Per Resolution vs. Per Conversation: AI Support Pricing Compared

Per-resolution pricing bills only after the AI meets a contract-defined success condition. Per-conversation pricing bills every eligible AI exchange, including exchanges that escalate or end without a clear result.

‍

Which model costs less comes down to the rate ratio, your actual resolution rate, and the fine print behind the meter.

‍

Fine print can turn similar unit prices into materially different invoices. Use the break-even formula and worked examples below to compare costs at your real support volume. Include platform fees, implementation work, repeat contacts, and human escalation in the math.

‍

Per-Resolution vs. Per-Conversation Pricing at a Glance

AI customer service pricing models can look simple on a rate card. Under a true per-resolution contract, the AI fee appears only after a successful outcome. Per-conversation contracts charge when engagement begins. The labels are not standardized, so treat the table as a common pattern and read your own contract word for word.

‍

Per Resolution

Billing trigger: Defined successful outcome
Failed attempts: No AI outcome fee
Forecastability: Varies with success
Incentive alignment: Vendor paid for outcomes
Main risk: Success definition can overcount
Best fit: Uncertain or complex workloads

‍

Per Conversation

Billing trigger: Billable AI engagement
Failed attempts: Usually charged
Forecastability: Varies with traffic
Incentive alignment: Vendor paid for activity
Main risk: Low resolution raises effective cost
Best fit: Mature, high-resolution workloads

‍

How Per-Resolution Pricing Works

Per-resolution pricing, often called outcome-based pricing, charges only after the vendor records a successful support result. A clear contract defines a successful outcome, whether the issue must avoid human escalation, and how long it must stay closed before the charge becomes final. Under a true resolution-based contract, failed attempts and default escalations should not trigger a resolution fee.

‍

This model protects the buyer because payment follows the result. Your AI support cost per resolution can climb as success improves, though. If the AI resolves twice as many cases next quarter, the variable part of the bill may double too. Finance gets accountability for outcomes, but the monthly total remains fluid.

‍

What Counts as a Billable Resolution?

Contracts diverge here. Intercom's Fin AI Agent outcomes documentation describes confirmed resolutions, where the customer signals that the answer helped, and assumed resolutions, where the customer leaves without asking for more help.

‍

It says an unsuccessful attempt or a request for a human does not count as an outcome. Zendesk's automated resolution tiers describes requests resolved by an AI agent without human escalation and adds value-based tiers to the billing structure.

‍

Neither approach wins automatically. A confirmation rule can miss customers who got the right answer and left, but an inactivity rule may count people who gave up.

‍

A no-escalation rule may miss a repeat contact that arrives later on another channel. Favor the definition your team can audit against conversation records and helpdesk events.

‍

Put these five contract terms in writing before comparing rates.

• Success signal
• Inactivity window
• Reopen window
• Treatment of human handoffs
• Handling of multiple issues in one thread

Ask whether the vendor can change any term mid-contract. These details matter more than the marketing name.

‍

When Per-Resolution Pricing Saves Money

Per-resolution pricing deserves a serious look when your expected AI resolution rate is uncertain or likely to start low.

‍

New deployments often begin that way because knowledge coverage is patchy, integrations still need tuning, and complex Tier 2 or Tier 3 work keeps escalating. The vendor absorbs the immediate fee risk when the AI misses.

‍

Picture 5,000 AI-engaged conversations with a 25% resolution rate. A true outcome contract bills 1,250 resolutions instead of 5,000 attempts.

‍

That gives your team breathing room to improve content, workflows, and evaluation while keeping the vendor focused on finished work.

‍

A smaller invoice can hide weak automation. The other 3,750 conversations still need somewhere to go, and many will bring human labor, slower replies, or another contact from the customer.

‍

Track coverage, escalation cost, reopens, and customer satisfaction (CSAT) beside the invoice. Otherwise, a low resolution rate can look deceptively efficient.

‍

How Should Ticket Complexity Change the Pricing Comparison?

Ticket complexity should change the comparison because one unresolved Tier 3 case can consume far more human effort than a routine Tier 1 ticket.

