When you start researching the cost of an AI agent for customer service, you quickly notice a problem: one platform charges per resolution, another per session, another mixes usage with a monthly minimum fee.
In the end, a comparison that seemed simple starts to get confusing and, in most cases, ends up leading to the wrong decision. The challenge, then, is to understand what each provider is actually charging for, how exchange rates affect that bill, and what other costs may appear beyond the main amount of the proposal.
That is why this article organizes this comparison more clearly, showing the most common billing models, the price differences among the main platforms, and the criteria that help assess the real cost before signing a contract.
What are the pricing models for AI agents in customer support?
When a company starts comparing AI agents for customer support, the first confusion usually appears even before the price. Some platforms charge per resolution, others per session or handled ticket, and others work with a subscription combined with a minimum monthly amount.
In practice, this changes the reading of the proposal quite a bit. A price that seems low may be tied to a broader billing unit or one less connected to the final outcome. The difference between these models becomes clearer in the table below.
Model | What is charged | What makes analysis easier | Where it tends to be confusing |
Per resolution | The charge occurs when the AI resolves the support request according to the platform's rules. | Brings cost closer to the actual outcome and tends to make the ROI calculation easier. | Each provider may define “resolution” differently. |
Per session or handled ticket | The charge occurs per interaction, support session, or processed ticket, even if that does not represent final resolution. | It may seem more affordable upfront and is easy to measure by volume. | A session is not the same as a resolution, so the comparison can distort the real cost. |
Subscription with monthly minimum | There is a fixed minimum monthly amount, sometimes combined with consumption or a usage allowance. | Provides contractual predictability and can work well in operations with stable volume. | Can be too heavy for operations that are still testing volume or have fluctuating demand. |
In the per resolution model, cost is closer to the delivered outcome. In the per session or ticket model, the amount may seem lower at first glance, but that does not mean each interaction was resolved. In contracts with a monthly minimum, the math can be predictable for those who already have a mature operation, but less comfortable for those still validating volume, channel, or return.
For this reason, price without context usually gets in the way more than it helps. Before looking at the provider, the first step is to understand which unit is being charged and how closely it matches the outcome your operation really wants to buy.
Why is charging per session different from charging per resolution?
In the session-based model, the charge happens for use or interaction within a time window, even if that does not necessarily end with the problem resolved. In the resolution-based model, the charge is more tied to the outcome of the support interaction.
Charging per session is different from charging per resolution because the two things measure different moments of the support interaction. That is why a platform may seem more affordable at first glance and, even so, not deliver the lowest cost per actual outcome.
When does the monthly minimum become a trap?
The monthly minimum starts to weigh when the operation does not yet have stable volume or enough predictability to safely absorb that commitment. In companies that are still validating the channel, demand, or real resolution rate, taking on a fixed amount can distort the calculation right from the start.
In this scenario, it makes more sense to prioritize a model with lower entry risk and a cost structure that better matches the current stage of the operation.
Why is comparing prices in dollars with prices in reais harder than it seems?
When billing is in dollars, the proposal amount stops being predictable for an operation that budgets in reais. This does not make Zendesk, Intercom, or Freshdesk less relevant, but it does require a more careful reading.
The number that appears on the pricing page changes when exchange rates are involved, when the billing unit is not the same across providers, and when the operation tries to turn all of this into actual cost per result.
Zendesk charges by automated resolution (AI-resolved support), Intercom charges by outcome (conversation completed with resolution or eligible handoff), Freshdesk/Freshworks charges by session, and Cloud Humans charges by resolution. Without this context, the comparison seems objective, but it can lead to a mistaken interpretation.
How do you convert and compare the actual costs of the main platforms?
To make the comparison more concrete, we have gathered below the public prices of Zendesk, Intercom, Freshdesk, and Cloud Humans on the same basis of analysis. The table organizes the numbers in a way that helps understand cost, billing model, and exchange-rate exposure, without erasing the differences between the platforms.
Platform | Billing model | Estimated cost in BRL for 1,000 units/month | Exchange-rate risk |
Zendesk | Automated resolution (AI-resolved support) | R$ 7,856.40 to R$ 10,475.20 | High |
Intercom | Outcome (conversation completed with resolution or eligible handoff) | R$ 5,185.22 | High |
Freshdesk / Freshworks | Session (interactions within a support window) | R$ 2,566.42 for 1,000 sessions | High |
Cloud Humans | Resolution (AI-resolved support) | R$ 4,855 (contractual minimum) | Low (value in reais) |
Table built based on public prices reported by the platforms in March 2026. As a reference, we used the scenario of 1,000 units (session or resolution) per month according to each provider's billing logic.
The table helps compare cost ranges, but it does not resolve the analysis on its own. This is because not all platforms are charging for the same thing.
Cloud Humans, Zendesk, and Intercom work with units closer to an outcome.
Cloud Humans charges per resolution in reais.
Freshdesk charges per session, and a session is not automatically a resolution.
