Top AI Chatbots for Customer Support Automation: The Second Meter

Why the AI meter is not like the seat meter

A seat meter charges you for a human who is available. You pay whether they handle ten tickets or a hundred. An AI meter charges you for what the AI does, and that is where the confusion starts.

Three pricing models appear in this market for the AI layer:

• Per resolved outcome. You pay each time the AI finishes a conversation without handing it to a human. This is the model that surprises people. A team that automates well resolves many outcomes, and the AI bill rises with that success.
• Per session or per message. You pay a volume-based rate for each AI interaction, whether it resolves or hands off. This is closer to a seat model in that usage can be predicted from traffic, but the unit economics are different.
• Flat AI add-on per agent. You pay a fixed monthly amount per human seat to unlock AI features. This is the simplest to budget but covers the least automation, because it does not scale with how many conversations the AI handles.

The trap is signing up for a plan based on the seat price, then discovering the AI charge doubles your monthly bill. The only way to compare honestly is to run both meters for the same site.

Intercom: seat and resolved outcome

The Asyntai row is ours; read it as a disclosure. One monthly plan covers the AI, so nothing is charged per resolved outcome on top of it.

Fin AI charges once per conversation, not per message. One conversation may contain twenty exchanges, but the bill arrives once when that conversation closes. This matters for how you count the cost.

Example site: 8 agents, 600 conversations in a month, with Fin AI handling 40% of them automatically.

The seat price was $680. At 240 resolved outcomes the AI charge is roughly a third of the seat price. That uplift comes from the AI meter, not the seat meter.

Freshdesk: agent, AI copilot and AI agent on different meters

Freshdesk separates its AI into two products with different billing logic. Freddy AI Copilot is a flat add-on per agent. Freddy AI Agent is a session-based charge after a free allowance. Connector tasks are a third meter entirely, for integrations that push data between systems.

Ours again, as a disclosure. No session bundle and no per-agent AI add-on sits on top of the plan.

Day passes are available at $2, $7 and $12, which is useful for seasonal spikes but does not apply to the steady monthly AI cost. Connector tasks are billed at $80 per 5,000 tasks, which applies only if you run automations that push data through integrations.

Example site: 6 agents, 400 conversations, Freddy AI Agent handling 50% as sessions.

At this volume the site fits inside the 500-session free allowance. The free allowance creates a cliff edge. Push to 1,100 conversations and 550 sessions, and the overage is 50 sessions at the published rate. You are inside it until you are not, and the jump from free to paying is sudden rather than gradual.

Zendesk: seat and Copilot as a separate add-on

Zendesk structures its AI as a layer on top of its seat tiers. The seat price is one meter. Copilot is a flat add-on per agent, separate from the seat. Workforce Engagement and Contact Center are further add-ons that sit alongside rather than underneath the AI layer.

Ours, as a disclosure. The AI is the product, not a paid layer over a seat.

Example site: 10 agents on Suite Professional, Copilot enabled. The Copilot charge is per agent, same as the seat charge, so both meters scale together.

The uplift from adding Copilot to Suite Professional is 43% on the seat price alone, before any conversation volume is counted. This is the clearest example of a two-meter bill in this set: the pricing page shows $115 and you write a purchase order for $165 per agent per month.

Tidio: conversation-based with AI on the same meter

Tidio prices on conversations, not agents. The AI layer is embedded in the tiers rather than charged separately. Free covers 50 conversations. Starter covers 100. Growth covers 250 and above at a volume rate. There is no per-resolved-outcome charge and no per-agent AI add-on.

Ours, as a disclosure. Read the unit carefully here: this vendor counts conversations and we count messages.

Tidio is the only platform in this set that does not have a separate AI meter. The AI cost is embedded in the conversation price. This makes the bill simpler to read, but the per-conversation rate is higher than it appears if you are used to per-seat pricing, because one conversation may involve multiple agents over time.

Example site: 300 conversations in a month.

