AI
AI customer service pricing in 2026: per resolution, credits, seats or your own model key
Vendors charge for AI support in very different ways. Here is how each pricing model works, what it does to your bill as volume grows, a worked example with the same monthly volume, and the questions to ask before you sign.
MonoChat Team Updated: 9 min read
On this page
AI customer service is priced in five main ways in 2026: per resolution or outcome, per AI conversation, credit packs, AI bundled into seats, and bring your own LLM key, where you pay the model provider directly and the platform charges for the software. Each model puts the cost risk in a different place. Per resolution scales with success, per conversation with volume, credits and seats with caps, and your own key with tokens.
None is automatically cheapest. The right one depends on your volume, how good the AI is at resolving, and how much you value a predictable invoice. This guide explains each model, runs the same month of support through all five, lists the costs every model leaves out, and ends with the questions to ask vendors. If you want the full cost of one WhatsApp agent, see our companion guide to AI chatbot cost.
The five AI customer service pricing models
| Model | You pay for | Grows with | Predictability | Vendor’s incentive |
|---|---|---|---|---|
| Per resolution or outcome | Each conversation the AI resolves (definitions vary) | AI success and volume | Medium | Count as many resolutions as possible |
| Per AI conversation | Each conversation the AI joins | Volume | Medium | More conversations reach the AI |
| Credits | Units consumed by AI actions or replies | Replies, actions and features used | Medium to high until the pack runs out | Features that use more credits |
| Bundled in seats | Human seats, with AI included up to a limit | Team size | High until the limit | Sell more seats or tiers |
| Bring your own LLM key | Platform plan, plus tokens to your AI provider | Tokens per conversation | Medium, controllable | Platform value, not AI usage |
Per resolution or outcome
You pay a fixed fee each time the AI resolves a conversation. The appeal is obvious: no resolution, no charge. The catch is in the definition. Ask whether a resolution needs the customer’s confirmation, whether silence after an answer counts, how long the vendor waits before assuming it, and whether a handover the AI completed is billed.
As of October 16, 2026, vendors using this model on their own pricing pages include:
- Intercom Fin: $0.99 per outcome, with a minimum of 50 outcomes a month on the plan for any helpdesk. Fin’s pricing counts resolutions and some handoffs and routings as outcomes, at most once per conversation, and does not charge when a conversation is simply passed to your team without an outcome.
- Zendesk: automated resolutions are its billing unit for AI agents. Suite plans include a number of automated resolutions per agent each month, with additional ones at $1.50 committed or $2.00 pay as you go.
- Text (text.com): charges per AI resolution, with a monthly allowance per plan and automatic 50-resolution refills at $49.50.
Per AI conversation
You pay for every conversation in which the AI replies, resolved or not. Cost tracks volume, not quality, so a weak AI costs as much per conversation as a strong one. It is easy to forecast if you know your inbound volume.
- Tidio Lyro: billed per Lyro conversation, meaning an interaction where Lyro sends at least one reply. Standalone Lyro starts at $32.50 a month for 50 conversations; per-resolution billing is offered on its Premium plan by quote.
Credits
You buy or receive a pool of credits, and AI features consume them. Credits let a vendor price very different actions (a reply, a summary, a voice minute) in one unit. Read the conversion table: how many credits a reply, a tool call or an image uses, whether unused credits expire, and what happens when you run out.
- respond.io: uses AI Credits as the unit for AI usage. Its pricing page says credits are included with every plan, reset each billing month and scale with Monthly Active Contacts, its main billing metric.
Bundled in seats
AI comes inside a per-agent plan, often with a monthly cap or fair-use limit. Budgeting is simple and the AI feels free, but the cost is hidden in the seat price, and you may pay for seats you only need to unlock AI volume. Many vendors combine this with another model, for example seats plus included resolutions.
Bring your own LLM key
The platform charges for the product (agents, channels, inbox, automation), and you connect your own account with a model provider. You pay the provider’s token prices with no markup, choose the model per task and keep your own data agreements with the provider. The cost per conversation is usually lowest, but it is variable: long prompts, big retrieved documents and agents that loop on tool calls use more tokens. Budgets per run and cheaper models for routine questions keep it in check.
Meta’s own agent for WhatsApp, Meta Business Agent, is a sixth variation: on the WhatsApp Business Platform it is charged per token at one global rate of $2.00 per million tokens, covering both AI and message delivery. We compare it with platforms in Meta Business AI vs an AI agent platform.
Worked example: the same month under each model
The numbers below are illustrative. Each rate is a round assumption chosen to show how the models behave, not any vendor’s price.
