How much does an AI customer service cost? The 3 pricing models in 2026

Direct answer: an AI customer service is priced today according to three models. Per resolution: around $1 per conversation resolved by the AI (Intercom Fin bills $0.99). On subscription: from 50 to 500 € per month depending on the tool and the number of agents (Crisp, Zendesk). Or custom, as a one-time flat fee: 10 to 40 k€ at an agency, or 2 640 € excl. tax with me, starting from a building block already proven in production. The right choice essentially depends on one variable: your conversation volume. Here are the verified figures, the hidden costs, and the calculation to run before signing anything.

In brief

• Pay per resolution: Intercom Fin bills $0.99 per resolved conversation, minimum 50 resolutions per month, human seats on top ($29 to $139 per agent per month) if you use its helpdesk.
• Subscription: Crisp ranges from 45 to 295 € per month per workspace; Zendesk from $55 to $115 per agent per month, plus $50 per agent for advanced AI, plus roughly $1.50 per automated resolution beyond the quota.
• Custom at an agency: 10 to 40 k€ depending on channels and integrations, lead time in months.
• The hidden cost: operations. A RAG assistant consumes a few cents of AI API per conversation, that is less than 60 € per month at 500 conversations.
• The right metric: the cost per resolved conversation over 12 to 36 months, not the sticker price. It is what separates the three models.

The 3 pricing models on the market

1. Pay per resolution

This is the model popularized by Intercom with its Fin agent: $0.99 per resolution, that is, per conversation the AI closes without human intervention. No resolution, no bill: on paper, it is the model most aligned with your interest, and that is true to get started. In practice, read the terms: a minimum of 50 billed resolutions per month, and if you use the Intercom helpdesk around it, human seats from $29 to $139 per agent per month. Above all, the unit price does not drop with volume: 2,000 resolutions per month is about $2,000 per month, every month, indefinitely. Your bill grows at exactly the rate of your success.

2. Subscription

Crisp bills per workspace, not per agent: 45 € per month (Mini), 95 € per month (Essentials, with an AI chatbot limited to 50 actions per day), 295 € per month (Plus, unlimited AI, 20 agents). Zendesk bills per agent: the Suite ranges from $55 to $115 per agent per month, advanced AI adds $50 per agent per month, and automated resolutions beyond the small included quota are billed at roughly $1.50 each. In other words, Zendesk now combines subscription AND pay per resolution. The subscription has one virtue: it is predictable. Its limit: the assistant stays generic. It answers from your help center, rarely from your back office, and its resolution rate plateaus quickly if your answers depend on live data (order status, contract, customer file).

3. Custom

An agency that develops a support assistant from scratch generally bills between 10,000 and 40,000 € depending on channels (chat, email, phone) and integrations (CRM, back office, telephony). You own the system, it speaks in your tone, it reads your data, and no one bills you by volume. The downside: a high entry ticket, a lead time counted in months, and a real project risk if the agency discovers RAG at the same time as you. It is the model that amortizes best at high volume, and the worst if your volume is low.

The hidden cost: operations

Whatever the model, one line item is almost always missing from comparisons: operations. In a SaaS, it is included in the price (that is even where the margin sits). In a custom system, it is on you, and that is good news, provided you cost it honestly:

• The language model API: a RAG assistant consumes embeddings (search) and generation (answer). Order of magnitude: 2 to 5 cents per conversation depending on the chosen model and the length of the exchanges.
• Hosting: on an edge stack (Cloudflare Workers, managed database), count 10 to 30 € per month, not hundreds.
• Content maintenance: the real recurring cost, and the most underestimated. An aging FAQ means an assistant that answers wrong. Plan a few hours per month in-house to keep the documentation up to date.

Typical calculation. At 500 conversations per month: around 10 to 25 € of API and 10 to 30 € of hosting, that is less than 60 € per month. At 2,000 conversations per month: 40 to 100 € of API plus hosting, that is on the order of 100 to 150 € per month. Compare that to the roughly $1,200 monthly that the same volume would cost under pay per resolution (at 60 % resolution), or the 295 € of Crisp's beefiest subscription.

The calculation that matters: cost per resolved conversation

Comparing a price per resolution, a subscription and a flat fee only makes sense when reduced to a common unit: the cost per resolved conversation, over the real lifespan of the system, that is 12 to 36 months. Let us lay out a worked case. Tisselin, a home-linen e-commerce shop (a fictional but realistic case, assembled from situations I encounter): 800 support conversations per month, of which 60 % resolved by the AI, that is 480 resolutions per month and 17,280 over 36 months.

• Per resolution: 480 × $0.99, about $475 per month, that is roughly $17,100 over 36 months, excluding seats. Cost per resolved conversation: ~$1, constant, with no end of lease.
• On subscription (Crisp Plus, 295 € per month): 10,620 € over 36 months, that is about 0.61 € per resolved conversation, provided you hold 60 % resolution with a generic assistant. If the real rate drops to 35 %, for lack of back-office access, the cost climbs back above 1 € and the team handles the rest by hand.
• Flat fee (Buildoto base): 2 640 € excl. tax once, plus about 60 € per month of operations. Over 36 months, this comes to on the order of 0.35 € per resolved conversation, and each additional month lowers the average further.

The tipping point reads in the numbers. For Tisselin, the flat fee overtakes pay per resolution around month 10, and the Crisp Plus subscription around month 17. Generalizing: below about 100 conversations per month, pay per resolution or a small subscription win, the flat fee does not amortize. Between 300 and 1,000 conversations, the flat fee amortizes between month 8 and month 30 depending on volume. Beyond 1,000, pay per resolution becomes the most expensive model on the market, by far: it is an annuity built on your growth.

The Buildoto offer

My position in this market is simple: the AI customer service block is already built and proven in production, I customize it to your business. The base feature is 2 640 € excl. tax, a fixed flat fee signed before starting: chat and voice, answers bounded to YOUR content by a RAG (the assistant cites its source and hands off rather than inventing), human escalation with full context, integration with your existing channels. No cost per resolution, no cost per seat: you pay for the build once, then operations at cost, generally a few dozen euros per month. I detailed how it works, the AI/human balance and a full case in automating your customer service with AI, and the phone side in the AI voice agent for SMBs.

Honesty compels me to say the opposite too: below about 100 conversations per month, a subscription SaaS (Crisp's free or Mini plan, for example) is often enough, and I will tell you so at the audit rather than sell you a flat fee. Custom is justified when the volume, the integration need (back office, CRM, telephony) or the reliability requirement exceeds what a generic tool can do.

Where to start

• 1. A free 30-minute audit: we count your monthly conversations, the share that is genuinely automatable and the state of your documentation. You leave with a costed recommendation, including if it is "take a 45 € subscription and come back later".
• 2. The 36-month calculation: for each relevant option, total cost (license or flat fee + operations) divided by the expected resolved conversations. Ten minutes of spreadsheet that save three years of volume billing.
• 3. A reduced scope to start: 2 or 3 high-volume request types, the assistant bounded to your content, human escalation on everything else. Live in 2 to 4 weeks, we measure, then we extend if the numbers justify it.

Sources

• Intercom · pricing grid (Fin at $0.99 per resolution, seats)
• Crisp · pricing grid (Mini, Essentials, Plus)
• Zendesk · Suite pricing grid and AI options