AI Customer Service Agent

A support agent trained on your business and wired to your tools — one you own, not a per-resolution subscription.

PriceFrom $25K
Timeline4–10 weeks
TermsFixed scope

We build the support agent that resolves the repetitive majority — and hands the rest to your team with context. Trained on your help docs and real tickets, wired to your order systems through scoped tools, guarded by evals and a human-in-the-loop path, and deployed on the channels you already run.

Per-resolution platforms scale against you as you grow. This is a system you own: your data, your guardrails, your cost per resolution — and you can extend it.

Week by week
01Week 1

Scope the tickets

We mine your real ticket history for the repetitive majority, pick what to automate first, and define what a good resolution means.

02Weeks 2–4

Build on your data

Retrieval over your docs and tickets, scoped tool actions on your systems, and the escalation path — behaving on real cases from day one.

03Weeks 5–7

Harden it

An eval harness built from your ticket history, tone and safety guardrails, and human-in-the-loop review — so it survives edge cases, not just the demo.

04Weeks 8–10

Ship + hand over

Deploy to your channels, wire the observability dashboard, and transfer full IP. You own it.

Scope

What the engagement covers

The ticket mix

What we build
  • Your real ticket history
  • The repetitive majority
  • Success criteria per intent
What it does

It automates the tickets that actually recur — not a vague assistant that does everything badly.

What you receive

A working support agent, delivered on your data — not a Figma prototype.

The conversation, the actions it takes, the evals it passes, and the numbers it moves. A sample is shown; yours runs your tickets, on your channels, owned by you.

A real resolution

Retrieval, a scoped tool call, and a clean resolve — the agent doing the job.

support · liveSample
I was charged twice for my March invoice.
Let me take a look — I can see two charges dated Mar 3.
lookup_charges(customer) → 2 matches, 1 flagged duplicate
Confirmed: a duplicate charge of $49. I've refunded it — you'll see it in 3–5 business days.
issue_refund($49.00) → ok · ref RF-2231
Anything else? I can connect you to a billing specialist if you'd prefer.

resolved · confidence 0.94 · escalation offered

The eval harness

Scored against your real tickets before it ever meets a customer.

evals/support-agent.jsonSample
Eval suite
Production-ready
A
Resolution accuracy92
Correct tool calls95
Escalation precision88
Tone + safety97
Grounded answers96

The actions it takes

Named, scoped, audited — with approval gates above your thresholds.

agent/tools.tsSample
1// scoped, audited actions the agent can take
2defineTool('lookup_order', {
3 access: 'read',
4 input: { orderId: z.string() },
5})
6
7defineTool('issue_refund', {
8 access: 'write · approval > $500',
9 input: { amount: z.number() },
10})

The numbers it moves

What the support workflow looks like 60 days after launch.

impact · 60 daysSample
Ticket deflection
63%▲ 63%
CSAT
4.6/5▲ 0.3
Median resolve
40s▼ from 6m
Cost / resolution
$0.12
PlusFull IP transfer · a deployment runbook · 30 days of async Q&A after launch.
Fit

Built for

Head of Support

Deflecting tickets without hurting CSAT

An agent that resolves the repetitive 60% and hands the rest to humans, with context.

COO / Founder

Watching per-resolution pricing scale against them

Owned economics — a cost per resolution measured in cents, not a platform fee that grows with volume.

Ecommerce ops

Automating WISMO, refunds, and exchanges

A 24/7 agent wired to the order system that actually re-routes, refunds, and updates.

FAQ

Questions, answered

Not in scope
  • Help-desk migration (we integrate with what you run)
  • Phone calls — that's the AI Voice Agent & Receptionist

Common questions

Start here

Book a free scoping call.

We scope the exact number with you, and sign a mutual NDA before any code or data is shared.

Newsletter

One letter, every week. Working systems — not hot takes.

Build logs, agentic engineering decisions, agent failures, evals, and what survives real users. Sent weekly, never more.

Weekly. No spam. Unsubscribe anytime.