The pricing page says "Growth from $49.17 250+." For 300 conversations, the rate is not published in the facts given. The honest comparison with Tidio at this volume is to contact them, because the self-serve tiers do not directly cover it.

The worked example across all four vendors

Same site each time: 10 agents, 800 conversations in a month, AI handling 45% of them automatically. For vendors with per-resolved-outcome AI, that is 360 outcomes. For vendors with per-session AI, that is 360 sessions. For vendors with included AI, the AI cost is inside the plan.

For Freshdesk Pro, the agent cost and AI add-on are both per-agent, so the total is straightforward to calculate.

For Zendesk Suite Professional, the same structure applies with different rates.

For Asyntai the two thresholds are the message allowance and the dashboard seats the plan carries. Ten agents need the Pro plan, which carries ten seats. Assume 6 messages per conversation for 800 conversations.

For Intercom Advanced, the seat charge is per seat and the AI charge is per resolved outcome. At 10 agents and 360 outcomes, both meters apply.

We make Asyntai, so read that row as a disclosure. One monthly price covers the AI. The Pro plan at $449 carries 50,000 messages and 10 dashboard seats, which is what this ten-agent scenario needs; Standard at $139 carries 15,000 messages and 3 seats. The count above assumes 6 messages a conversation; use your own average when you have one.

At this volume the three established platforms charge between $840 and a seat-plus-AI figure that varies with outcome volume. Asyntai charges $449 for the plan that carries all ten seats. The difference is not a feature gap: it is the meter. When AI is a separate charge per agent or per outcome, the bill reflects both the human infrastructure and the automation layer. When AI is included in a flat plan, the automation layer does not add to the invoice.

The meter that surprises people

Intercom's Fin AI is the clearest example of the counterintuitive meter. Most buyers expect to pay for the software that handles easy tickets so their agents can focus on hard ones. That logic suggests automation should reduce the total cost: fewer agents needed because the AI handles volume. Fin AI charges per resolved outcome, which means the more tickets the AI resolves, the higher the AI bill runs.

On a team that resolves 500 outcomes in a month, Fin AI charges at that volume. On a team that resolves 1,000 outcomes, Fin AI charges at double that volume. The seat price is the same in both cases. The AI cost doubled, and it doubled because the AI worked better.

This is not a flaw in the pricing. It reflects a genuine cost to Intercom: each resolved outcome uses AI compute. But it catches buyers who are comparing the headline seat price against their current headcount without modelling the AI layer separately.

The question to ask before signing is not "what does AI cost per agent" but "what does AI cost when it works well." If your current deflection rate is 30% and you expect to reach 60%, model both scenarios. The AI meter may make the better outcome more expensive than the seat price suggests.

What to ask your vendor before you sign

1. Is the AI layer included in the seat price, or is it a separate meter? If it is separate, get the rate in writing.
2. What counts as one unit on the AI meter? A conversation, a message, a session, an outcome? These are not interchangeable.
3. Is there a free allowance, and what happens when I exceed it? Freshdesk publishes a 500-session allowance before the per-session rate starts. Intercom publishes a per-outcome rate and no allowance alongside it, so ask what happens on the first resolved outcome of the month.
4. Does the AI charge increase when automation improves? Per-resolved-outcome meters do. Flat add-on meters do not. Know which one you are on.
5. What is the cliff edge? At what volume does a free allowance run out, and what is the rate after it?
6. Is there an enterprise tier without published pricing? Zendesk Suite Enterprise is quote-only. Any comparison that excludes it is incomplete.

The AI layer changes what you should compare

When you buy a help desk on seat price alone, you are buying human availability. The AI layer is sold separately, and it is priced to reflect usage, not headcount. The two meters together are the real cost of running automated support.

Asyntai prices include the AI layer in the plan. There is no per-outcome charge, no per-session overage, no per-agent AI add-on. The message count is the only variable, and it scales with conversation volume rather than automation success. When the AI works better, the message count does not jump: the same conversations are answered with fewer wasted exchanges.