When The Employees started growing, manual support stopped scaling. Every incident—a bot not responding, a failed checkout, a broken deep link—ended up in the same place: context-switching to diagnose something we'd already seen before.
We built a separate support agent that doesn't try to be smart. It just follows rules.
The 5 phases
1. Intake and classification
The customer describes their problem. The system identifies if it's onboarding, access, deploy, checkout, configuration, or a bug, and collects minimum data: which bot, what environment, when it started.
2. Triage
Before acting, it validates: is there enough evidence? Is there a playbook for this? If there's no context, it doesn't guess. It asks for more information or escalates.
3. Safe mitigation
Only executes documented and reversible actions: health checks, container reprovision, configuration validation. No explicit playbook = no automatic action.
4. Escalation with evidence
When the problem requires development or business decisions, the system creates a ticket in Odoo with logs, hypothesis, and clear next step. The technical team gets clean cases, not "something broke, check it" messages.
5. Closure and metrics
Every incident leaves a trace: was it solved with a playbook? Did it require escalation? How long did triage take? This tells us where to improve the product, not just the support.
What it doesn't do
It doesn't deploy product changes without human validation. It doesn't touch billing or sensitive data. It doesn't invent solutions without evidence. It doesn't decide pricing or commercial terms.
The result
Customers get immediate response and clear diagnosis. The technical team works with clean, traceable tickets. The business spends fewer hours on repetitive support and more time evolving the platform.
We're now in validation phase: playbooks are tested with real cases and metrics guide what to automate and what to keep manual.