Per-Action vs Per-Resolution AI Pricing: Which Is Right for You? (2026)

This is a decision guide, not a taxonomy. If you want the full rundown of all five AI agent pricing models, we cover that in AI agent pricing models, explained. Here we go narrow and deep on the one comparison most buyers actually agonize over — per-action vs per-resolution — with verified 2026 rates, three worked cost scenarios, and a clear read on which fits which team.

One disclosure up front: Macha (who publishes this) sells an AI agent layer priced per AI action. We'll cover that model with its real trade-offs, not as the obvious winner. Per-resolution is genuinely the right call for some teams, and we'll say so.

The quick answer

• Per-resolution wins when your tickets are simple, your AI resolves a high and honest share of them, and you value paying only for outcomes — and you're willing to audit how "resolution" is counted.
• Per-action / credits wins when your workflows are multi-step, your volume is high and your tickets are simple-but-many, or you want your bill to map transparently to work done with no fuzzy "was this resolved?" judgment in the middle.
• Per-conversation is the predictability compromise — you pay per conversation handled regardless of outcome, which avoids the "what counts as a resolution?" fight but charges you for failures too.

The rest of this article shows the math behind that.

Two philosophies: paying for the outcome vs paying for the work

Every pricing model is an answer to one question: what should the vendor get paid for?

• Per-resolution says: pay for the outcome. The meter only ticks when the AI fully closes an issue without a human.
• Per-action says: pay for the work. The meter ticks for each step the AI performs — understanding a ticket, searching your knowledge base, calling an API, drafting a reply, tagging, routing — whether or not that adds up to a tidy "resolution."

Neither is dishonest. They optimize for different things. Per-resolution aligns price to value in the clean case but depends entirely on a definition the vendor controls. Per-action aligns price to effort transparently but asks you to understand your usage. Hold three lenses up to each: predictability (can you forecast the bill?), incentive alignment (does the vendor earn more by helping you, or just by doing more?), and where the risk sits (volume risk on you, or definitional risk on a number the vendor sets?).

Per-resolution pricing, and what it really costs in 2026

You pay a flat fee each time the AI resolves an issue end-to-end. It's the headline model of the current AI-agent wave because it sounds maximally fair: only pay when it works. Verified current rates:

Rates current as of mid-2026; confirm on each vendor's pricing page before budgeting.

The pros are real. It's intuitive — you pay for results, not effort. In the clean case, incentives point the right way: if the vendor's model improves from resolving 60% to 75% of tickets, they earn more and you get more value.

The cons are equally real and worth naming:

1. "Resolution" is defined by the party who profits from it. Most vendors count an assumed resolution — if the customer doesn't reply or reopen within a window, it's billed as resolved. A customer who gave up in frustration looks identical to one who got a perfect answer. (Zendesk's move to LLM-verified resolutions is a direct response to this critique.)
2. Cost is hard to forecast. Tie the bill to a fuzzy event that scales with traffic, and a viral moment or an outage spikes your invoice. Zendesk's no-cap auto-billed overages make that sting concrete.
3. Hidden double-counting. Gorgias billing a resolution and a ticket, or Fin's separate $9.99 qualification tier, means the "$0.99/$0.90" headline isn't the whole story.

Per-conversation pricing: the predictability middle ground

A close cousin worth a mention, because it's where the "I just want a predictable number" buyers land. You pay per conversation handled — resolved or not — usually by committing to a monthly volume.

• eesel AI prices per interaction: Team $299/mo ($239 annual) for 1,000 interactions (~$0.24–$0.30 each), Business $799/mo ($639 annual) for 3,000 (~$0.21–$0.27 each), overage $0.15/interaction.
• MyAskAI prices per conversation: Pro $199/mo (~$133 annual) for 1,000 conversations, overage $0.12. Watch the fine print: through a help-desk integration, MyAskAI counts one conversation per 2 AI responses — so a long back-and-forth can register as several billable conversations.

