This first phase is about building a clear foundation before you add data or train your AI agent. By setting goals, identifying where AI can help, and drawing boundaries for human-only tasks, you’ll save time later and avoid missteps.
Think of this phase as writing the rules of the game for your AI agent.
Steps
1. Define your goals
Decide what you want to achieve with Ordemio during the free trial.
Why it matters: Clear goals help you measure success and show quick wins to your team.
Examples:
Deflect 20% of repetitive tickets during the trial
Provide 24/7 frontline coverage when your team is offline
Increase first-contact resolution rate so fewer tickets need follow-up
Example pie chart of request types by percentage, highlighting common categories to prioritize for automation.
2. Identify your AI touchpoints
List all the places where your customers currently reach out for support.
Why it matters: Mapping these channels helps you choose where to deploy Ordemio first for maximum impact.
Common touchpoints include:
Website chat widget
Help Center / FAQ pages
Intercom or Zendesk chat
Email inbox / Tickets
Community spaces (Slack, Discord)
💡 Tip: Start with one or two high-volume touchpoints for a simple rollout.
3. Decide boundaries (AI vs. human)
Not every request should be automated. Draw a clear line between what your AI agent will handle and what must be escalated to a human.
Why it matters: Customers lose trust if sensitive or complex cases are left to automation without clear rules.
Examples of common boundaries:
FAQs and product setup → handled by AI
Troubleshooting basics → AI for common fixes, escalate if unresolved
Account changes, cancellations, billing → often escalated, sometimes automated with controls
Legal or compliance → always human-only
Request type matrix:
Request type | AI handles | Human | Notes |
FAQs (product info, setup) | ✅ |
| Great starting point for automation |
Troubleshooting basics | ✅ |
| Limit to common, repeatable fixes |
Account changes | ⚠️ Optional | ✅ | Always requires human approval |
Billing & refunds | ⚠️ Optional | ✅ | Escalate for financial accuracy/trust |
Legal & compliance |
| ✅ | Too sensitive for AI |
Cancellations | ⚠️ Optional | ✅ | Must be reviewed by human agent |
General product guidance | ✅ |
| Deflects tickets when tied to documentation |
⚠️ Optional = depends on your company’s policy
Example matrix showing how tasks can be split between the AI agent and human agents.
Privacy, Security & Compliance
Check these before launch:
GDPR: lawful basis, DSR flows (access/delete), retention, DPAs.
SOC 2: request Type I/II report from vendor.
PII: redact sensitive data in transcripts, restrict log access.
Plan handoff rules
Document when and how the AI agent should escalate a conversation to a support agent.
Escalation & Handoff Blueprint
Make handoffs foolproof so customers never have to repeat themselves.
Handoff triggers:
Customer asks for human (once → re-offer help; twice → escalate)
Refund/overcharge, identity/security, legal/compliance
2× fallback in a thread
Negative sentiment detected
Pre-handoff data capture:
Always collect and pass to the agent:
Customer email and plan/tier
Order/tenant ID or relevant error codes
Last three bot turns
Detected intent + tags
SLA alignment:
Max bot dwell time before auto-handoff
Differentiate business hours vs. off-hours
Apply stricter limits for VIPs
Escalation guidelines (example):
- Provide up to 3 assistance attempts before escalating.
- Each attempt should add value:
• Clarification question, OR
• Troubleshooting step, OR
• Quick explanation of why AI help might be faster.
- If the user requests a human:
• First request → respond politely, offer AI help once more.
• Repeated requests → acknowledge, try one last clarification.
- Confirmation step:
• Ask: “Would you like me to connect you to a human now?”
• Escalate only if the user confirms clearly.
Note: When escalation happens, the full conversation history is passed automatically to the support agent.
2-Week Trial Rollout Plan
A sequenced rollout helps you launch smoothly and keeps ownership clear.
Days 1–7 – Scope & Knowledge Setup (Support Lead + Docs Owner)
Publish a “never automate” policy (e.g. billing, refunds, cancellations, legal/compliance, security).
Define hard handoff triggers.
Draft escalation macros and routing (Intercom/Zendesk).
Prioritize top intents by volume and risk.
Upload Help Center first; split by OS/plan/role; remove duplicates; set version tags.
Add “narrow” facts as Q&A (e.g., license limits, API quotas, rate limits).
Days 8–14 – Evaluation & Pilot (QA Owner + Support Lead)
Build a golden set of 50–100 real customer questions with correct answers and citations.
Score each for correctness, groundedness, tone, and refusal correctness.
Pilot in 1–2 channels with agent-assist mode (AI first responder + fast handoff).
Set improvement goals: +10 pts containment, −20% ART, +2 pts CSAT during trial.
Best Practices / Tips
Set 1–2 clear goals for the free trial — expand later.
Involve your support or ops lead in defining boundaries and handoff rules.
Focus AI on repetitive, knowledge-based questions first.
Keep touchpoints manageable — better to nail one channel than juggle many.
Common Mistakes to Avoid
Trying to automate sensitive cases (cancellations, refunds, disputes) before setting up basic responses.
Skipping goal-setting — without metrics, you can’t prove value.
Launching on every channel at once (overwhelm).
Forgetting escalation rules — customers get stuck in loops or dropped chats.
Cross-references
For adding knowledge later, see Knowledge Base – Add data sources.
For integrations, see Intercom or Zendesk setup (covered in Phase 4).
✅ Expected outcome: Clear boundaries and escalation rules are defined.