 and now lists $1.50 per committed automated resolution. Intercom charges $0.99 per Fin outcome. You pay when the job gets done.

Value-unit pricing is underrated. The vendor picks one unit the buyer already understands and charges for that, nothing else. It is easy to forecast, easy to compare and hard to game.

Credits are where I am most skeptical. Credits are flexible, and for mixed workloads they can be the right answer. But the vendor sets the exchange rate between credits and actions, and can change it. Too often, credits are usage pricing in nicer packaging.

Who already prices for the AI era

Here is how a mix of incumbents and newer vendors price today, based on their public pricing pages as of October 2026. Prices are list prices in USD unless noted.

Company	Category	What you pay for	Entry point	Free tier
GrackerAI	AEO/GEO visibility	Tracked prompts, AI engines and content volume by tier	$99/mo Starter ($79/mo billed annually); Enterprise quoted per brand	7-day trial
Intercom Fin	Support AI agent	Fin outcomes, no seats required	$0.99 per outcome	14-day trial
SSOJet	Enterprise SSO and SCIM	SSO and SCIM connections, not users	$59 per connection billed annually, falling to $35 at 50	Unlimited users, no SSO connections
Fidiora	AI support desk	Resolutions only; no seats, ticket or AI action fees	$0.69 per resolution, $0.45 above 1,000/mo	No credit card to start
Salesforce Agentforce	CRM AI agents	Flex Credits, per conversation, or per-user add-on	$500 per 100,000 credits; $125/user/mo add-on	100,000 credits for Enterprise Edition and above via Foundations
Clerk	Developer authentication	Monthly retained users who return 24+ hours after sign-up	$25/mo Pro ($20/mo billed annually); $0.02 per user above 50,000	50,000 retained users per app
Mailazy	Behavior-driven email	Products and team members; you bring your own sending provider	$49/mo Pro with events and automations	1,000 contacts, 5,000 emails/mo
Microsoft Copilot Studio	Agent platform	Copilot Credits consumed per answer and action	$200/mo for 25,000 credits; $0.01 per credit pay-as-you-go	Employee use included with Microsoft 365 Copilot licenses
MojoAuth	B2B and B2C authentication	Monthly active users	$150/mo at 2,500 MAUs; $0.06 per MAU overage	Up to 25,000 MAUs
Zendesk	Support AI agents	Automated resolutions	$1.50 per committed resolution; $2.00 pay-as-you-go	Starter usage level included
WorkOS	Enterprise SSO and user management	SSO connections; users free up to 1 million	$125 per connection, falling to $65 at 51 to 100	1 million MAUs
HubSpot (EMEA, new customers)	CRM and marketing	Core or Front Office seats plus credits	€20 Core seat on Starter, with 5,000 credits a month	Free tools with limits
Resend	Developer email	Monthly email volume plus automation runs	$20/mo for 50,000 emails; $0.90 per extra 1,000	3,000 emails/mo (100 a day)
Sierra	Enterprise customer AI agents	Successful outcomes, such as a resolved conversation	Not public; custom contracts	None

GrackerAI: price the visibility, not the headcount

GrackerAI is my company, so read this with that in mind. We price AEO (answer engine optimization) and GEO (generative engine optimization) by what a marketing team is actually trying to measure. That means how many prompts we track, across how many AI engines, how often, and how much optimized content we produce. Starter is $99 a month for 100 prompts across three engines. Enterprise is quoted per brand. Seats come bundled with each tier instead of being sold per head, because charging a CMO extra for a teammate to read a visibility report makes no sense.

SSOJet: charge for the connection, not the user

Enterprise SSO is the classic seat-pricing trap. A B2B SaaS company signs a 5,000-employee customer and suddenly its identity bill jumps. SSOJet prices by SSO and SCIM connection instead, with unlimited users on every plan. A connection is the unit that maps to a closed enterprise deal, so cost tracks revenue. I priced that link in the identity tax: what auth costs per closed deal. On annual billing, the per-connection rate drops from $59 to $35 as you grow.

MojoAuth: pay only for people who log in

MojoAuth charges per monthly active user, defined in its pricing FAQ as someone who authenticates at least once in a calendar month. Dormant accounts cost nothing. The free tier covers 25,000 MAUs, which is enough for most early-stage products to reach real traction before paying.

Mailazy: unbundle the email bill

Most email platforms charge on contacts or sends, so your bill grows as your list grows, whether or not the emails work. Mailazy charges a flat fee for products and team members, and you send through your own provider, such as Amazon SES or Postmark. The pricing page's own calculator puts 50,000 emails at about $54 a month: $49 for Pro plus about $5 in provider fees at an assumed $0.10 per 1,000 emails. Event-driven automations are included from Pro. I compare it with Resend and three others in the top behavioral onboarding email tools of 2026.

