AI in credit control does four practical jobs: it spots which invoices are likely to be paid late, writes and times reminders for each customer, sorts the replies that come back, and carries out the routine follow-ups you've allowed, such as a statement or a late fee. You set the rules, and you keep the decisions that need judgement, like a dispute or a client in difficulty.
In this article
The short version
- AI in credit control predicts late payers, drafts and times reminders, sorts replies and carries out follow-ups you allow.
- Late payment is rarely spread evenly, so spotting likely late payers is often the most useful first step.
- Disputes, hardship conversations and decisions to escalate stay with a person.
- UK law allows automated chasing; keep client data out of public AI tools and promotional lines out of reminders.
- Choose a tool where you can approve actions, pause single customers and send from your own email address.
What AI does at each stage of credit control
Three quite different technologies get called "AI" in credit control, and each one does a different job. Prediction models learn from payment history and score customers and invoices by how likely they are to be paid late. Language models write and read messages: they draft a reminder, or work out what a customer's reply means. And plain rules, such as "send the second reminder seven days after the due date", have been around for years, though they're often sold under the same label.
Across the credit control process, the work divides up like this:
| Stage | What AI can do | What you decide |
|---|---|---|
| Before the due date | Flag invoices likely to be paid late and send those customers a note first | Your payment terms, and who gets an early nudge |
| Chasing | Write each reminder and pick the day and channel from what that customer has responded to before | How often anyone can be contacted, and which channels are allowed |
| Replies | Read a reply and sort it: a promise to pay, a question, a dispute | Your answer to a dispute or a hardship request |
| Follow-up | Send a statement when an account goes quiet, add statutory late fees, offer an instalment plan | Whether each action runs on its own or waits for your approval |
| Payment | Match the payment to the invoice and retry a failed collection | Write-offs, and whether to stop work for a customer |
Spotting the likely late payers is often the most useful place to start, because late payment is rarely spread evenly across a ledger. In Adfin's own payment data, among business customers who'd paid at least five times, about a quarter were late on four or more invoices in every five, and that group accounted for about half of all late payments. The data leans towards small accountancy practices, so your figures will differ, but the shape is common: a short list of names creates most of your chasing.
For drafting, a language model writes your fortieth reminder as carefully as your first, uses the invoice number and amount the customer will recognise, and can follow your tone.
If a client ignores email but usually pays within a day of a text, a prediction model working from that customer's record can send the next reminder by text.
Replies are the part people tend to underestimate. Every chasing sequence produces answers, and each needs a different next step. Software that reads "we'll pay on Friday" and holds the next reminder, or spots a dispute, stops chasing and passes it to you, saves you from one of the more damaging things automated chasing can do: send a reminder after the customer has already told you there's a problem.
When a customer who's already paid gets chased anyway, the cause is often a payment that arrived with the wrong reference and was never matched, so the last stage matters too. Keeping reconciliation in the same system as your chasing makes that less likely.
Which decisions stay with you
Some of credit control is commercial judgement, and that's better left with you. Whether to extend terms for a client going through a bad quarter, whether to keep working while an invoice ages, whether a disputed invoice was right in the first place: those decisions are yours.
So is the conversation with someone who tells you they're struggling. That one needs a person at both ends, and it usually ends in an instalment arrangement or an honest date for payment.
Where you draw the line on automation depends on what each action does. Routine reminders can usually run on their own. Late fees, account stops and referrals to a collection agency all change a customer's position, so most businesses keep a person's approval on those, at least until they trust how the software has been set up.
What UK law expects when software chases for you
Chasing an unpaid invoice is ordinary use of your customers' data, and doing it with software doesn't change that. Since February 2026, UK data protection law has allowed decisions with a significant effect on a person to be made automatically, provided the person can get a human to look at the decision and challenge it. A reminder is unlikely to count as that kind of decision. An automatic account stop against a sole trader might, so make sure your customers can always reach a person.
