AI Voice Agent for NetSuite AR Collections

Most AR teams I've worked with run the same process: pull a report of overdue invoices, hand it to someone, and hope they get through the call list before end of week. It works, but it doesn't scale. Two hundred accounts sitting past 30 days is a full-time job just to dial.

I built a different approach. An outbound AI voice agent reads overdue invoices from NetSuite, calls each customer, and logs what happened automatically. No spreadsheet. No one needs to remember to make the call.

How the flow works

Two phases, both automated.

The first is the trigger. Every night, a scheduled workflow in n8n queries NetSuite for invoices past 30 days due. For each customer it finds, it fires an outbound call through Retell AI, a voice agent that already knows the customer's name and the exact invoice amount before the phone rings.

The second phase is the callback. When the call ends, Retell fires a webhook back to n8n with the full transcript and outcome. n8n receives it and writes a Phone Call activity directly to the NetSuite customer record. The activity includes a plain-English summary of the conversation, the invoice number in the subject line, and a sentiment tag (Positive, Neutral, or Negative) so the AR team can triage at a glance without listening to recordings.

One detail worth getting right: the Phone Call activity links to the invoice itself, not just the customer. So when you open an invoice in NetSuite, the call is already there.

What the agent actually says

This surprised me during testing. Because the agent pulls real invoice data from NetSuite before dialing, it says "I'm calling about invoice 384 for $2,150 due April 4th." Not a generic script. Customers notice. A few asked follow-up questions the agent handled cleanly.

On one test call, the customer said they were confused about the charge. The agent offered to have someone follow up with details and ended the call professionally. The note that wrote itself to NetSuite read: "The agent contacted the customer regarding a past due invoice. Customer expressed confusion about the charge and the agent offered to have someone follow up to clarify. Sentiment: Neutral."

That's exactly what an AR rep would write. Except no rep touched it.

The stack

Three tools, each with a specific job:

• NetSuite: source of invoice data, destination for call logs
• Retell AI: outbound voice agent, post-call webhook
• n8n: orchestration, webhook receiver, NetSuite write-back

The demo runs on n8n Cloud. A production version would replace n8n with a Node/Express service for better error handling and multi-tenant support, but the core logic is the same: query NetSuite, trigger Retell, receive webhook, write back.

The part that takes the most time to get right is NetSuite auth (Token-Based Authentication across multiple HTTP nodes) and mapping the Retell webhook payload to the right NetSuite activity fields. Everything else clicks into place once those two are working.