Salesforce: three models, one decision

Salesforce now offers per-conversation pricing at $2, Flex Credits at 10 cents per standard action, and a $125 per user per month add-on for employee-facing agents. It also sells an Agentic Enterprise License Agreement: a fixed price over a fixed term, with unlimited use of consumption services and no overage charges. Choice is good. But you need your own usage model to pick correctly, because the break-even points are not obvious.

HubSpot: the incumbent that blinked

HubSpot's EMEA model is the clearest sign that seats alone no longer work, even for a company built on them. Seats remain for people, and credits now cover system activity like emails, workflows and agents. Existing customers stay on their current model for now, and HubSpot has not announced timing for other regions.

Intercom Fin: the per-outcome benchmark

Intercom prices its Fin AI agent at $0.99 per outcome, with no seats required. That number has become the reference point for the whole support category. Read the definition before you sign, though. An outcome counts when a customer confirms the issue is resolved, but also when they simply stop asking for help, or when Fin completes a workflow, including a handoff. Model your bill on all three.

Clerk: count the users who come back

Clerk charges by monthly retained users, meaning people who return at least 24 hours after signing up. Drive-by sign-ups and bots that never come back cost nothing. The free tier covers 50,000 retained users per app, and the rate per user drops as volume grows. It is a smart twist on MAU pricing for consumer and developer products.

Microsoft Copilot Studio: credits with a visible rate card

Microsoft meters agents in Copilot Credits. Its billing rates are 2 credits for a generative answer, 5 for an agent action and 10 for grounding in tenant data. Packs cost $200 for 25,000 credits, or $0.01 per credit pay-as-you-go. The rate card is public, which I respect. But costs stack inside a single interaction, so model a realistic conversation before you commit.

WorkOS: free users, paid connections

WorkOS gives away user management up to 1 million monthly active users and charges for enterprise SSO connections, from $125 each down to $65 at 51 to 100 connections. Like SSOJet, it ties cost to the moment a B2B customer brings enterprise SSO, which usually means a bigger contract. The difference between the two is mostly price per connection, so compare at your expected connection count. My buyer-stage framework for Auth0, Okta, Stytch, WorkOS and SSOJet walks through that comparison.

Resend: email priced for developers

Resend starts free at 3,000 emails a month and charges $20 for 50,000, with automation runs metered separately above 10,000 a month. It shows how far the cost of sending has fallen when the vendor skips the contact-based pricing that marketing suites rely on.

Sierra: outcome pricing at the enterprise end

Sierra charges when its agent completes a task, such as resolving a conversation or saving a cancellation. Unresolved conversations typically are not billed. None of the pricing is public. That is the trade-off at the top of the market: aligned incentives, but you need strong negotiators and a precise definition of success.

Fidiora: a support desk that only bills when the customer is helped

Fidiora takes outcome pricing all the way down to the help desk itself. There are no seat fees and no charge per ticket or AI action. You pay per resolution, defined as an issue that was actually closed, not an abandoned chat or a silent timeout. The first 1,000 resolutions each month cost $0.69 each and every one after that costs $0.45, so 3,000 resolutions come to $1,590. Human handoffs are free, and a monthly spend cap keeps the bill from running past a number you set.

This is the model I expect more support tools to copy. It is published, easy to forecast and cheaper per resolution than the better-known names. It also puts the definition of a resolution on the pricing page instead of leaving it to the contract.

How to buy software now: my five-question test

I run every software decision through five questions. If a vendor cannot answer them clearly, that tells you more than any demo.

1. What is the value unit, and does the price follow it? Name the one thing you are buying: a resolved ticket, a closed enterprise SSO deal, an active user, a tracked AI prompt. Then check whether the bill moves with that thing. If you are paying for logins while an agent does the work, you are funding the vendor's legacy model.
2. Can I forecast next year's bill in one spreadsheet row? Ask for the pricing formula, not a quote. Model your expected growth and a 3x spike. If you cannot predict the bill without a vendor rep, the model is too complex, or deliberately opaque.
3. Should I build this instead? AI has lowered the cost of building, so the build-versus-buy line has moved. My rule: buy what is regulated, security-critical or needs constant upkeep, such as authentication, SSO and email deliverability. Build the thin workflow glue that is specific to your business. Never build identity yourself to save a subscription; I explain why in build vs buy auth is the wrong question.
4. Who keeps the savings when the vendor gets more efficient? Vendors' AI costs are falling. Ask how that shows up in your price over a multi-year term. A vendor that passes efficiency through on its own is a partner. One that needs to be pushed is a supplier.
5. How easily can I leave? Look for short terms, monthly billing, data export and standard protocols such as SAML, OIDC and SCIM. In a market repricing this fast, flexibility is worth more than a multi-year discount.

Two practical habits make this work. First, keep a live inventory of every tool, its pricing unit and its actual usage. In Flexera's 2026 State of ITAM research, only 36% of organizations report complete visibility into their IT estate. Second, start renewal conversations six to nine months early, because that is when you hold the leverage.

