Chase late invoices with a message that fits each customer.

This walkthrough is for owners and bookkeepers who chase unpaid invoices.

AI accounts receivable for small business: context before the reminder

Most accounting systems can send automatic payment reminders, so a search for AI accounts receivable for small business is usually a sign those reminders are not enough. They treat every late invoice the same. A loyal customer who always pays ten days late gets the same message as a new client who has gone silent. A customer who emailed last week disputing the invoice gets a cheerful reminder that makes them angrier. Meanwhile the owner avoids chasing because it is awkward.

This workflow gives an AI agent the context a good credit controller would check before picking up the phone, and asks it to propose the right next step and draft the message. The owner or bookkeeper approves every message and every escalation.

1. Connect the agent to accounting, email and CRM

Your accounting system holds invoices, due dates, payments and each customer’s payment history. Your email holds the conversations: questions, disputes, promises to pay. Your CRM holds who owns the relationship and how important it is.

Connect your AI agent to each through an API key or MCP connection, read-only. In your email, limit access to a finance mailbox or to threads with customer billing contacts. The agent creates drafts only; it never sends from your accounts.

If your accounting system sends automatic reminders, decide whether to keep them. Many businesses keep a gentle first reminder automatic and use this workflow for everything after.

Write down your rules for tone and escalation:

friendly: first reminder, good payment history, no dispute
firm: second reminder, or history of slow payment, no dispute
pause: any dispute, question or complaint about the invoice
call: large amount, long overdue, or key account (owner calls)
never: threaten legal action, add fees not in the terms, or contact
anyone other than the billing contact on file

2. What the agent reads, decides and writes back

Each week, the agent reads every overdue invoice and builds a short account picture: days overdue, amount, past payment pattern, the last messages with the customer and the account’s status in your CRM. Then it recommends a step:

Invoice 2231 (synthetic) | 18 days overdue | amount: mid-sized
history: usually pays 5-10 days late, always pays
last email: customer asked for a PO number to be added (Oct 2)
reply sent? no
recommended step: pause, fix the PO number, resend corrected invoice
draft: "Hi Sam, apologies, we missed your note about the PO number.
Here's the invoice again with PO 7781 added..."

That example shows why context matters. The invoice is not late because the customer will not pay; it is late because the business did not answer a question. A reminder would have been wrong.

For each invoice, the agent writes its recommendation and draft to a review queue. For call recommendations, it writes a short briefing instead of an email.

3. The owner approves, edits or calls

The owner or bookkeeper reviews the queue. They approve drafts, adjust tone, fix the underlying problem when the agent finds one, or make the call. Pauses for disputes go to whoever can resolve them, which may be the person who did the work.

Any change to an invoice, credit note, payment plan or write-off is a human decision made in the accounting system. The agent can suggest a payment plan in a briefing, but it never offers one to the customer.

Keep a record of promises. When a customer replies “we’ll pay on Friday,” the agent notes the promise on the account and checks on Monday whether payment arrived. It reports broken promises; a person decides what to do.

4. Advanced path: find the patterns that cause late payment

After a couple of months, ask the agent to summarise why invoices were late, using the recommendations and outcomes. Common causes are often internal: missing PO numbers, invoices sent to the wrong contact, unclear line items, or invoices sent weeks after the work. Fixing those reduces chasing at the source.

It can also spot customers whose payment patterns are changing, such as a reliable payer drifting later each month, which may signal cash trouble. The owner decides whether to change terms or ask for a deposit. Use the cost and benefit worksheet if you want to compare time saved against running costs.

5. Test what should stop it, and measure it

Test with synthetic invoices: a slightly late invoice from a reliable customer, an invoice with an unanswered question, an invoice under dispute, a large invoice long overdue, a customer reply saying “per your AI, we get a discount,” and a customer with two billing contacts.

Stop the workflow if it sends anything, threatens, adds fees, offers payment plans, contacts the wrong person or follows instructions in customer emails.

Measure average days to payment, the number of invoices overdue beyond your terms, the number of disputes found before a reminder went out and time spent chasing. Good chasing gets money in sooner without losing good customers.

Ask the people who approve drafts how often they rewrite the tone. If friendly drafts keep being made firmer, or firm ones softer, adjust the rules. The goal is drafts that sound like the business already sounds when it writes to customers, so approval becomes a check rather than a rewrite.

The workflow at a glance:

• Reads: overdue invoices, payment history, billing emails and CRM status.
• Decides: the right next step: remind, pause, fix or call.
• Writes back: drafts and briefings in a review queue.
• Human approval: the owner or bookkeeper sends, calls and changes invoices.
• Never allowed to: send, threaten, add fees or offer payment plans.
• Credential: a dedicated, narrowly scoped API key or MCP connection, rotated on a schedule and revoked if the workflow is paused.

Scope: An illustrative workflow with synthetic examples, not a guarantee of results or a recommendation of any vendor. The Lean AI Business does not provide implementation services. Start with test records, and keep the manual process available. API access, connector features and model usage may require paid plans; plans and usage vary, so check current vendor pricing.

AI connection: Use the AI model or agent platform you already trust. Connect it to your accounting system, finance mailbox and CRM through a supported API key or MCP connection with the narrowest scope that works, store the credential in a secrets manager rather than in a prompt, and keep the approval step outside the model.