The Challenge
Create an enterprise platform where support teams could monitor AI conversations in real-time, intervene when necessary, and train the system continuously. The challenge was making complex technology accessible without sacrificing power and control. Approach
I spent 3 weeks with ControleNaMao’s customer service team observing real operations. I sat alongside agents during their shifts, followed problematic handoffs, and mapped friction points. What I discovered: Agents wasted time contextualizing each conversation from scratch. They had to read complete history before responding anything. They took over conversations unnecessarily or left customers waiting too long due to lack of clear signaling.
Managers had no visibility. They didn’t know how many conversations were active, which needed attention, or if the AI was improving. Generic reports didn’t show where to invest training time. Knowledge system was chaotic. Information scattered across docs, spreadsheets, messages. Updating AI responses required involving IT. Product team wanted to adjust tone but had no access.

Information Architecture
I structured the platform around real operation workflows: Service as the central hub with active conversation list and filters, Questions with chat view and complete history, Training to manage AI knowledge, Reports with metrics and evolution, Intelligences to configure assistants, and Team management with permissions. Fixed side navigation maintains context between sections. Company selector at the top allows switching between multiple clients (multi-tenant).

Design
I structured the platform around real operation workflows: Service as the central hub with active conversation list and filters, Questions with chat view and complete history, Training to manage AI knowledge, Reports with metrics and evolution, Intelligences to configure assistants, and Team management with permissions. Fixed side navigation maintains context between sections. Company selector at the top allows switching between multiple clients (multi-tenant).

Resultados
Colors define status in the list: green (AI processing), blue (human in command), gray (closed). Chat distinguishes AI through orange avatar. Side history loads 30 days of interactions, providing necessary context for agents to decide whether to take over or not. Dashboard top concentrates 6 critical metrics: service count, automation in hours, escalation percentage, handoff latency, human effort, and satisfaction score. Complementary visualizations reveal trends and individual gaps. Priority algorithm automatically identifies urgent conversations—high idle time or patterns indicating dissatisfied customer gain prominence. The platform offers complete dark theme, essential for teams monitoring conversations 24/7. Same semantic palette adapted, with contrast adjustments to ensure readability. Automatic or manual switching.