AI Customer Support Agent

Problem

Support agents spent their day context-switching: pulling order details, product specs, and account history from four systems before they could even start writing a response.

Approach

I built an AI agent with a toolbox of scoped actions instead of a single monolithic prompt. The agent decides which tools to call — look up the order, fetch product details, check Salesforce — then composes a grounded, on-brand reply for a human to approve.

What I built

• An orchestrating Make AI agent with tools: get_order, get_product_details, email_customer, and update_zendesk.
• A human-in-the-loop step in Microsoft Teams so agents approve or edit before anything sends.
• Guardrails and logging on every tool call for auditability.

Results

• ~40% faster first-response times.
• Consistent, accurate answers grounded in live system data.
• Agents focus on judgment, not copy-paste.