‍

QueryPal is built to resolve Tier 1 through Tier 3 tickets inside helpdesks such as Zendesk, Intercom, and Freshdesk. Use that complexity range as a second cost test.

‍

Group historical tickets by tier, assign each group its loaded human cost, and compare the work each model leaves with agents. A low AI invoice loses its appeal when the remaining queue consumes senior support and engineering time.

‍

How Per-Conversation Pricing Works

Per-conversation pricing bills each AI exchange that crosses the vendor's meter. The charge usually lands whether the interaction resolves, escalates, or ends without a clear result. You know the cost per conversation, but the cost of a successful resolution can still swing.

‍

Forecasting gets easier when traffic is steady and the billing boundary matches your helpdesk data. That forecast loses value when AI performance falls. A useful answer and a failed attempt carry the same unit charge. The failed case may still need paid human work.

‍

What Starts and Ends a Billable Conversation?

Start with the event that turns the meter on. Vendors may use the first customer message, first AI reply, completed authentication, or a minimum engagement threshold.

‍

Then pin down when the conversation ends. Inactivity windows, channel changes, reopened threads, and several issues inside one thread can all move the count.

‍

Salesforce currently lists Agentforce Conversations at $2 per conversation on its public pricing page. Treat that as a current public example, not a universal benchmark.

‍

Your order form can include commitments, included usage, multipliers, or other terms that change effective cost. The public page notes that pricing information can change and directs buyers to a sales representative for details.

‍

Before signing, ask for a sample usage report and reconcile it with one month of helpdesk exports. Every billed conversation should trace back to an event your team recognizes.

‍

Test the hard edge cases too, including a customer who returns after two hours, switches from chat to email, raises two issues in one thread, or comes back the next day.

‍

When Per-Conversation Pricing Saves Money

Conversation pricing can undercut outcome pricing when the AI already resolves a large share of the traffic it touches. Clean knowledge, repeatable Tier 1 questions, dependable integrations, and visible seasonal patterns raise the odds. In that setting, a low traffic-unit price converts into finished work with little waste.

‍

Efficiency changes everything. At $0.40 per conversation and a 60% resolution rate, the AI fee works out to roughly $0.67 per successful resolution. Drop the rate to 20%, and the figure jumps to $2.00 before human escalation. The sticker price never moved.

‍

Teams with stable traffic and tight control over duplicates or reopened threads may prefer conversation pricing. Predictable traffic does not guarantee predictable total cost, though. Escalations, harder questions, and repeat contacts can make the operation more expensive even when the AI invoice barely moves.

‍

During a pilot, keep a shadow invoice for both meters when the vendor sells only one. Apply the second formula to the same eligible conversations and verified outcomes.

‍

You will see how quickly the proposal changes when performance moves, and procurement gets one yardstick for rate cards written in completely different language.

‍

The Cost Crossover: Compare Both Models With Your Numbers

An AI support pricing calculator needs monthly AI-engaged conversation volume, expected AI resolution rate, and both unit prices. Use identical eligible traffic for each proposal. If one vendor handles chat alone but another covers chat and email, normalize the scope before touching the calculator.

‍

Monthly per-resolution AI cost = conversations × resolution rate × per-resolution price.

Monthly per-conversation AI cost = conversations × per-conversation price.

‍

Ticket volume determines the size of the budget, yet it drops out of the basic break-even formula. It still shapes minimum commitments, overages, seasonal spikes, and annual planning. Under equal assumptions, volume alone does not decide which meter is cheaper.

‍

Use the Break-Even Resolution Rate Formula

Break-even resolution rate = per-conversation price ÷ per-resolution price.

‍

Consider two illustrative rates, $0.40 per conversation and $1.00 per resolution. They break even at a 40% resolution rate. Below that point, per-resolution produces the lower AI fee because fewer outcomes are billed. Above it, the cheaper engagement fee gives per-conversation the edge.

‍

The formula covers AI usage fees and nothing else. Add base platform charges, seats, implementation, integrations, analytics, commitments, and unresolved-case labor before choosing. Test several performance levels too. The contract that wins at 35% may lose at 55% once knowledge and workflows improve.

