42 AI agent pricing trends shaping the future of software monetization

Market data and pricing benchmarks revealing how AI agents are transforming billing models, cost structures, and revenue design for software companies

AI agents have moved from experimental pilots to permanent line items in enterprise software budgets, and pricing them has become one of the most complex challenges in modern software monetization. Unlike traditional SaaS products with predictable seat-based models, AI agents consume variable compute resources, process unpredictable token volumes, and increasingly bill against outcomes that resist simple per-unit measurement. Companies building AI-powered products need usage-based billing infrastructure capable of handling this complexity while maintaining the financial accuracy that enterprise customers demand.

Key takeaways

• Hybrid pricing is now the default, not the exception: 95% of the 80 AI agent companies in Orb's 2026 study use hybrid pricing, up from 92.4% a year earlier and effectively at saturation
• Usage-based pricing is near-universal: 91.3% of AI agent companies meter and charge for usage, while 71.3% still carry a subscription component
• Pure outcome pricing remains rare and is not growing: 3.8% of the sample uses outcome-based pricing, down slightly from 4.5% in 2025, even as vendors publish headline per-resolution rates
• Published per-resolution prices sit just under $1: Gorgias charges $0.90 per resolved interaction, $1 on Starter, and Intercom charges $0.99 for a resolution outcome, while Zendesk meters verified automated resolutions without publishing a rate
• Enterprise budget mix is shifting: Deloitte, citing Gartner, projects that by 2030 at least 40% of enterprise SaaS spend could move toward usage-based, agent-based, or outcome-based models
• Pricing is iterated, not set: 92% of AI businesses that charge for usage subsequently changed their pricing, making pricing agility an infrastructure requirement

How AI agent companies actually price today

The clearest picture of current practice comes from Orb's 2026 study of 80 AI agent companies, supplemented by Deloitte's 2026 SaaS predictions and Stripe's 2025 pricing research. Orb analyzed 80 AI agent companies at a 90% confidence level. Together the sources show a market that has already consolidated around hybrid monetization, with the year-over-year numbers now moving only incrementally.

1. 95% of AI agent companies use hybrid pricing

Orb's 2026 research found that 95% of the 80 AI agent companies studied combine multiple pricing mechanisms rather than relying on a single model. Hybrid is no longer an emerging strategy; it is the operating norm, and single-model billing systems are now the constraint rather than the simplification.

2. 91.3% of AI agent companies use usage-based pricing

Usage-based pricing appears in 91.3% of the companies in Orb's 2026 sample, up from 83.3% in 2025. Consumption has effectively become the default variable component of AI agent monetization, which makes accurate, real-time usage metering a prerequisite rather than an optimization.

3. 71.3% of AI agent companies still charge a subscription

Subscriptions have not disappeared. 71.3% of the companies studied retain a subscription element, up from 62.1% in 2025, typically as a platform or access fee that anchors predictable revenue underneath variable consumption.

4. 94.7% of subscription-charging companies also charge for usage

Among companies that use subscriptions, 94.7% pair them with usage-based pricing. The standalone seat subscription is close to extinct in the AI agent category, replaced by a base plus consumption structure.

5. Outcome-based pricing is rare and slipped year over year

Despite the volume of commentary, outcome-based pricing sits at just 3.8% in Orb's 2026 sample, down slightly from 4.5% in 2025. Orb attributes the gap to instrumentation rather than appetite: defining, measuring, and attributing outcomes reliably is hard, and most legacy billing infrastructure was not built for it.

6. Hybrid adoption is now saturating rather than climbing

Within Orb's own year-over-year series, hybrid pricing moved from 92.4% in 2025 to 95% in 2026, a small step because the model was already near saturation. Stripe's 2025 research found 56% of AI company leaders blending subscription and usage-based fees across a broader population of AI companies, so the two figures are not points on one line, and the 56% number should no longer be cited as the current AI agent benchmark.

7. 83% of AI-native SaaS companies offered usage-based pricing

Deloitte's 2026 technology predictions cite survey data showing that 83% of AI-native SaaS companies offered usage-based pricing. Companies born after the shift are not retrofitting consumption billing; they start with it.

8. At least 40% of enterprise SaaS spend could shift to usage, agent, or outcome models by 2030

Deloitte, citing Gartner, projects that by 2030 at least 40% of enterprise SaaS spending could move toward usage-based, agent-based, or outcome-based models. This is the single most consequential number for revenue design teams, because it reprices a large share of the addressable budget.

Billable units and pricing metrics

The hardest question in AI agent pricing is not how much to charge but what to charge for. The market has converged on a wider set of billable units than traditional SaaS ever required.