Per-conversation is more predictable than per-resolution (conversation volume is easier to forecast than fuzzy "resolutions") and sidesteps the "what counts as resolved?" fight entirely. The downside: you pay even when the AI fails and a human takes over, so a weak agent costs you twice — the conversation fee and the agent's time.

Per-action / credit pricing: paying for the work itself

Here you're billed for each automated step the agent takes, typically through a credit system where different actions (or different underlying models) cost different amounts. This is Macha's model: Macha bills per AI action — each step the agent performs draws credits — and frames it as automation and orchestration rather than a packaged "outcome."

The pros:

• It's the most honest mapping of cost to work. You pay for what the AI did, with no fuzzy "was this resolved?" judgment call deciding your bill. A one-step "where's my order?" lookup costs less than a five-step workflow that touches three systems.
• No incentive to over-count wins. Because the vendor bills for steps, not for declaring victory, there's no definition to game.
• Granular and transparent. You can watch usage action by action.

The cons, stated plainly:

• It asks you to understand your usage. A per-action bill is only predictable once you know your actions-per-ticket pattern. A chatty or badly-configured agent can rack up actions.
• It's less intuitive than "$0.99 when you win," and it pushes the modeling work onto you.

That last point is the honest trade-off of the whole model: per-action rewards transparency but asks more of the buyer up front.

The trade-offs, side by side

Worked cost scenarios: the model that wins flips with your traffic

There is no universally cheapest model — only the cheapest model for your traffic shape. Three scenarios make that concrete. (Per-action math below assumes a representative ~$0.10 per action, in line with credit-based platforms; your rate and actions-per-ticket will vary — model your own.)

Scenario A: Simple tickets, high volume, high resolution rate

A consumer brand: 5,000 conversations/month, AI resolves 70% (3,500), each resolution takes ~2 actions (understand → reply).

Per-action and per-conversation win when tickets are simple. You're doing little work per ticket, so paying per step (or per cheap conversation) beats paying a flat outcome fee.

Scenario B: Complex, multi-step workflows

A B2B SaaS team: 2,000 conversations/month, AI resolves 50% (1,000), but each resolution averages 8 actions (multi-system lookups, API calls, multi-turn reasoning).

Now it's close, and per-resolution can win — because per-resolution caps your cost per outcome no matter how much work it took, while per-action exposes you to the step count. The heavier your workflows, the more per-resolution's "pay once" looks attractive.

Scenario C: Low resolution rate

An early deployment with a thin knowledge base: 5,000 conversations/month, AI resolves only 30% (1,500).

The lesson: **per-resolution rewards a good agent** — you only pay for the 30% that worked. Per-conversation punishes a weak agent — you pay for all 5,000 attempts, most of which failed. Per-action sits in between, billing only for steps actually taken. As your resolution rate climbs, per-resolution gets more attractive relative to per-conversation.

Which model is right for you

• Choose per-resolution if: your tickets are relatively simple, your agent already resolves a high and verifiable share, and you want to pay only for outcomes. Non-negotiable: make the vendor show you exactly how a resolution is counted (assumed vs. verified, double-counting, minimums) and audit it against your own CSAT and re-contact data.
• Choose per-conversation if: finance needs a flat, forecastable number above all, and you're comfortable paying for conversations the AI doesn't solve. Best once your resolution rate is high enough that you're not paying mostly for failures.
• Choose per-action / credits if: your workflows are multi-step and agentic, you want cost to track actual work with no definition to dispute, and you're willing to model your actions-per-ticket. Best for teams that want transparency and an agent that layers onto the helpdesk they already run.

And whatever you pick, do the one thing most buyers skip: run your own volume through every model like the scenarios above, using your real resolution rate, conversation volume, and ticket complexity. The model that looks cheapest in a vendor's example almost never wins on your traffic. For the broader budgeting picture, our help desk software pricing comparison puts the platform fees alongside these AI usage charges.