‍

Model Three Monthly Volume Scenarios

The table uses illustrative rates of $1.00 per resolution and $0.40 per conversation across three monthly volumes and several performance levels.

‍

These are not vendor quotes. The effective figure divides the monthly AI fee by successful resolutions, giving both meters the same unit of value.

‍

At 25% resolution, the conversation model costs $1.60 per successful resolution, but the outcome model stays at $1.00. At 40%, both equal $1.00. At 60%, conversation pricing falls to about $0.67 per successful resolution. The pattern is identical at every volume, but the dollar exposure grows sharply.

‍

Stretch the model across 6-12 months. Give the launch, normal operations, and seasonal peaks their own assumptions. Let performance improve over time, then add any tiers or commitments that switch on as volume grows. One average month can hide the quarter when the cheaper contract flips.

‍

Volume: 1,000

25% rate

• Per resolution: $250, effective $1.00
• Per conversation: $400, effective $1.60
• Lower AI fee: Per resolution

40% rate

• Per resolution: $400, effective $1.00
• Per conversation: $400, effective $1.00
• Lower AI fee: Tie

60% rate

• Per resolution: $600, effective $1.00
• Per conversation: $400, effective $0.67
• Lower AI fee: Per conversation

Volume: 5,000

25% rate

• Per resolution: $1,250, effective $1.00
• Per conversation: $2,000, effective $1.60
• Lower AI fee: Per resolution

40% rate

• Per resolution: $2,000, effective $1.00
• Per conversation: $2,000, effective $1.00
• Lower AI fee: Tie

60% rate

• Per resolution: $3,000, effective $1.00
• Per conversation: $2,000, effective $0.67
• Lower AI fee: Per conversation

Volume: 20,000

25% rate

• Per resolution: $5,000, effective $1.00
• Per conversation: $8,000, effective $1.60
• Lower AI fee: Per resolution

40% rate

• Per resolution: $8,000, effective $1.00
• Per conversation: $8,000, effective $1.00
• Lower AI fee: Tie

60% rate

• Per resolution: $12,000, effective $1.00
• Per conversation: $8,000, effective $0.67
• Lower AI fee: Per conversation

‍

Hidden Contract Terms That Change the Real Price

Unit rates show only the visible edge of AI support pricing, especially when per-seat AI pricing sits beneath separate usage fees. The goal is to reduce customer costs without hiding service tradeoffs.

‍

A bargain rate can sit on top of an expensive commercial stack. Build separate first-year and renewal-year total cost models so one-time implementation work stays visible and recurring add-ons stay distinct from a low launch quote.

‍

Include the following costs in first-year and renewal-year models.

• AI usage fees and core platform subscriptions
• Required agent seats
• Implementation and integrations
• Knowledge preparation, analytics, quality assurance, and internal engineering
• Human handling for unresolved conversations

Then add the contract mechanics that change what you pay.

‍

Minimum Commitments, Overages, and Unused Credits

Minimums push volume risk onto the buyer. Commit to 10,000 units and use 7,000, and your real unit cost comes from what you paid for rather than what you consumed.

‍

Credits that expire make the shortfall permanent. During a launch, outage, or seasonal surge, overage premiums can hurt from the other direction.

• Do unused units expire at the end of the month, quarter, or contract term?
• Are overages billed at the contracted rate or a higher rate?
• Can committed volume move between chat, email, and other channels?
• Do seasonal spikes trigger a new tier, multiplier, or minimum?
• Can the vendor change the billing definition without a price cap or audit trail?

Ask every finalist to simulate six months of invoices using your historical volume and a low, expected, and high resolution rate.

‍

Your procurement team should be able to reproduce the result. If nobody can explain why a line appears, the proposal is too murky to compare.

‍

Setup, Integrations, and Human Escalation Costs

Implementation rarely ends with a one-time services line. Someone has to connect the helpdesk, clean up knowledge, map workflows, set permissions, build evaluations, review answers, and tune the system. The vendor will own some of that work. Support operations, engineering, security, and legal will absorb the rest.

