Money owed to you does not collect itself. Invoices go out, due dates pass, and the polite reminders that would bring the cash in get pushed down the list behind work that feels more urgent. Weeks later you have a receivables ledger full of overdue balances and no clean view of who owes what or who has already been chased. An AI agent can keep that view current and do the patient drafting: read the ledger every day, age the overdue invoices, and prepare a reminder for each customer that fits how late they are, so a person only has to review and send.

This guide covers using an agent for accounts receivable follow-up in five steps. It builds on how to set up your first AI agent, and it is the mirror image of the payables side in AI agent for accounts payable automation. Here the agent helps you collect; there it helps you pay.

What receivables follow-up means

Receivables follow-up is the ongoing work of turning overdue invoices back into collected cash through timely, well-judged reminders. An agent handles the legwork: it reads which invoices are open and past due, sorts them by how late and how large they are, and drafts the appropriate reminder for each. It does not move the ledger and, by default, it does not send. It prepares a run that a person approves.

This fits an agent because most of receivables is reading and routine writing, not decision-making. The ledger says what is owed and when it was due. A sensible reminder follows from the age of the balance and the customer's history. Producing that, consistently, across dozens of accounts, is tedious for a person and reliable for a language-model agent on a task with clear inputs and a checkable output (Anthropic, "Building Effective Agents", 2024). The judgment, when to soften, when to escalate, when a phone call beats an email, stays with you.

Why a person stays in the loop

A reminder to a customer is a relationship action, not just a transaction. Send the wrong tone to a good client and you can damage an account worth far more than the invoice. So the send stays human by default. The agent's job ends at a drafted, prioritized run. If you are deciding whether an agent or a simpler assistant fits this connected, ledger-reading task, AI agent vs chatbot vs assistant draws the line.

Follow-up vs single-invoice chasing

These overlap, so it helps to be precise. Single-invoice chasing is the narrow act of nudging on one outstanding invoice, the job covered in AI agent for invoice chasing. Receivables follow-up is the whole-ledger program: it ages every open invoice, prioritizes across all customers, and runs a tiered sequence over time. Chasing is one move; follow-up is the system that decides which moves to make and in what order.

The practical difference is scope and cadence. Chasing answers "remind this customer about this invoice." Follow-up answers "across everything owed to us, who should hear from us today, with what message, and what comes next if they still do not pay." If your need is a single reminder, the chasing post is the simpler fit. If you want a standing process that keeps the whole ledger moving, follow-up is the right frame. A related but separate case, recovering failed card payments, lives in AI agent for Stripe failed payment recovery.

1. Define the outcome

Write the result in one sentence first. For example: "An aged list of overdue invoices, grouped into 1 to 30, 31 to 60, and 60-plus days late, with a drafted reminder per customer that matches the bucket, ready for me to approve and send." That sentence fixes the scope to overdue items, names the deliverable, sets the aging buckets, and puts you on the send button.

Why outcome-first matters here

An outcome keeps the run focused and the tone deliberate. Describe the result and the constraints follow: which invoices count as overdue, how the buckets map to message tone, and who approves. This is the describe-the-outcome approach the platform is built on, set out in how to set up your first AI agent. You are stating what a good collections run looks like, not scripting each email.

2. Connect read access

To prepare a run the agent needs to read your receivables: open invoices, due dates, amounts, payment history, and customer contacts. It also needs to draft messages for review. It does not need to send mail, mark invoices paid, or write off balances. Grant read access plus draft-only creation, and keep sending and any ledger change with people. With no send permission, the agent cannot email a customer the wrong thing.

Scope financial access tightly

Give the narrowest access that does the job and confirm what the agent can see before connecting it. The no-send, no-write boundary is what lets you hand it the ledger without worry: it can read everything and still cannot act on a customer or change a balance. Treat receivables data with the same care as any sensitive system, as the broader AI agent security best practices guide explains.

