Use AI to draft overdue payment chasers without losing the human touch

What shipped

Zapier recently published a useful small business guide to Claude, one of the better known AI assistants. The article is broad, but one practical idea sits underneath it: an AI assistant becomes more useful when it is connected to the work already happening in the business, rather than treated as a separate chat window.

For an Irish SME, a good finance example is overdue payment review and follow-up drafting.

The workflow is straightforward. It checks the accounting system, or an exported aged debtors list, for overdue invoices. It groups the invoices by client and ageing period, such as 1 to 14 days overdue, 15 to 30 days, 31 to 60 days and beyond. It then drafts polite follow-up emails using the invoice details, client history and agreed tone of voice.

Nothing is sent automatically. The output is decision support. A finance or admin person reviews the overdue list, checks each suggested message, edits where needed and approves before anything goes out.

The named human review step is: Finance Review and Send Approval.

That review step matters. Credit control is not just a data task. There may be a disputed delivery, a payment plan, a long-standing relationship, a recent bereavement, a major order in progress or a client who always pays on the 20th regardless of invoice terms. AI can prepare the work, but a person should make the judgement.

For many smaller firms, this is not a project that needs model training, a data science team or a large platform. It can often be prototyped using exported data from Xero, Sage, QuickBooks, Surf Accounts or another accounting package, plus a small set of approved email templates.

Who should care

This is worth looking at if overdue invoices are a regular annoyance rather than a formal credit control operation.

It is a fit for firms where one person in finance, admin or operations is checking aged debtors every few days, copying invoice details into emails, trying to remember what was said last time and deciding whether to chase now or leave it until Friday.

Typical examples include professional services firms, trades and contractors, distributors, training providers, agencies, clinics, engineering firms and B2B service businesses. In these firms, the pain is rarely one huge unpaid invoice. It is the steady drag of small and medium overdue amounts that need a polite nudge.

The value signal is modest but real. If the firm has regular credit control, this workflow can save 1 to 3 hours a week. More importantly, it can shorten the time between an invoice becoming overdue and the first chase. A reminder sent three days after the due date is usually easier than a frustrated message sent five weeks later.

It also reduces avoidance. Chasing money can feel awkward, especially when the client relationship is good. A prepared draft lowers the effort needed to act. The person still decides what to send, but they are not starting with a blank email.

It is not a fit where payment collection requires complex legal processes, heavy dispute management or sensitive consumer debt handling. It is also not a fit if the business wants fully automated debt collection with no review. That is the wrong standard for an SME workflow. The safer and more useful version is a prepared review pack for a person who understands the customer.

A worked example

Take a 35-person engineering services company in Cork. It uses Xero for accounts and Outlook for email. The office manager checks aged debtors once a week, usually on a Friday afternoon. Some invoices are only a few days late. Others are 45 days overdue. The office manager knows many of the clients, but the follow-up process is inconsistent because other work gets in the way.

The first version of the workflow starts with a weekly export from Xero. The export includes client name, invoice number, invoice date, due date, amount, days overdue and payment status. If available, it also includes the client contact and any notes from previous follow-ups.

The workflow groups the overdue invoices by client. This is important because sending three separate chasers to the same accounts payable contact looks careless. A grouped view might show:

• O'Brien Facilities Ltd: two invoices, one 9 days overdue and one 34 days overdue
• Munster Plant Hire: one invoice, 16 days overdue
• Northside Retail Group: four invoices, all under 14 days overdue

The AI assistant then applies agreed rules. For invoices 1 to 14 days overdue, it drafts a light reminder. For 15 to 30 days, it drafts a firmer note asking for an expected payment date. For 31 to 60 days, it suggests a message that references previous reminders and asks whether there is a query blocking payment. For older invoices, it flags the account for manual attention rather than drafting a strong message automatically.

A draft for the first ageing band might read:

“Hi Mary,

I hope you are well. I am just following up on invoice INV-1847 for €2,450 plus VAT, which fell due on 4 March. Could you confirm whether this has been scheduled for payment?

I have attached the invoice again for convenience.

Kind regards, Accounts Team”

A firmer draft might read:

“Hi John,

I am following up on invoice INV-1792 for €6,800 plus VAT, now 32 days overdue. Could you let me know when payment is expected, or whether there is any issue with the invoice that we need to resolve?

Thanks, Accounts Team”

The finance review screen or spreadsheet then shows the client, invoices, ageing band, suggested action and draft email. The office manager checks the list, adjusts the wording, removes any clients who should not be chased and approves the final emails.

That is the Finance Review and Send Approval step. It is the control point that keeps the workflow practical and safe.

Over time, the company can improve the workflow. It might add a “do not chase” tag for clients with agreed payment plans. It might include customer-specific notes. It might draft internal Slack or Teams reminders for account managers when a key client is overdue. It might update a log when a follow-up has been sent.

But the first version does not need all of that. The useful starting point is a reliable draft pack that reduces the weekly burden.

Where I'd start

Start with the aged debtors list, not the AI tool.

Export the list from your accounting system and check what fields are available. At minimum, you want client name, invoice number, amount, due date, days overdue and contact email. If client notes or previous chase history are available, include them, but do not delay the pilot waiting for perfect data.

Next, define the chase templates by ageing band. Keep them short. Most firms need three or four tones: light reminder, payment date request, query check and manual escalation. Agree who signs the messages and what language should never be used.

Then test against 10 overdue invoices. Pick a mix of simple and awkward cases. Include one client with multiple invoices, one with a known dispute and one that should not be chased. Ask whether the workflow groups correctly, drafts sensibly and flags the cases where a person needs to decide.

The review questions are simple:

• Is the invoice information correct?
• Is the tone right for the client and ageing period?
• Has the workflow missed any reason not to chase?
• Would the finance person send this after a quick review?
• What should be logged after the email is sent?

If the answers are good, connect the workflow more tightly. That might mean a scheduled export, a direct connection to the accounting system, or a small internal tool where finance can approve, edit and send from one place.

For an SME, the win is not “AI does credit control”. The win is that the boring preparation is done before the finance person sits down. They get a clear list, grouped by client, with sensible drafts ready for review.

Artellis can usually prototype this kind of workflow in days, then refine it over a few weeks with the people who actually chase payments. The aim is not to remove judgement. It is to make sure overdue invoices are noticed sooner, followed up consistently and reviewed by a person before any message leaves the business.