‍

Human escalation is often the biggest missing cost in customer support total cost of ownership. Multiply unresolved conversations by the loaded cost of human handling, then add repeat contacts and reopens when they create more work. A failed outcome may carry no AI fee, but support still pays for it.

‍

Complex B2B support makes that gap much bigger. A Tier 2 or Tier 3 case may pull in a senior agent, engineer, log analysis, and several handoffs. Price that work differently from FAQ automation. The real prize is cheaper resolution without a drop in quality.

‍

Keep fixed costs and variable AI fees separate, then calculate avoided human work on its own. Fixed costs reveal what you spend before the first outcome.

‍

Variable fees show how the invoice reacts to usage and performance. Avoided cost tells you whether the program creates economic value. Blend those figures too early, and a weak proposal can look healthier than it is.

‍

How Do You Test Pricing Against Your Real Support Queue?

Test pricing against your real queue by applying each meter to the same historical tickets, comparing the price per support ticket, and adding the human cost that remains.

‍

QueryPal's ROI methodology starts with ticket volume, agent cost, and expected automation instead of a vendor benchmark.

‍

Build three cases using today's performance, a conservative improvement, and a high-performance case. Compare the vendor charge with the loaded cost of escalations in each case.

‍

The result shows whether the pricing model stays attractive as resolution improves and workload grows.

‍

Which Pricing Model Fits Your Support Operation?

No pricing model wins everywhere. Resolution maturity, ticket complexity, traffic stability, budget tolerance, and data transparency all change the answer.

‍

Start with current performance, then model the improvement you expect during the contract. A structure that protects a shaky pilot may become pricey once the AI gets good.

‍

Use the shortlists below to narrow the field. Then test the choice against your own break-even rate and full cost model.

‍

Choose Per Resolution If

• Your resolution rate is unproven, volatile, or likely to start below the break-even threshold.
• Your queue contains complex tickets that often need Tier 2, Tier 3, or engineering escalation.
• You want the vendor to share the financial risk of unsuccessful AI attempts.
• You can audit the success signal, reopen window, escalation rules, and usage report.
• Finance accepts spend rising as the AI successfully resolves more work.

A strong outcome contract starts with a billable resolution definition you can verify independently. Improvement becomes obvious as well.

‍

When the resolution rate rises, spend rises because more work is getting done. Compare that extra AI fee with the human work it replaces, not last month's invoice by itself.

‍

Choose Per Conversation If

• Your AI already resolves a high share of the conversations it engages.
• The conversation rate is materially lower than the outcome rate and stays below the break-even point.
• Traffic is predictable across channels and seasonal patterns are visible in historical data.
• You can identify conversation starts, endings, reopens, and duplicates in your own helpdesk.
• You prefer a known engagement cost and can absorb fees for cases that still escalate.

Conversation pricing works when a billed engagement is already a dependable route to resolution. If that connection weakens, the meter rewards activity, but your team carries the downstream cost.

‍

Watch effective cost per resolution beside CSAT and re-contact rate so low-cost units do not create dead ends for customers.

‍

A Third Option: Ticket-Volume Pricing

Outcome and conversation meters are only two AI agent pricing models.

‍

Ticket-volume pricing attaches a subscription or tier to an operating number your team already tracks, such as the tickets entering the support environment.

‍

That can quiet arguments over vendor-defined outcomes and give finance a budget tied to a familiar workload.

‍

The tradeoff shifts rather than disappears. You may buy ticket capacity that the AI never touches, and the plan still needs clear rules for scope, overages, and included features. For an established support team, ticket volume is often easier to forecast than next quarter's resolution rate.

‍

Our public pricing aligns plans with ticket volume and supports teams using existing helpdesk workflows. This structure is relevant for complex B2B operations that want predictable planning without reducing every interaction to a vendor-owned success label.

‍

Plan-specific pricing is custom, so use the public pricing page and a scoped proposal rather than assuming a rate.

‍

Whichever structure you shortlist, connect it to one operating goal. Scale customer support without scaling cost, backlog, or agent burnout.

