Use AI to prepare the follow-up, not to fire reminders at everyone on the aging report. It reads the overdue line, the customer's history and the last emails, works out why the invoice is late, and drafts the right message in Arabic or English. A collector approves or edits it before it goes, and every send is logged.
#Key takeaways
- Most overdue invoices are not refusals to pay; they are stuck on a missing purchase order, a disputed line or an approval inside the customer's company.
- The useful AI work is sorting each overdue invoice by reason and drafting the message that fixes that reason.
- A collector approves every message, and anything about disputes, credit holds or legal steps stays with a person.
- Your accounting system's built-in dunning templates are enough when customers are few, pay mostly on time and rarely dispute.
- Measure overdue balance by reason, disputes resolved, and how many drafts the collector sends unchanged.
#What does the collections week look like for an AR team?
The AR analyst opens the aging report from SAP, Odoo or Dynamics 365, filters what is past due, and starts working down the list. For each customer they check the inbox, the CRM notes and the statement, then write an email or a WhatsApp message to whoever is responsible on the other side. Half the replies say the invoice never arrived or needs a new PO number.
The problem is rarely the reminder itself. The atradius press release, august 2025 on its UAE survey reported that 58% of credit sales were paid late, mainly because of administrative bottlenecks or financial distress at the customer. A generic reminder does nothing for a bottleneck. A message that sends the missing document to the right person does.
#Which collections tasks can AI take on safely?
- Sort each overdue invoice by reason: no PO, disputed line, waiting for approval, partial payment, no response, or real cash trouble.
- Draft the follow-up for that reason, with the invoice copy, the delivery note or the corrected PO attached.
- Match the tone to the account: a first gentle note for a good customer, a firmer one after repeated silence, in Arabic or English.
- Read the replies, log the promised payment date in your system, and remind the collector when a promise is broken.
- Summarise each customer's position before the weekly cash meeting.
Keep with a person: credit holds, payment plans, write-offs, anything mentioning a dispute in detail, and every legal or escalation step. Those touch the relationship and the balance sheet.
#What does a practical collections playbook look like?
- Agree reason codes with sales and customer service, so everyone tags the same way.
- Write message rules per reason and per step, including who is copied and when sales is told.
- Connect the aging report, the shared AR inbox and the invoice archive, read-only at first.
- Run the drafts beside your current process and let collectors mark each one sent, edited or rejected.
- Move to approve-and-send once the edit rate is low, keeping a person on every message.
Customer contact data is personal data. In the UAE, Federal Decree-Law No. 45 of 2021 governs it, as the UAE government portal, checked november 2026 sets out, so decide where the drafts are processed before the build starts.
#When are dunning templates enough?
If you have a few hundred active customers, low disputes and one collector who knows every account, the reminder schedules in Xero, Zoho Books, Odoo or your ERP do the job. A cheaper step is to clean your reason codes and contact list. AI earns its place when volume, languages and dispute reasons outgrow what one person can track.
#How does Stennir run collections with AI?
Stennir's AI consulting audit checks whether collections is your best first workflow, next to invoice intake or reconciliation. If it is, AI implementation builds it into your aging report and AR inbox in Arabic and English, hosted where your rules require, with a named collector approving every message. You pay for finished work, and terms are agreed on the call.
AI training then runs the AI Operator Program with the AR team on their real overdue list, so collectors learn to correct drafts quickly, with usage measured before and after.
We are
Stennir builds the collections workflow into your own systems, with a collector approving every message and every action logged.
We aren't
Stennir is not a debt collection agency, and does not contact your customers on its own.
#What should a finance lead do next?
Pull your aging report and tag the top fifty overdue invoices by reason. If most are paperwork, not cash trouble, the workflow is worth building. Read more at AI for finance and in AI for invoice processing, then take the free readiness assessment or book a 30-minute call.
#Common questions
Will customers know an AI drafted the message?
The message is sent by your collector, from your address, after they read it. Many teams add a line to their email policy saying drafts may be prepared with AI. What matters to the customer is that the message is correct, polite and fixes the reason the invoice is stuck.
Does it work with WhatsApp and Arabic replies?
Yes. Many Gulf customers reply on WhatsApp in Gulf Arabic or a mix with English. Stennir builds for Arabic and English, reads replies in both, and drafts in the customer's language. WhatsApp sending runs through your own business account, with the same approval step as email.
What could go wrong?
The main risks are a reminder sent to a customer with an open dispute, a wrong amount, or a wrong contact. The build checks for open disputes and credit notes first, pulls amounts only from your ledger, and holds any message where the data does not match until the collector reviews it.
What stays with us?
The accounts, the data, the message rules and the off switch. Your collectors keep every decision about credit, payment plans and escalation. Stennir hands over the workflow documented, so your team can change a rule or a template without calling anyone.