The month:
- 2,000 customer conversations on WhatsApp, all to UK numbers
- The AI joins every conversation and resolves 60 percent (1,200); 800 go to people
- 5 business replies per conversation on average, 4 of them from the AI
| Model | Illustrative rate | AI line this month |
|---|---|---|
| Per resolution | $1.00 per resolution | 1,200 × $1.00 = $1,200 |
| Per AI conversation | $0.50 per conversation | 2,000 × $0.50 = $1,000 |
| Credits | 1 credit per AI reply, $0.02 per credit | 8,000 × $0.02 = $160 |
| Bundled in seats | AI included in 4 seats, up to a cap | $0 extra, until you pass the cap |
| Bring your own key | $0.002 per conversation in tokens | 2,000 × $0.002 = $4 |
The $0.002 figure is the average AI cost per conversation that MonoChat’s AI cost calculator uses for an efficient model. A heavier model could cost ten times as much and still be a small line.
Now add the costs that are the same under every model:
| Shared cost | Calculation | Monthly |
|---|---|---|
| Meta service messages | 10,000 replies × $0.022 | $220 |
| Human time | 800 × $0.68 | $545 |
Meta’s UK service rate comes from our bulk messaging calculator, based on Meta’s October 2026 rate card. The human cost uses the AI cost calculator’s assumption of one agent handling 4,400 conversations a month on a $3,000 salary. Platform or seat fees are left out because they vary by vendor and plan.
What happens when the AI gets better?
Say you improve your knowledge base and resolution climbs from 60 to 80 percent:
- Per resolution: the AI line rises from $1,200 to $1,600, while human time falls by about $273. You pay more to the vendor for the same customers.
- Per conversation, credits and your own key: the AI line stays roughly flat, and you keep the full human-time saving.
- Seats: you may be able to drop a seat, if the plan’s AI allowance does not depend on seat count.
That is the core trade-off. Outcome pricing shares the upside of a better AI with the vendor; usage pricing keeps it with you but leaves you the risk of a poor AI.
Hidden costs every AI pricing model leaves out
- Channel fees. On WhatsApp, Meta has charged every non-template reply as a service message since October 1, 2026, at the per-market utility and authentication rate, with no volume tiers. Templates are charged separately. Messages in the 72-hour free entry point window after a click-to-WhatsApp ad are free for delivery. See WhatsApp Business API pricing.
- Message count. An AI that answers in four short messages costs more in channel fees than one that answers in two complete ones, under every model.
- Human time. Handovers are usually the biggest line. A cheap AI that hands over half the chats is expensive.
- Setup and upkeep. Writing and maintaining the knowledge base, testing, and connecting order and CRM systems.
- Minimums and seats. Monthly minimums, required base plans and seat prices that come with the AI.
- Overage. What the per-unit price becomes after the allowance, and whether the AI stops or keeps billing.
Questions to ask AI customer service vendors
- What exactly is the billing unit, and how is it counted? Ask for the written definition.
- Is a conversation billed once, or can one conversation create several charges?
- Are handovers to my team billed? Are failed or abandoned conversations billed?
- What is included in my plan, and what is the overage price?
- Can I set a hard monthly cap or a budget per conversation?
- Which model powers the AI, and can I choose or change it?
- Can I bring my own LLM key, and is there a markup on AI usage?
- Are channel fees, such as Meta’s WhatsApp charges, passed through at cost?
- What reports show AI usage and cost per conversation?
- What happens to my bill if my resolution rate doubles?
How MonoChat prices AI
MonoChat keeps AI cost separate from the platform, so you can see and control it.
- Two ways to pay for AI. Use MonoChat AI APIs, billed by token consumption according to MonoChat pricing, or bring your own OpenAI, Gemini, Anthropic or other provider key and pay that provider directly. With your own key, MonoChat charges no AI usage fee.
- Bring your own key on every plan. It is included on every subscription plan, including the free Basic plan (one agent, one channel).
- Choose and route models. Custom LLMs connect providers such as OpenAI, Anthropic, Google, Azure OpenAI, AWS Bedrock, OpenRouter, xAI, DeepSeek, Groq, Cohere or self-hosted models through Ollama, with routing by cost, latency, capability or language, and fallback chains.
- Budgets per run. In the Agent Harness you set a budget per run, so an agent never spends more than planned on one conversation, and a fallback model can take over after a soft limit.
- Platform plans. Subscription plans cover agents, channels and outgoing messages. Incoming messages are free, and you can add message packs for more outgoing volume.
- Meta fees at Meta’s rates. Estimate them by country and category with the bulk messaging calculator.
Use the AI cost calculator to compare an agent’s cost per conversation with AI, check plans on the pricing page, and start free at app.monochat.ai.
Sources
Put this into practice with MonoChat