‍

Ticket volume tells you how much work is arriving. Resolution quality tells you whether the technology is removing that work.

‍

Forecasts help only when every proposal covers the same work. Confirm the ticket sources, channels, integrations, and automation features inside the volume band.

‍

Stress-test normal growth, seasonality, and a product incident that briefly doubles demand. A transparent model should absorb those scenarios without inventing a fresh definition of success.

‍

Questions to Ask Every AI Support Vendor Before You Sign

A proposal becomes comparable once the meter, scope, and contract behavior are explicit.

‍

Send the same questions to every finalist and require the answers in the order form or an attached rate card. A polished sales deck alone does not count.

• What exact event creates a billable resolution or conversation?
• How are inactivity, abandonment, reopens, repeat contacts, and human handoffs treated?
• Can we audit billed units against conversation records and our own helpdesk export?
• Which channels, languages, integrations, and platform features are included in the unit rate?
• What minimums, overages, expirations, multipliers, and annual prepayment rules apply?
• What implementation, professional services, seat, analytics, or support fees are required?
• Can the billing definition or rate card change during the term, and what notice or cap applies?
• Who owns the usage data, evaluation records, prompts, and exported conversation history at exit?
• Can you model our last six months at low, expected, and high resolution rates, including total first-year cost?

One question cuts through the detail. Who gains financially when the AI improves, and who absorbs the cost when it fails?

‍

That answer exposes the incentive structure. Your team needs performance signals it can verify outside the vendor's dashboard.

‍

Frequently Asked Questions

The same neutral break-even framework applies to each answer below. Replace every illustrative assumption with your own rates, performance, and contract definitions before making a buying decision.

‍

Is Per-Resolution Pricing Cheaper Than Per-Conversation Pricing?

Per-resolution is cheaper when your AI resolution rate sits below the break-even point, calculated by dividing the conversation rate by the resolution rate. Above that point, per-conversation carries the lower AI usage fee. Both models cost the same at the threshold when eligible traffic is identical.

‍

That is only the first layer of cost. Add platform fees, implementation, commitments, and human escalation before calling a winner. A tiny usage charge can still produce an ugly total when interactions reopen or spill into expensive human queues.

‍

How Do You Calculate Effective Cost per Resolved Ticket?

Add AI fees, applicable platform costs, and human escalation, then divide the total by successfully resolved tickets.

‍

Keep the resolution definition consistent across both proposals. If one vendor treats inactivity as success but another requires customer confirmation, normalize the outcome count before comparing.

‍

Effective cost per resolved ticket is more useful than sticker price because it measures what finance and support leaders buy, which is completed work.

‍

‍Our customer support ROI methodology offers a broader framework that combines ticket volume, labor cost, and expected automation for planning.

‍

What Counts as a Resolution in AI Customer Support Pricing?

The definition varies by vendor. Common signals include explicit customer confirmation, inactivity after an answer, completion without human escalation, a defined reopen window, or automated quality evaluation.

‍

Some contracts separate different outcome types or assign them different values.

‍

Require the definition in contract language, test it against sample conversations, and compare it with your ticket deflection rate, reopen rate, and CSAT.

‍

A closed or deflected interaction may still leave the issue unresolved. A useful meter makes that distinction visible instead of burying it in a dashboard.

‍

Price the Outcome, Not Just the Unit

When comparing per-resolution vs. per-conversation pricing, run both contracts against the same historical workload, vary the resolution rate, and calculate effective cost per successful outcome.

‍

Add implementation, platform fees, commitments, quality effects, and the human cost of every unresolved case. A support leader and CFO can defend that decision together.

‍

Compare that decision with a more forecastable operating metric through our ROI calculator and review our ticket-volume pricing. Choose the model that lets support scale and keeps cost, quality, and accountability visible.

‍

For a practical AI support ROI test, run your ticket volume, labor cost, and expected automation through QueryPal's ROI calculator, then compare the result with both vendor meters.

‍

JetBrains reported that QueryPal helped it meet SLAs during a 62% year-over-year increase in ticket volume. Bring the model to a pricing review and test whether it still holds during your next peak.