9. AI usage is billed across actions, compute, API calls, tokens, duration, or combinations

Deloitte documents that AI monetization now spans actions and tasks, compute, API calls, tokens, duration, or combinations of those measures. A single vendor may run several of these simultaneously across different products and customer tiers.

10. Agent usage maps naturally to discrete events

Orb's 2025 analysis of AI agent pricing notes that agent activity is naturally represented by events such as calls, tokens, jobs, documents, or leads. Choosing the event is the pricing decision; everything downstream, from rate cards to invoices, inherits that choice.

11. Effort-based pricing is emerging as a distinct model

Orb's 2026 research identifies effort-based pricing, where compute time, number of steps, or task complexity become the value and usage measure. This sits between raw consumption and pure outcomes, and it fits agentic workloads where a single request may trigger long, variable execution chains.

12. Some vendors explicitly reject seats, messages, and tokens as billable units

Gorgias states that its AI Agent is not billed by seats, messages, or tokens, but by fully resolved interactions. Naming the metrics it does not bill is itself a positioning move, signaling that the customer pays only for completed work.

13. Credit systems translate heterogeneous actions into a single unit price

Salesforce sells Flex Credits at $500 per 100,000 credits, with one Agentforce action consuming 20 credits, or $0.10. Credits let a vendor price wildly different actions on one scale, which is why prepaid credit and token packages have become standard components of AI rate cards.

14. Per-agent and digital worker flat pricing persists as one model in a wider mix

Deloitte documents flat-fee, per-agent, or digital worker pricing as one of several emerging models. It has not disappeared, but it now typically appears as one rail in a hybrid structure rather than as the whole commercial model, though no market-wide monthly benchmark range for it is well established.

Outcome pricing benchmarks from first-party vendor documentation

Where outcome pricing does exist, the published mechanics are more precise than the summary coverage suggests. These figures come from vendor documentation rather than secondary aggregation.

15. Intercom charges $0.99 for a resolution outcome

Intercom's Fin documentation prices a resolution at $0.99, alongside procedure handoffs and disqualifications, for Fin over chat and email. The unit is the outcome, not the conversation, the message, or the token.

16. Different outcomes carry different unit values

Intercom does not price all outcomes identically. Its documentation lists $0.99 for several outcome types and $9.99 for qualification, a roughly tenfold spread that reflects differing commercial value. Outcome pricing is therefore a rate card, not a single number.

17. Vendors cap billable outcomes per conversation

Intercom charges at most one outcome per conversation. Caps of this kind protect customers from compounding charges inside a single interaction, and they have to be enforced at the metering layer rather than at invoice time.

18. No charge when an eligible outcome is not achieved

Intercom does not charge when an eligible outcome is not achieved. That single rule is what makes outcome pricing commercially credible, and it forces the vendor to define success precisely enough to be auditable.

19. Gorgias charges $0.90 per resolved interaction, $1 on Starter

Gorgias prices its AI Agent at $0.90 per resolved interaction on most plans and $1 on Starter. The billable unit is a resolved interaction rather than a conversation, and that distinction matters for how usage is metered.

20. Zendesk restructured packaging around automated resolution tiers

Zendesk introduced automated resolution tiers on May 18, 2026. Automated resolutions are the unit used to calculate and bill AI agent usage, accounts are credited with a resolution allowance that those resolutions draw down, and the tiers are designed so customers do not overpay for straightforward tasks. This is a shift from the earlier framing of a flat overage bolted onto seat pricing.

21. Only independently verified resolutions draw down the allowance

Under Zendesk's tiers, an assisted escalation and a contained resolution do not count against the resolution allowance. A verified resolution is counted only after a 72-hour window with no customer follow-up, once a large language model evaluates the conversation and confirms the request was satisfactorily resolved. Billing gated on automated verification is a meaningful step beyond charging for any contained conversation.

22. Salesforce moved from a single conversation metric to multiple consumption rails

Salesforce's $2 per conversation metric remains available, but its 2025 packaging update introduced Flex Credits and multiple consumption and buying models. The trend is not the $2 price point; it is the deliberate move from one metric to a portfolio of consumption rails within a single product.

Unit economics pressure on AI pricing

Outcome and resolution pricing only works if the underlying cost curve cooperates. Analyst projections suggest it may not, indefinitely.

23. Gartner projects GenAI customer service cost per resolution will exceed $3 by 2030

Gartner predicts that generative AI cost per resolution for customer service will exceed offshore human agent costs by 2030, with cost per resolution exceeding $3, higher than many B2C offshore human agents. Vendors publishing sub-$1 resolution prices today are pricing against a cost base that may not stay where it is.

