← Winston Diaz

Handybots · 2023–present

Conversational AI Customer Support Agents

AI Product Management Consultant

Chat and voice support agents that resolve customer requests end to end and get smarter with every conversation.

24/7
availability
15–40 hrs
saved per week
4+
channels handled
40–70%
workload deflected

The problem

Small businesses are usually fielding the same support and scheduling requests over and over, across email, chat, SMS, and phone. Staff can spend hours on repetitive questions, and anything after business hours simply went unanswered.

How it works

Customers reach a Retell AI agent, chat or voice, backed by a shared knowledgebase, with a clean path to a live human whenever they want one. Behind it, custom tooling on Vercel and Neon calls managed agents and automation tools (Make, n8n, Zapier) to actually get things done.

The agent works through the client's own systems of record: the CRM doubles as the RAG source (HubSpot, GoHighLevel, Sheets, Airtable), alongside scheduling (Google Calendar, Fillout, Calendly), billing (Xero, Square, Stripe), and support (Intercom, Zendesk). Complex requests escalate to a human.

The learning loop

Every interaction is analyzed and stored, and the useful signal feeds back into the knowledgebase and models, so the agent grounds its next answer better than its last. Telemetry and a reporting dashboard give the team and leadership visibility into resolution rate, deflection, and cost saved.

The impact

For a video production studio, the scheduling and support voice agent saved 15 to 20 hours a week and took availability from eight hours a day to 24/7. The same pattern generalizes across clients, typically deflecting 40 to 70 percent of the repetitive workload.

How it works

Grouped into three zones: the customer, the AI support agent (with its tooling and systems of record), and outcome and learning. Every conversation resolves, is analyzed, and retrains the shared knowledgebase, which loops back to ground the agent. An observability dashboard aggregates signals from the agent, the outcome, and the analysis.

The value isn't a single bot. It's the loop: every conversation makes the next one better.

More work