3. Age and prioritize

With read access in place, the agent ages the ledger and ranks what matters. It groups overdue invoices into your buckets, then orders them so the largest and latest balances rise to the top. It also notes context that changes the message: a customer who always pays a few days late is not the same as one who has gone quiet for two months. The output of this stage is a prioritized worklist, not yet any drafts.

read_open_invoices(ledger)   -> all unpaid invoices + due dates
age_buckets(invoices)        -> 1-30, 31-60, 60+ days late
rank(by amount, by lateness) -> biggest, latest first
add_context(payment_history) -> habitual-late vs gone-quiet

Aging and ranking turn a flat list of overdue balances into a plan for the day. You collect more by spending your limited attention on the balances that matter most, and the agent makes that ordering explicit instead of leaving you to eyeball a long report.

4. Draft the sequence

Now the agent writes the reminder for each customer, matched to the bucket. An invoice five days late gets a light, friendly nudge. One past sixty days gets a firmer, clearer note about next steps. Each draft references the specific invoice, amount, and due date, and reads like a person wrote it, because a generic blast annoys good customers and gets ignored by late ones.

What a good reminder reads like

A good reminder is specific, courteous, and proportionate to the lateness. "Hi Sam, invoice 1043 for $2,400 was due on May 30 and is now three weeks past due. Could you let me know when we can expect payment, or flag if something is holding it up?" is firm without being hostile. The agent drafts to the tone you set per bucket; you adjust any message before it goes. If a run looks large and you want to size it first, how to estimate agent cost before deploying shows how.

5. Route to a human

The final stage delivers the aged list and drafted sequence to you. You approve the routine reminders quickly, rewrite anything that needs a personal touch, and decide which sensitive accounts deserve a call instead of an email. The agent then logs what was sent so the next run knows where each customer stands. The send and the judgment stay with a person; the preparation and the record-keeping do not.

This is not financial or legal advice

A receivables agent is a preparation tool, not a credit manager or a lawyer. It does not know your contracts, your local rules on collections, or when an account should go to a formal process. A drafted reminder is a proposal, not a decision about a customer relationship or a legal step. Treat the run as a starting point for a qualified person and you get the speed of consistent follow-up without ceding judgment that should stay with you.

The Gravity way to run it

On a platform like Gravity you do not build any of this. You describe the outcome, "age our overdue invoices and draft a reminder for each customer matched to how late they are, ready for me to approve," and an expert-built agent handles the read access, aging, ranking, and drafting, then hands back the run in about 60 seconds. You pay only when it runs, at $1 for 1,000 credits. The steps above are what a good agent does under the hood; you describe the outcome and approve the sends. To keep the books tidy alongside collections, see AI agents for bookkeepers.

Frequently asked questions

Can an AI agent chase overdue invoices?

Yes, the preparation part. An AI agent reads your receivables ledger, finds invoices that are past due, ages them into buckets, and drafts a reminder for each customer matched to how late they are. It does not send anything by default. A person reviews the drafts and approves which go out and when.

Does the agent email customers on its own?

Not unless you choose to allow it, and even then approval first is the safer default. A well-built receivables agent drafts the messages and routes them for review, because the tone of a reminder to a paying customer is a relationship decision. Keeping a human on the send button protects that relationship.

How is this different from invoice chasing?

Receivables follow-up is the whole-ledger view: it ages every open invoice, prioritizes by amount and lateness, and runs a tiered sequence across all customers. Invoice chasing is the narrower job of nudging on a single outstanding invoice. Follow-up is the program; chasing is one action inside it.

Is it safe to connect an AI agent to my receivables?

It can be, with tight scope. The agent needs read access to your AR ledger and the ability to draft messages for review. It should not send mail or change the ledger on its own. Grant the narrowest permission that does the job, review what it can see, and keep the send and any write-off with a human.

How do I set up a receivables follow-up agent?

Define the outcome first: an aged list of overdue invoices with a drafted, tiered reminder per customer, ready to approve. Connect read access to your AR ledger, set your aging buckets and tone, and route the drafts for review. On a platform like Gravity you describe the outcome and an expert-built agent prepares the run in about 60 seconds.

Three takeaways before you close this tab

Sources