24. Cost-savings framing is weakening as a pricing rationale

Gartner cautions that returns on AI customer service investments are far from guaranteed and that full automation will be prohibitively expensive for most organizations, expecting leading organizations to use AI to drive customer engagement rather than to cut costs. For vendors, that erodes deflection-and-savings as the anchor for a price and shifts the justification toward engagement, retention, and lifetime value.

Pricing iteration as an operating discipline

The evidence is consistent across sources: AI pricing set at launch does not survive contact with real usage.

25. 92% of AI businesses charging for usage have changed their pricing

Stripe found that among AI businesses charging for usage, 92% subsequently changed their pricing. Repricing is the expected case, not the exception, which makes the cost and latency of a pricing change a competitive variable.

26. AI pricing frequently requires iteration after launch

Stripe's pricing framework treats post-launch iteration as a structural feature of AI products rather than a sign of a mistake, because early usage data is the strongest input available for calibrating rates against real consumption patterns.

27. Pricing is becoming a continuously iterated system, not a one-time packaging decision

Orb's 2026 research frames AI agent pricing as an ongoing system rather than a single packaging exercise. Companies that can version, simulate, and roll out pricing changes continuously hold a structural advantage over those that treat pricing as a launch artifact.

28. Execution is constrained by technical complexity, operational friction, and pricing knowledge gaps

Orb's 2026 study identifies technical complexity, operational friction, and pricing knowledge gaps as the practical limits on pricing execution. The bottleneck is rarely the strategy; it is the systems and workflows required to run it.

29. Pricing is simulated against real usage before rollout

Orb's 2025 analysis emphasizes testing pricing against actual usage data before rollout. Simulating a proposed rate card against historical events turns repricing from a revenue gamble into a modeled decision.

Commercial guardrails and contract mechanics

Variable consumption creates variable bills, and enterprise buyers will not accept unbounded exposure. Guardrails have become part of the pricing model itself.

30. Caps, pooling, commitments, credits, and overage terms manage variable spend

Orb identifies caps, pooling across teams, seasonal overage protections, and credit bundles that roll over as the mechanisms that make consumption pricing acceptable to buyers. These are commercial terms with billing consequences, and each one adds an aggregation rule the billing system must enforce.

31. Hybrid structures pair a base fee with variable usage

The dominant structure combines a base platform fee with variable usage charges. The base fee funds predictable delivery and gives finance a revenue floor; the variable component captures expansion as consumption grows.

32. Billable outcomes and attribution rules are defined to be auditable by the customer

Orb stresses defining billable outcomes and attribution rules precisely enough for customers to verify them. When the invoice line is an outcome rather than a unit of input, the customer's ability to reconstruct the charge from event history becomes a commercial requirement.

33. Prepaid credits and token packages formalize committed consumption

Prepaid credits and token packages let buyers commit budget in advance and give vendors cash and forecastability. Managing drawdown, expiry, and top-ups correctly is a billing engineering problem, not a spreadsheet exercise.

34. Real-time spend monitoring and configurable thresholds are now table stakes

Spend controls with real-time monitoring, configurable thresholds, alerts, and workflow triggers address a leading buyer objection to consumption pricing: bill shock. Vendors that cannot show a customer their spend in real time struggle to sell variable pricing into procurement.

35. Threshold billing converts consumption into predictable invoicing events

Threshold billing invoices when accrued usage crosses a defined level rather than waiting for a calendar boundary. For high-velocity AI consumption, this reduces credit exposure and smooths cash collection.

36. Dimensional pricing varies rates by model, region, and feature set

AI rate cards increasingly vary by dimension: model version, region, and feature set. Dimensional pricing is what allows a vendor to pass through materially different underlying costs without maintaining a separate product for each combination.

Metering, revenue recognition, and finance operations

Pricing strategy is only as good as the system that executes it. In AI agent monetization, the metering and finance layer has become part of the pricing model rather than a downstream reporting concern.

37. Raw usage event history underpins outcome attribution

A preserved raw usage event layer is what makes outcome billing defensible. Event-level history can support auditability, corrections, and finance workflows, and it is what lets a vendor reconstruct why a specific outcome was billed.

38. Backdating and backfill make retroactive corrections possible

Orb documents backdating price changes and backfilling missing events, with invoices and credit ledgers recalculating automatically rather than through manual correction events. In agentic systems where events arrive late or contracts are renegotiated after a service period, retroactive correction is a routine operation rather than an edge case.

39. Metrics defined in SQL and versioned pricing decouple rate changes from code releases

Defining usage metrics in SQL over raw usage events, with versioned pricing and logs, means a new billable metric can often be defined without a fresh instrumentation project. This is what turns the 92% repricing statistic from a crisis into a workflow.