Where Macha fits (honestly)

Macha is an AI agent layer that runs on top of your existing helpdesk — Zendesk or Freshdesk — rather than replacing it. It bills per AI action: each step an agent takes (understanding a ticket, pulling from your connected knowledge, drafting a reply, tagging, routing, resolving) draws credits, varying by the action and the model used.

The honest reason we don't bill per "resolution" is the argument running through this whole article: a resolution is a fuzzy, sometimes-fictional event, and charging for one we can't always verify would be exactly the misalignment we just flagged in the per-resolution model. Outcomes vary with how good your knowledge base is and how messy your tickets are, so we price the work the agent actually does and call it automation, not a guaranteed outcome.

That model's real trade-off, stated plainly: per-action asks you to understand your usage. If you want a single "pay only when you win" number and you trust a vendor's resolution definition, a per-resolution agent may suit you better. If you want cost that maps to actual work, no incentive to over-count wins, and an agent that layers onto your current stack, the per-action model is built for that — and you can watch the usage yourself. 7-day free trial, no credit card required.

Frequently asked questions

What's the difference between per-action and per-resolution AI pricing? Per-resolution charges a flat fee each time the AI closes an issue end-to-end (e.g., Intercom Fin at $0.99). Per-action charges for each step the AI takes — understanding, searching, drafting, routing — usually via credits. Per-resolution prices the outcome; per-action prices the work. Per-resolution is more intuitive but depends on a vendor-controlled definition; per-action is more transparent but asks you to understand your usage.

Is per-resolution or per-action cheaper? It depends entirely on your traffic. For simple, high-volume tickets that resolve in 1–2 steps, per-action (or per-conversation) is usually cheaper. For complex, multi-step workflows (8+ actions per ticket), per-resolution can win because it caps cost per outcome no matter how much work it took. Model your own resolution rate and actions-per-ticket before deciding.

How much does Intercom Fin cost vs other AI agents in 2026? Intercom Fin is $0.99 per outcome (resolution, procedure handoff, or disqualification), billed at most once per conversation, with lead qualifications at $9.99 and a ~50-resolution/month minimum. Zendesk AI agents run ~$1.50 per verified resolution committed (~$2.00 pay-as-you-go). Gorgias Automate is ~$0.90–$1.00 per automated resolution but also counts each as a helpdesk ticket. Confirm all rates on each vendor's pricing page.

What is "assumed resolution" and why does it matter? Most per-resolution vendors count a resolution when the customer doesn't reply or reopen within a window after the AI's last message. The problem: a customer who gave up looks identical to one who got a great answer, and you're billed for both. It can inflate cost and misalign the vendor's incentive toward counting generously. Zendesk's LLM-verified resolutions and per-action models both sidestep this in different ways.

Does per-conversation pricing charge me for failed conversations? Yes. Per-conversation models (eesel, MyAskAI) bill for every conversation handled, resolved or not — so a weak agent costs you twice (the conversation fee plus the human agent's time). Watch for double-counting too: MyAskAI counts one conversation per two AI responses through a help-desk integration.

The bottom line

There's no single "best" AI support pricing model — only the one whose risks you can live with for your traffic. Per-resolution aligns price to outcomes and rewards a strong agent, but only as honestly as the vendor's definition of "resolution," and it can surprise you at scale. Per-conversation buys predictability at the cost of paying for failures. Per-action / credits maps cost to work transparently with no win to game, but asks you to know your usage. Run your real volume, resolution rate, and ticket complexity through all three, demand the exact definition behind any "outcome" you're billed for, and weigh predictability, incentive alignment, and where the risk sits. The headline per-unit price is the least useful number on the page.

Vendor rates cited were current as of mid-2026 from each company's pricing pages and secondary coverage; several (Zendesk, Gorgias) are search-cited and should be confirmed directly. Pricing in this category changes frequently — next review by December 2026.