B2B Collections: Can AI Make Your Overdue Invoice Calls?

Raj runs finance for a sixty-person packaging supplier. At any time he has around four hundred open invoices, and a handful of people, himself included, chasing them between everything else. The email reminders go out on schedule. Most are ignored. The money that arrives late is nearly always the money somebody picked up the phone about.

The phone call works because it finds out why the invoice is late, which an email never does. It is also the step that never gets done consistently, because nobody enjoys it. That makes it a natural thing to hand to an AI voice system, and an easy thing to get wrong. This post covers what the call should achieve, which rules apply to business-to-business collections, and where a person has to take over.

Here is what matters most:

• Most overdue invoices are late because of a blocker, such as a missing purchase order, a wrong contact, a dispute or a pending approval, not because the customer will not pay. The call's job is to find the blocker and get a date.
• The US Fair Debt Collection Practices Act covers consumer debt. Business invoices fall outside it, but phone rules, state rules and your own customer contracts still apply.
• If you also collect from individuals, Regulation F limits how often you may call about one debt, and the rules are strict.
• An AI is best used for the first call on a recently overdue invoice. Disputes, large accounts and anything sensitive go to a person.
• What the call finds must land in your accounting system as a promised date or a task, not sit in a transcript.

What B2B collections is, and where the phone fits

B2B collections means getting paid by another business after the invoice is due. It is different from consumer debt collection in almost every way that matters. The customer is a company with an accounts payable team, an approval process and a relationship with you that you want to keep. The money is usually owed for goods or services, on agreed terms, and the delay is more often caused by process than by unwillingness.

Businesses use a handful of approaches, often in combination.

In-house accounts receivable follow-up. Your own team sends reminders, rings customers and escalates. It keeps the relationship in your hands and is limited by people's time.

Accounts receivable automation. Software sends the invoices, schedules the reminders, applies payments and flags overdue accounts. It is often sold as debt collection software or collection agency software, though a B2B version looks nothing like the tools a collection agency uses. This is the written side: emails, payment links, statements. The reminder sequence is sometimes called dunning, which is simply the schedule of increasingly firm nudges.

Accounts receivable outsourcing. An outside firm handles the follow-up for you, including calls. It frees your team and costs a fee, and its tone becomes your tone.

A collection agency. For invoices that are seriously late, a third party takes over, usually for a share of what it recovers. This is the point where a business "sends someone to collections", and it changes the relationship with the customer.

The phone call sits in the middle of this. Receivables software handles the written reminders well but cannot find out why an invoice is stuck. A person on the phone can, and it is the step many teams skip because it takes time. That gap is where an AI voice agent fits: it can make the first call on every recently overdue invoice, find the blocker and agree a date, and leave the difficult accounts for your team.

What the call is for

Think about what a good collector does on a call. They do not demand payment. They confirm the customer received the invoice, ask whether anything is stopping it being paid, and agree a date. If there is a problem, they find out what it is and who can fix it. Then they write it down.

That is a short, structured conversation, and it is the kind an AI can handle when it is set up carefully. The questions are the same every time: did the invoice arrive, is there an issue with it, is a purchase order or approval outstanding, can you give me a payment date, is this the right person to speak to. The answers fall into a handful of outcomes. The customer will pay on a stated date. There is a dispute. The invoice never arrived or went to the wrong person. Someone needs an approval first.

Each outcome should trigger something in your systems. A stated date becomes a follow-up task for that day. A dispute creates a ticket for the person who owns the account. A wrong contact updates the customer record and sends the invoice again. A call that cannot reach anybody schedules a retry. The work that matters is connecting the conversation to your accounting and customer records, so that nothing depends on a person remembering to type it up. We build these connections as part of our AI voice agent development work, and they are where the value sits.

To hear what an automated conversation of this kind sounds like, try the live demo below. It is one of our demo voice agents, and the heating-and-cooling business is closest to a company that qualifies a request and agrees next steps. Talk to it as a customer would, interrupt it, and change your mind.

The rules: what applies to business debt and what does not

I am not a lawyer, and the right answer depends on your location, your customers and how you collect. But there is a distinction that every finance lead should have clear.

The consumer and business line. The Fair Debt Collection Practices Act (FDCPA) is aimed at collecting consumer debts, meaning money owed for personal, family or household purposes. A trade invoice between two businesses is generally outside it. That does not mean anything goes. Some states have collection rules that reach commercial debts, and a third-party collector working for you may need a licence in the states where it operates. If your customers include individuals, for example a supplier that sells to both trade and the public, treat the consumer accounts separately.

Call frequency for consumer debts. Under Regulation F, the CFPB rule that implements the FDCPA, a collector is presumed to violate the law by calling more than seven times in seven consecutive days about one debt, or by calling within seven days of having an actual conversation about it. If any of your accounts are consumer debts, build the limits into the system so it cannot exceed them, whatever the schedule says.