That is why the Freshdesk line may seem more affordable at first glance without necessarily representing the lowest cost per actual outcome.
What are the most common pitfalls when hiring an AI agent?
AI agent proposals often seem simple on first reading, but several important points only appear once the operation starts running or when the contract is reviewed more carefully.
That’s where the differences between the promised rate and the actual rate come in, along with costs that were left out of the main calculation, escalation rules to a human agent, monthly minimums that are misaligned with the current volume, and requirements to change the stack that make the decision much more expensive than it seemed.
Promised automation rate vs. real rate: how to validate before signing
A high automation rate in the sales pitch says little on its own. What matters is the real performance in an operation similar to yours, with comparable volume, channel, demand type, and knowledge base. Before signing, it is worth asking for concrete examples, understanding what is included in this calculation, and confirming how the platform measures resolution, handoff, and service success.
Invisible setup, integration, and migration costs
The proposal’s monthly price does not always show the total entry cost. Setup, integration with existing systems, workflow adjustments, operation training, and possible process changes can significantly alter the bill. For that reason, the comparison is only complete when these items are shown clearly, and not as a detail to be discussed later.
Human escalation SLA: what should be in the contract
When AI does not solve the issue, the billing rule needs to be clear. The customer needs to know whether they will be charged even when the interaction ends in escalation to a human agent. Depending on the model, this detail can significantly change the operation’s real cost. That is why it is worth understanding from the start what counts as a resolution, what is included as a handoff, and in which cases the charge does or does not apply.
Monthly minimum incompatible with current volume
The monthly minimum starts to weigh on the operation when there is not yet enough stable volume or predictability to absorb that commitment safely. In companies that are still validating the channel, demand, or real resolution rate, taking on a fixed amount can distort the numbers right from the start. In this scenario, it makes more sense to prioritize a model with lower entry risk and a cost structure that better matches the operation’s current stage.
Mandatory help desk switch: when this makes the decision unfeasible
In some cases, adopting an AI agent ends up driving a bigger change than the team had planned. When the solution requires switching help desk systems or a deeper reconfiguration of the operation, the cost is no longer just the AI itself and starts to include migration, training, and process adaptation.
How do you calculate the ROI of an AI agent before signing a contract?
Before looking only at the proposal amount, it’s worth organizing the calculation in a simple sequence. The goal here is to understand how much of the operation the AI can absorb, what savings that generates, and how that gain compares to the total cost of the solution.
Step 1 — Define the monthly volume that goes into the calculation
Start with the number of monthly support interactions that currently pass through the operation and that, in theory, could be absorbed by the AI. This is the starting point of the calculation.
Step 2 — Estimate a conservative resolution rate
Instead of starting from a high sales promise, use a conservative scenario. If the AI resolves 30% to 40% of the volume with acceptable quality, you already have a more realistic basis for projecting return.
Step 3 — Calculate the average cost of human support
Here you bring in the operating cost of the support that currently depends on the team. This calculation can consider team hours, allocated infrastructure, and the average cost per handled contact.
Step 4 — Estimate the avoided cost with AI
Multiply the volume the AI can resolve by the average cost of human support. This number shows how much of the operation can stop consuming direct team effort.
Step 5 — Compare that amount with the total cost of the solution
Now include the real cost of the solution: monthly fee, minimum consumption, integration, setup, and any other relevant onboarding or operating item.
Simple example:
Imagine an operation with 1,000 support interactions per month and an average cost of R$ 12 per human support interaction. If the AI resolves 40% of that volume, that represents 400 absorbed interactions and R$ 4,800 in avoided cost for the month.
If the contracted solution costs R$ 4,855, the direct return is already close to break-even, even without including scale gains, time freed up for the team, and reduced operational pressure.
This is when models charged per resolution in reais tend to become easier to forecast and justify internally, because they reduce uncertainty in the calculation for whoever needs to approve the investment.
What questions should you ask in the vendor demo before deciding?
Before moving forward with the proposal, it’s worth leaving the demo with clear answers to these questions:
What exactly counts as a resolution?
What counts as a session, outcome or billed interaction?
If the AI does not resolve it and the case goes to a human, will there be a charge?
Is there a minimum monthly amount? How does it work in practice?
Which setup, integration, or migration costs are included in the project?
Does the solution work with the helpdesk we use today?
What real resolution rate have you already achieved in operations similar to ours?
Which AI agent makes the most sense for Brazilian operations?
For Brazilian operations, the choice tends to make more sense when it combines financial predictability, low entry risk, and a billing logic that is easy to follow on a day-to-day basis.
That's why models charged per resolution in reais are usually more comfortable for companies that are still validating volume, channel, and operational return.
In this scenario, the Cloud Humans stands out for offering a simpler structure to predict, compare, and defend internally, without adding the currency-exchange component to the bill and without requiring the operation to accept, right from the start, a greater level of exposure than it needs.
Want to learn more? Schedule a demo here. Talk with our team, ask your questions, and compare (the right way) with other players in the market.