Three other points tend to catch people out. A neutral reminder is a service message, but if the software adds a line promoting your other services, the email becomes direct marketing, and the consent rules for marketing emails apply to sole traders and many partnerships. Accountancy and tax practices should keep client data out of public AI tools, because their professional bodies' guidance treats that as a likely breach of confidentiality unless the client has agreed. And it's an offence to harass an individual with demands for payment so frequent, or made in such a way, that they're calculated to cause alarm or distress, so set a cap on how often anyone can be contacted inside the software itself.
How to choose and set up an AI credit control tool
Ask each vendor which of the three kinds of AI they mean, and what you can switch off. These questions tend to separate one tool from another:
- Can you review every action before it goes, or choose which actions need your approval?
- Can you pause chasing for one customer without affecting anyone else?
- Do messages go from your own email address, with replies coming back to you?
- Is your customer data kept out of public AI models?
- Does it read replies and stop chasing when a customer disputes an invoice?
- Does it sync payments both ways with Xero or QuickBooks?
Be careful with percentage claims about getting paid faster. They're usually a vendor's own figures, and there's no independent UK study comparing AI-timed reminders with a well-run schedule. What you can test in a trial is what the tool does, and whether it stays within your rules.
Adfin's Customer Agents are built around that kind of control. You group customers by how you want them chased, choose the channels (email, SMS or WhatsApp) for each group and decide how much runs without you. The agents can learn the timing and channel each customer responds to, triage basic invoice questions, send statements, add statutory late fees to business invoices and offer instalment plans, and you can pause chasing for any single customer. Payments are matched to invoices and synced with Xero or QuickBooks.
However you do it, start small. Switch automation on for routine, undisputed invoices first, read what goes out for a few weeks, and widen what runs unattended once you're comfortable with it.
Common questions
What does AI do in credit control? It predicts which invoices are likely to be paid late, drafts and times reminders for each customer, sorts replies by what they mean, and carries out follow-ups you've allowed, such as statements or late fees.
Is it legal to use AI to chase invoices in the UK? Yes. Chasing a debt is an ordinary use of customer data, and since February 2026 automated decisions with significant effects are allowed with safeguards, but take advice if software will stop accounts or refer debts without a person approving.
Does AI chasing get invoices paid faster? There's no independent UK study showing that, so treat any percentage you're quoted as the vendor's own claim. What it does change is consistency, because reminders go out on the day they're due.
What's the difference between automated reminders and AI chasing? Automated reminders follow one schedule you wrote for everyone. AI chasing varies the timing, channel and wording for each customer, within limits you set.
Does AI credit control replace a credit controller? It takes over the repetitive sending and sorting. Disputes, hardship conversations and decisions about escalation still need a person, and they tend to get more of that person's time.
Which parts of credit control should stay with a person? Disputes, conversations with customers in difficulty, decisions to escalate or stop work, and any message stating a legal consequence or a specific figure you'd need to defend.
Sources
- legislation.gov.uk - UK GDPR, Article 22C: safeguards for automated decision-making (accurate as of October 2026) https://www.legislation.gov.uk/eur/2016/679/article/22C
- ICO - Direct marketing guidance: identifying direct marketing (accurate as of October 2026) https://ico.org.uk/for-organisations/direct-marketing-and-privacy-and-electronic-communications/direct-marketing-guidance/identify-direct-marketing/
- ICAEW - Topical guidance: application of PCRT to ethical use of artificial intelligence tools (accurate as of October 2026) https://www.icaew.com/technical/tax/working-in-tax/pcrt/topical-guidance-application-of-pcrt-to-ethical-use-of-artificial-intelligence-tools
- legislation.gov.uk - Administration of Justice Act 1970, section 40 (accurate as of October 2026) https://www.legislation.gov.uk/ukpga/1970/31/section/40
- Adfin - Credit control (accurate as of October 2026) https://www.adfin.com/credit-control