Phone rules apply regardless of debt type. The Telephone Consumer Protection Act treats an AI-generated voice as an artificial voice. A call to a mobile phone using an artificial voice needs prior consent, and plenty of accounts-payable contacts give a mobile number. The safer approach is to call the business landline or main number published on the invoice or the account record, to record in your terms of sale that you may contact them about payment, and to read our guide to AI outbound calling before switching anything on. Some states also require everyone on a call to agree to a recording, so open each call with a short notice.

Honesty on the call. Say who is calling, on whose behalf and why. If the person asks whether it is an automated assistant, say yes. A customer who feels misled will remember it the next time they decide which supplier to pay first.

For a wider view of where voice agents work and fail, see our guide to AI voice agents for business.

Where the AI stops and a person starts

The first call on a recently overdue invoice is a good job for an AI. Day eight on a routine invoice from a customer who usually pays on time is nobody's idea of a relationship-defining moment. Put the people on the calls where they matter.

Route to a person: any invoice above a value you set, your largest and most strategic accounts, any customer with an open dispute, any call where the contact sounds annoyed or says they have already paid, and any customer who asks for a human. These are exactly the calls where tone, history and authority decide the result, and they are the ones your team should spend their time on.

An example ladder for an invoice on thirty-day terms might look like this. Email reminders before and just after the due date. An AI call about a week after, to find the blocker and agree a date. A call from a named member of your team if the promised date passes. An escalation to the account owner after that. Adjust it to your own terms and customers. What matters is that each step has an owner and that nobody is chasing the same invoice twice from different directions.

A short example shows how the first call might go. The system rings the accounts payable contact on file and says who it is calling for and why. It asks whether invoice 4471 for the March delivery has arrived. The contact says it has but is waiting on a purchase order number. The system asks who issues it, records the name, offers to resend the invoice with the order number field highlighted, and asks for a date by which the order number will arrive. It writes all of that into the account: blocker, owner, date. A person on Raj's team then follows up on the order number, not on the invoice, which is the thing that was actually stuck.

How to know it is working

Track the numbers you already care about. Days sales outstanding, the average number of days it takes you to be paid, is the headline measure. Alongside it, watch the share of promised payment dates that were kept, how early disputes are surfacing, and how many hours your team spends chasing. The dispute figure is often the surprise: invoices that sat unpaid for weeks turn out to have been blocked by a missing purchase order that a single phone call would have found.

Run it first on a slice, such as one customer segment or one week's overdue invoices, and listen to the calls. Compare cost per resolved invoice against your current process using your own team's time. If the AI route does not beat it on the first slice, you have found out cheaply.

FAQ

Is it legal to use AI to call customers about overdue invoices?

For business invoices, the main federal collection law, the FDCPA, does not apply, but the phone rules still do. An AI voice counts as an artificial voice under the Telephone Consumer Protection Act, which means consent rules apply to calls to mobile numbers. State laws and your contracts also matter, and some states cover commercial debt, so confirm the position with a lawyer who knows collections where you operate.

Does the FDCPA apply to B2B collections?

It generally does not, because it covers debts incurred for personal, family or household purposes. Invoices between businesses are normally outside it. If you also collect from individuals, those accounts are covered, and Regulation F's call limits apply to them. Keep the two groups separate in your systems so the right rules apply to each.

How does a business send someone to collections?

You hand the overdue account to a third-party collection agency, usually after your own reminders and calls have failed, and usually for a share of what they recover. Before you do, make sure the invoice is not in dispute, that the amount and terms are documented, and that you are content for the agency's tone to speak for your company. Many businesses use calls to find and fix blockers first, so fewer invoices reach this point.

Will an AI caller damage customer relationships?

It can if it is used carelessly, which is why I would not use it for your largest accounts or for anything in dispute. Used for routine reminders on invoices a few days late, with a clear route to a person and honest identification, it is usually received as one more administrative call. The calls that damage relationships are the ones that ignore context, such as chasing an invoice the customer disputed last week.

How is this different from accounts receivable automation software?

Receivables software sends invoices, reminders and payment links, and tracks what is owed. It is good at the written side. The phone call is the part it does not cover, and it is the part that finds blockers and gets dates. The best setup connects the two: the software knows what is overdue, the call finds out why, and the result goes back into the software.

Should I outsource accounts receivable or automate it?

It depends on volume and on how much control you want over tone. Outsourcing removes the work and charges a fee, but the outside team speaks for your company. Automation keeps the conversation in your own voice and systems, and costs more to set up. A common route is to automate the routine first calls and keep a person on the large or sensitive accounts.