Use AI voice agents with Xero to chase overdue invoices - Here's how

Why chasing late payments is still a pain

Chasing overdue invoices is nobody’s favourite job. It’s time-consuming, awkward, and often pushed to the bottom of the list. But what if accountants and bookkeepers could automate the whole process?

That’s exactly what

Beau Gaudron

from Growth Wise set out to show. In a

recent video

, he demonstrated how Xero, n8n, VAPI and Twilio can all be connected to automatically call clients with overdue invoices. The result? A working phone call from an AI voice agent that politely reminds someone about an outstanding payment.

So how does it work in practice? Let’s break it down step by step.

Setting up the workflow in n8n

Beau begins the demonstration inside

n8n

, a tool that works much like Make.com or Zapier. “We’re using it because it is something you can actually set up for free yourself,” he explained, adding that it’s a good way to get familiar with workflows before scaling up.

The process starts with a blank canvas. From there, Beau clicks the plus button to add nodes, which act as the building blocks of the workflow. These nodes include connectors for different apps such as Xero. Each one can be customised and renamed, so that the workflow is easy to follow later.

One of the first nodes sets up the

assistant ID and phone number ID

. These values can then be reused later in the flow, making configuration much smoother.

Pulling overdue invoices from Xero

With the basics in place, the next step is to connect to the Xero demo account. The workflow pulls in invoices where:

• The amount due is greater than zero

• The invoice is overdue (due date earlier than today)

• The type is accounts receivable

As Beau points out, this filter means the firm can ignore small amounts. “If you don’t want to chase people that have a debt that’s below, say, $50 you can change that amount due figure,” he said.

Testing this step showed three overdue invoices. Perfect for the demo.

Narrowing down the invoice data

Once the invoices have been captured, n8n allows the workflow to strip out only the fields that are needed. “The fields that we need are amount due, contact name, invoice number, and then the due date,” Beau explained.

Additional details like descriptions can also be included if required, but keeping it simple makes the next steps easier. This is where n8n really shines, letting accountants pass only the most relevant information along the chain.

Getting contact details

The workflow then switches back to Xero, this time looking up the

contact details

associated with each overdue invoice. Using the contact ID pulled from the previous step, n8n fetches details like names and phone numbers.

As Beau showed, “when it processes the data for each contact ID, it will get a contact.” The result is a neat package of information about the debtor, ready to be used by the AI agent.

For testing purposes, the limit was set to just one contact, preventing the system from trying to launch hundreds of calls at once. A handy safeguard when experimenting with new automations.

Connecting to the AI agent

The next part of the workflow calls out to

VAPI

, an AI voice platform. This is where the actual phone call is generated. Using the configuration set earlier, n8n retrieves the agent details and prepares them for the next step.

Inside VAPI, Beau creates a new assistant. The setup includes:

• Choosing the AI provider (in this case OpenAI, though others are supported)

• Selecting the model (GPT-4.1 Mini for faster response times)

• Adjusting settings for latency and responsiveness

• Adding a system prompt that tells the AI what to say, using variables like invoice number and amount due

• Configuring the voice with 11 Labs, including options for background noise to make the call sound more realistic

One small but important detail is setting the first message mode to

wait for user

. This ensures the AI doesn’t launch straight into a script, but pauses until the person on the other end has said hello.

Phone numbers and Twilio

Of course, to make a call, the agent needs a phone number. In the US, VAPI provides numbers for free testing. In Australia, Beau opted for

Twilio

, which costs around $3 per month for a number. Once added into VAPI, that number becomes the dial-out option for the AI assistant.

As Beau explained, “it’s not an incredibly expensive exercise to get this part of it set up.”

The live demo call

With everything connected, it was time to test the workflow. Beau clicked

execute workflow

and waited. Within seconds, his phone rang.

“Hello Beau, my name is Barry, and I’m contacting you on behalf of Growth Wise in relation to an outstanding payment,” the AI agent said. It continued with the invoice number, the exact amount due, and the overdue date, before politely asking when payment could be made.

This call was powered entirely by the workflow. n8n pulled the data from Xero, VAPI generated the voice call, and Twilio delivered it to Beau’s phone.

Fine-tuning for real business use

The demo ended with a reminder that some tweaking is needed before using this in practice. Firms would need to test different voices, prompts, and models to get the right tone.

Background noise can also be added to make calls more believable, whether it’s the sound of typing or even a recording of a real office.

As Beau put it, “you do need to take a little bit of time to actually try out different models, different voices, and different system prompts to get it set up in such a way that it’s actually appropriate to the tone of your own business.”

Bringing it all together

For many firms, chasing late payments is a constant struggle. Automating that process with tools like Xero, n8n, VAPI and Twilio could save hours of admin, reduce awkward phone calls, and speed up cash flow.

Growth Wise’s experiment shows it’s not just possible, but relatively low-cost to set up.

So here’s the question: if an AI agent can make those uncomfortable calls for you, why wouldn’t you give it a try?