The traps inside consumption pricing

Usage pricing is not automatically cheaper. It moves risk, and if you are not careful, it moves it onto you.

The AI tax. Tropic uses this term for the 20 to 37% renewal uplift that comes from bundling AI features into mandatory packages. Forced SKU migration is the usual delivery mechanism. You did not ask for the AI features, but the renewal price assumes you did.

The hybrid double charge. Bain's review of AI pricing found that most software companies moving beyond seats are layering new meters on top of them rather than replacing them. That is not a new model. It is the old one with an extra meter.

The expiring credit. HubSpot's included credits reset monthly and do not roll over. Salesforce Flex Credits do not roll into the next subscription term. Many credit systems work this way, so unused credits are simply lost.

The shifting exchange rate. If a vendor can change how many credits an action costs, your contract price is not really fixed.

The vague outcome. If "resolution" is not defined in writing, the vendor's dashboard decides what you owe.

What to put in the contract:

• A cap on annual price increases, ideally 3 to 5% or tied to CPI
• A lock on your current SKU so you cannot be force-migrated mid-term
• A carve-out so new AI features do not trigger automatic uplifts
• A hard consumption ceiling that throttles usage instead of billing overages
• A written definition of every billable outcome or credit action
• A rate card freeze for credit-to-action conversions for the full term
• A mid-term review at 12 to 18 months to switch models if usage changes
• Full data export in a standard format at no cost

My take: where software pricing goes next

These are my predictions, not reported facts.

Seats survive, but only for humans doing human work. HubSpot's model points the way: seats for people, a meter for machines. I expect most large vendors to land on this hybrid within two years. The ones that keep pure seat pricing will lose customers who are replacing tasks with agents.

Specialists will out-price suites. A focused vendor with a clear value unit can price below a suite's bundled line item and still make money, because AI has cut its cost to build and serve. Expect more buyers to unbundle: an identity vendor for SSO, a separate one for customer login, a lean email tool on their own sending provider, an AEO platform for AI visibility.

Transparency becomes a sales advantage. When a vendor publishes its full pricing, with the per-unit math, it signals that its price can survive comparison. In my experience, buyers increasingly shortlist vendors whose pricing they can understand without a call. If you sell software too, run your own numbers through the SaaS metrics calculator before you copy any of these models.

Procurement becomes a FinOps function. Teams that already manage cloud spend by usage will take over SaaS spend too. Software buying will look less like signing a contract and more like managing a meter.

The vendors who win this transition will be the ones who make their price boring: easy to predict, tied to something you value, and lower each year as their costs fall.

When did you last check whether the unit you pay for still matches the value you get?

FAQ

Why is per-seat SaaS pricing declining in 2026?

AI agents now do work that used to require human users, so seat counts no longer track value. In Kyle Poyar's 2025 State of B2B Monetization survey, seat-based pricing fell from 21% to 15% of companies in a year, while hybrid pricing rose from 27% to 41%.

What is consumption-based software pricing?

You pay for what you use, such as active users, connections, actions or outcomes, instead of a fixed fee per login. It can be cheaper and fairer, but only when the unit is clearly defined and the vendor cannot change its rate mid-contract.

How did Salesforce and HubSpot change their pricing?

Salesforce prices Agentforce through Flex Credits at $500 per 100,000 credits, per conversation at $2, or a $125 per user monthly add-on. HubSpot partners report that, from October 1, 2026, new EMEA customers buy one edition with seats for people and credits for system activity.

Should companies build software with AI instead of buying it?

Build the workflow glue specific to your business. Buy anything regulated, security-critical or maintenance-heavy, such as authentication, enterprise SSO and email deliverability. AI lowers build costs, but not the cost of getting security wrong.

What should I negotiate in a usage-based software contract?

Cap annual increases, lock your SKU, freeze credit conversion rates, set a consumption ceiling that throttles instead of billing overages, define billable outcomes in writing, and add a mid-term review.

Sources

• HubSpot: Limited Release pricing (seats and credits)
• Salesforce: new flexible Agentforce pricing (May 15, 2025)
• Salesforce: Agentforce pricing
• Salesforce: flexible buying models
• Growth Unhinged: 2025 State of B2B Monetization
• Deloitte: TMT Predictions 2026, SaaS and AI agents (citing Gartner)
• Zendesk: outcome-based pricing for AI agents (August 28, 2024)
• Zendesk pricing
• Tropic: the AI tax in software pricing
• Bain: AI pricing, a reality check on effort, usage and outcomes
• Flexera: 2026 State of ITAM and SaaS sprawl
• Microsoft Learn: Copilot Studio billing rates
• Microsoft: Copilot Studio licensing guidance
• Sierra: outcome-based pricing for AI agents
• GrackerAI pricing
• Intercom pricing
• Fidiora pricing
• SSOJet pricing
• Clerk pricing
• MojoAuth pricing
• WorkOS pricing
• Mailazy pricing
• Resend pricing