What is AI customer service?
AI customer service uses language models to draft replies for agents and answer basic customer questions automatically, billed per resolution or per conversation depending on the vendor.
Updated September 23, 2026Also available in: Deutsch, Português
AI customer service is the use of language models to handle customer questions, either by drafting replies for human agents or by answering customers directly. In 2026 every helpdesk vendor ships some version of it. The differences sit in how it is billed and how well it is grounded in your content.
The two roles AI plays in support
The first role is the AI Copilot. It sits next to a human agent and drafts a suggested reply when a customer message comes in. The agent reads the draft, edits it if needed, and sends. The Copilot is meant to speed up the agent, not replace them.
The second role is the AI Bot. It talks directly to the customer, in chat or email. If it can answer the question from the knowledge base, it does. If not, it hands the conversation off to a human. The bot is meant to deflect simple questions so the team can focus on the hard ones.
How the AI knows your business
An AI model that has only been trained on the public internet does not know your prices, your refund policy, or your edge cases. To answer correctly it needs your content fed in at runtime. The technical name for this is RAG, which stands for retrieval augmented generation.
In practice it means the vendor's system reads your help center articles, your training documents, sometimes your past tickets, and packs the relevant ones into the AI's context window before the AI generates anything. The reply that comes back reads like your team would have written it because the model is literally working from what your team wrote before.
An ungrounded AI sounds confident but gets the details wrong. A grounded AI sounds like your team.
How vendors bill AI
Two patterns matter in 2026.
The first is per resolution. The AI gets billed every time it marks something solved. Zendesk does this at $1.50 to $2.00 per resolution. The bill is unpredictable: a viral spike doubles your AI cost the same month it doubles your volume.
The second is per conversation. The AI gets billed for every conversation it touches, no matter the outcome, at rates close to what the model itself costs. Deskwoot does this at $0.03 to $0.07 per conversation. The bill scales evenly with traffic.
At 5,000 AI conversations a month, the difference between the two models is roughly $7,500 vs $150.
What to compare
- Billing model: per resolution, per session, or per conversation?
- Foundation model: which one is under the hood (Claude, GPT, Gemini)?
- Grounding: knowledge base only, or training documents and past tickets too?
- Handoff: when the AI cannot answer, does it transfer cleanly to a human?
- Prompt injection posture: does the vendor have guardrails against malicious customer input?
Where Deskwoot fits
Deskwoot ships AI Copilot in every paid plan. The AI Bot Fynn is grounded in your help center plus the Training Hub, which can ingest website crawls, PDFs, and FAQ pairs. AI is billed per conversation at $0.03 to $0.07, with no per resolution surprise charges.