Get Started

Your First Agent

Two playbooks for deploying your first enterprise-grade Support AI Agent on raia, following the Agent Factory lifecycle: five phases and fourteen steps, from defining the intended outcome through to auditing live conversations and feeding what you learn back into the agent.

Written for business sponsors, support leaders, Functional Experts, Agent Users, Technical Agent Builders, Agent Integration Specialists, and agent owners.

The two documents

Both cover the same lifecycle. One is the reference guide you work from; the other is the deck you run a session with. Most teams use both.

PDF · Guide

Deploying Your First Support AI Agent

The full business-led guide to the five-phase, fourteen-step lifecycle — the reference document for the agent owner and the delivery team.

Includes sixteen sections plus three appendices: a preparation and decision workbook, a historical ticket export specification, and a setup documentation checklist.

Download PDF

PDF · Session deck

Support Agent Playbook Walkthrough

The condensed working-session deck: what your team prepares, who is involved, how long it takes, and what has to be true before each phase closes.

Best used to run the kickoff session — phase exit criteria, the four roles, where the effort goes, and the four commitments to leave with.

Download PDF

Which one do I start with?

Run the walkthrough deck first to align sponsors and name the team, then use the full guide as the working reference through Scope, Training, Build, QA and Launch. The appendices in the guide are where decisions and evidence get recorded.

The Agent Factory lifecycle

Each phase produces the inputs the next phase needs. Launch is not the end — monitoring results, human interventions and conversation audits flow back into Build and QA through a governed change process.

Phase	Purpose	Primary output	Where the work happens
1. Scope	Define the agent's intended outcome, operating blueprint, and delivery team.	Approved agent definition and blueprint	Workshops and your own documentation — no platform work
2. Training	Gather the information the agent needs and connect the required applications.	Validated sources and application connections	raia Command
3. Build	Create the agent's instructions, workflows, and reusable skills.	Functional agent configuration and workflow	raia CX — the Launch Pad
4. QA	Test the agent, collect human feedback, and document quality findings.	Validated agent and audit report	raia Copilot
5. Launch	Verify safeguards, introduce human oversight, and audit live conversations.	Production agent with ongoing governance	Your channel, watched from Copilot

Phases close on evidence, not on dates

A phase is finished when its exit criteria are met and a named person accepts them. Any phase can close with a different answer: proceed, narrow the scope, do more discovery, or hold. A hold is a valid and useful outcome.

The fourteen steps

Every step has a deliverable and an owner. Nothing in Build should be discovered during configuration — every input was decided in Scope or Training.

Scope

1 · Define agent outcome
2 · Build blueprint
3 · Assign team

Training

4 · Collect data sources
5 · Connect applications

Build

6 · Build instructions
7 · Develop workflow
8 · Set up skills

9 · Run simulated tests
10 · Human feedback
11 · Build audit report

Launch

12 · Verify security and compliance
13 · Monitor and human in the loop
14 · Audit conversations

The delivery team

These are raia's four defined roles. Each has its own training path, so the person you name can be brought up to speed before the work starts.

Role	Who this is	Involved in
Functional Expert	Your support domain specialist — defines what a correct answer looks like and which source wins a conflict	Scope through QA · 6 modules
Agent User	The people who work the queue today — realistic testing, ratings, and takeover in Copilot	Scope, QA, Launch · 3 modules
Technical Agent Builder	Your developer or prompt engineer — instructions, functions, webhooks, the API, and workflow integration	Build through QA · 6 modules
Agent Integration Specialist	Your IT, DevOps, or platform engineer — connectors, ingestion, MCP setup, identities, permissions, and the Live Chat SDK	Training, Build, Launch · 3 modules

Two more names need no training path: a sponsor who owns the value decision and accepts each phase, and an agent owner who coordinates the four roles, keeps the decisions and evidence, and runs the beta. Start the training during Scope — there is no platform work in that phase, so it is free time to get each named person through their path.

Effort and schedule

A first narrow agent is roughly a hundred combined hours — about sixty on your side and forty on ours, weighted toward Training and QA. The platform work is fast once decisions, sources, access and owners exist.

Effort is not schedule

Elapsed time is set by how quickly your side can decide, extract and approve. Timelines stretch on waiting for a ticket export, unowned knowledge, a security review started late, expert time that keeps slipping, and approvals waiting on one person.

Before Scope begins

Two of these have long lead times. Start them today, not at the phase where they are needed.

Name one person for each of the four roles, plus a sponsor, an agent owner, and a deputy for the Functional Expert — with confirmed hours and training paths started.
Pick a single support domain as the candidate, and decide who the first users are and whether they are internal.
Find out where the historical tickets live and who can extract them — this export is needed for the simulated tests in QA.
Decide whether any of this needs a privacy, legal, or security review, and start it early.