AI agents · Customer support
AI agents for customer support
A support agent reads an incoming ticket and resolves it against your knowledge base and internal systems. Zero-Shot connects these agents to your existing helpdesk; the agent cites the source of its answer, refuses to invent one when no source exists, and routes the ticket to the right person instead.
What we build
- 01A classification layer that sorts tickets by topic, urgency and customer type
- 02An agent that answers from your documentation and attaches the source link
- 03Tool calls that read the customer's account for account-specific answers
- 04Handover rules: anger, cancellation, legal language and unknown topics go straight to a human
- 05A draft mode where the support team edits and sends the agent's answer
How it runs
01
Evaluation set from past tickets
We sample resolved tickets, where the correct answer is already recorded, and measure the agent against that history.
02
Draft mode first
For the first weeks the agent never writes to a customer — it proposes drafts to the team. Acceptance rate is measured.
03
Hand over the easy topics
Where acceptance is high, the agent starts answering directly. Everything else stays in draft mode.
04
Close gaps with documentation
Topics the agent could not answer are listed. That list is the most honest report you will get on what your knowledge base is missing.
What we measure
- Share of tickets resolved automatically, broken down by topic
- Draft acceptance rate — how often the team sends the answer unedited
- Customer satisfaction after handover, compared with tickets that never touched the agent
- Number of answers produced without a source — a metric that should stay at zero
- Cost per ticket
When this is the wrong fit
If you have no knowledge base, or it is out of date, an agent has nothing to answer from. Do the documentation work first — the agent uses it, it does not replace it.
Common questions
Will the agent invent an answer it does not know?
The setup does not allow it. Answers are generated only from retrieved sources and the source link travels with the answer. Without enough source material the agent hands the ticket over instead of writing. 'Answers without a source' is tracked as its own metric.
Can we test it before it faces customers?
Yes, and that is the default path. The agent starts by proposing drafts to your team only. Once the team sends drafts unedited, topics are opened to customers one at a time.
Which helpdesk tools does it work with?
Any tool with an API: Zendesk, Freshdesk, Intercom, HubSpot or your own system. Email and WhatsApp channels can be connected too. You do not need to change your helpdesk.
Other services
What we've written on this
- Why AI agent projects failAgents that impress in a demo stall in production. Four observations on tool design, error handling, handing off to a human, and the cost ceiling.
- Where to start with enterprise AIHow to choose a first AI project: which tasks qualify, in what order to proceed, and what to expect in the first eighty days.
- Writing prompts for productionWhy a prompt that works in a demo breaks in production. Treating prompts like code: structure, versioning, boundaries and measuring every change.
Last updated: 2026-09-05
Let's talk about this one
Tell us what you want to build and within two days we'll tell you who takes it and how we'd measure it.