Skip to content

Prepare for Launch

Your AI system is built and tested. Now you need to make sure you're ready to go live safely.

Step 1
Checklist
Step 2
Monitor
Step 3
Rollback
Step 4
Communicate
Step 5
Launch

Step 1: Pre-Launch Checklist

Verify you've completed all necessary preparations.

Technical readiness:

  • Model validated on holdout data
  • Performance metrics documented
  • Bias testing completed (Check for Bias journey)
  • Model card created (Model Cards Guide)
  • Integration testing complete
  • Load/stress testing complete

Governance readiness:

Technical readiness:

  • Prompt templates tested and reviewed
  • Guardrails and content filters configured
  • Rate limiting and cost controls in place
  • Output validation logic implemented
  • Fallback behaviour defined

Governance readiness:

  • Acceptable use policy defined
  • User training materials ready
  • Data handling procedures documented
  • Vendor terms reviewed and accepted
  • Human-in-the-loop requirements defined

Step 2: Set Up Monitoring

You can't manage what you can't measure. Establish monitoring before go-live.

Read: Monitoring Guide

Monitor for:

Metric Why It Matters
Prediction accuracy Model degradation
Input data drift Distribution shift
Latency User experience
Error rates System health
Fairness metrics Bias emergence

Monitor for:

Metric Why It Matters
Token usage/costs Budget control
Response quality User satisfaction
Guardrail triggers Safety issues
User feedback Continuous improvement
Inappropriate content Reputational risk

Step 3: Define Rollback Plan

Know exactly how you'll revert if things go wrong.

Document your rollback plan:

  1. Trigger criteria - What conditions trigger a rollback?
  2. Decision authority - Who can authorise rollback?
  3. Rollback procedure - Step-by-step process
  4. Communication plan - Who to notify and how
  5. Recovery timeline - Expected time to restore previous state

Test your rollback

A rollback plan that hasn't been tested isn't a plan—it's a hope.

Step 4: Prepare Communications

Stakeholders need to know what's happening.

Communications to prepare:

Audience Message Timing
End users What's changing and how to use it Before launch
Support teams How to handle issues Before launch
Executives Launch confirmation and metrics At launch
Affected staff Impact on their work Before launch

For GenAI deployments, also address:

  • What the AI can and cannot do
  • When to trust vs verify outputs
  • How to report problems
  • What data not to input

Step 5: Go Live

Execute your launch with confidence.

Launch day checklist:

  • Monitoring dashboards open and visible
  • On-call team briefed and available
  • Rollback procedure accessible
  • Communication channels ready
  • Stakeholders notified of go-live

First 24-48 hours:

  • Monitor metrics closely
  • Gather early user feedback
  • Document any issues
  • Celebrate responsibly