40. ERP-aligned finance workflows connect usage data to the general ledger

Finance workflows covering ASC 606-aligned reporting, enterprise contract handling, and ERP integrations such as NetSuite close the loop between metered consumption and recognized revenue. Usage data that cannot reach the ledger cleanly is a month-end problem waiting to happen.

41. Outcome-based pricing creates new revenue recognition judgments

Deloitte's June 2026 analysis notes that outcome-based pricing introduces new ASC 606 judgments for SaaS companies, including how performance obligations and variable consideration are identified and measured. Pricing model selection is now an accounting decision as well as a commercial one.

42. Auditable pricing changes reduce engineering dependence in revenue design

Pricing versioning with auditable change history and simulation reduces engineering dependence when testing and operationalizing pricing changes. As pricing shifts from a static decision to a continuously iterated system, the ability of finance and pricing teams to act without a deployment cycle becomes a durable advantage.

Strategic implications for billing and revenue design

These trends converge on a small number of concrete requirements for companies building AI-powered products.

Hybrid pricing demands flexible infrastructure

With 95% of the AI agent companies in Orb's 2026 study using hybrid pricing and 94.7% of subscription users pairing subscriptions with usage, billing systems must support layered structures natively:

• Base platform fees combined with per-event, per-action, or per-outcome charges
• Volume tiers with distinct unit economics above included allowances
• Dimensional pricing across regions, models, and feature sets
• Spend controls and threshold alerts so customers can manage variable cost in real time

Outcome-based billing requires event-level tracking

Pricing per resolution, per resolved interaction, or per task depends on granular event history. Event-level records can support:

• Accurate outcome attribution and the rules that govern it, including per-conversation caps
• Retroactive adjustments and corrections through backdating and backfill
• Audit trails that enterprise customers can reconcile against their own systems
• Finance workflows and ASC 606-aligned reporting

Enterprise adoption accelerates billing complexity

As AI agents move from departmental tools to enterprise infrastructure, billing requirements expand:

• Multi-dimensional pricing across business units and product lines
• Contract-based commitments with usage overages, pooling, and caps
• Credit systems for prepaid consumption, including drawdown and expiry handling
• Integration with ERP and financial systems for recognition and close

Companies building AI solutions need billing infrastructure that evolves with pricing strategy rather than constraining it. When 92% of usage-charging AI businesses have already repriced, the ability to test new models, apply retroactive changes, and maintain financial accuracy across complex scenarios is not a back-office concern. It determines how fast a company can respond to a market that is still repricing itself.

Frequently asked questions

What are the most common pricing models for AI agents?

Hybrid pricing dominates. Orb's 2026 study of 80 AI agent companies found 95% use hybrid pricing, 91.3% use usage-based pricing, and 71.3% use subscriptions, with 94.7% of the subscription users pairing subscriptions with usage charges. Pure outcome-based pricing appears in only 3.8% of the sample, down slightly from 4.5% in 2025, which makes it the most discussed and least adopted model in the category.

How do AI agent vendors define a billable unit?

Deloitte documents billing across actions and tasks, compute, API calls, tokens, duration, or combinations. In practice, published examples include resolutions at $0.99 from Intercom, resolved interactions at $0.90 from Gorgias, verified automated resolutions at Zendesk, for which no list rate is published, and credit-denominated actions from Salesforce, where a standard 20-credit action equates to $0.10.

What billing infrastructure do companies need for AI agent pricing?

AI agent billing requires real-time usage metering over a preserved raw usage event layer, support for layered subscription plus usage structures, dimensional pricing across models and regions, credit systems for prepaid consumption, spend thresholds and alerts, and finance workflows that connect metered usage to ASC 606-aligned reporting and the ERP. Because pricing changes frequently, the system also needs versioning, simulation against historical usage, and backdating so corrections can be applied without manual reconstruction.

How is AI agent pricing different from traditional SaaS billing?

Traditional SaaS relies on predictable seat-based or flat subscription pricing. AI agents introduce variable consumption driven by tokens, API calls, task complexity, and compute, and rates may differ by model version, region, or feature set. Orb's 2026 research also identifies effort-based pricing, where compute time, steps, or task complexity become the usage measure. Outcome models add a further layer by charging for results rather than resources, which requires attribution rules, per-conversation caps, and the ability to not charge when an outcome is not achieved.

How often should AI agent pricing be expected to change?

Frequently. Stripe found that 92% of AI businesses charging for usage subsequently changed their pricing, and Orb's 2026 research describes AI agent pricing as a continuously iterated system rather than a one-time packaging decision. Cost pressure reinforces this: Gartner projects that generative AI cost per resolution for customer service will exceed offshore human agent costs by 2030. Teams should plan for repricing as a recurring operation and build the tooling to support it.