AI Agents for Payment Reminders: Getting Invoices Paid Without Awkward Calls

Every SME owner knows the moment. An invoice is three weeks overdue, the client is otherwise good, and now someone has to pick up the phone and ask for money they’re already owed. It’s an uncomfortable call to make, so it gets pushed to tomorrow. Tomorrow becomes next week.

That delay has a price. AI payment reminder automation takes the reminder cycle off a person’s plate entirely: it watches which invoices are due, sends the right message at the right time in the right tone, and only pulls a human in when a decision actually needs one. Nobody has to make the awkward call just to ask when payment is coming.

What Late Invoices Actually Cost, in One Illustrative Case

Take a 15-person engineering firm billing CHF 2.4 million a year, issuing around 140 invoices a month on 30-day terms. The numbers below are constructed to illustrate the mechanics; treat them as a model to run with your own figures rather than a sector benchmark.

Their average days sales outstanding sits at 48 days instead of the 30 they invoice on. That gap of 18 days, spread across a year of revenue, works out to roughly CHF 118,000 tied up in receivables at any given moment. It’s not lost money. It’s cash sitting with clients instead of in the company’s account, often covered in the meantime by a bank overdraft that costs interest.

Alongside that, someone (usually the office manager or the owner) spends an estimated 6 to 8 hours a month tracking who’s paid, updating a spreadsheet, drafting reminder emails, and deciding when a phone call is warranted. It’s not a huge time sink on its own. What it costs is attention and goodwill: the same person who chases invoices also has to keep the client relationship warm, and those two jobs pull in opposite directions.

The pattern tends to repeat itself. A client pays late once, nobody follows up firmly, and the next invoice to that same client drifts even further past due because there was no consequence the first time. Left alone, average DSO creeps upward year over year. Rarely is any single client being deliberately difficult; there’s simply no consistent mechanism enforcing the payment terms everyone agreed to at the outset.

Getting the reminder cycle right takes more than bolting an email template onto the current process. It usually means rethinking who owns which step and why the invoice sat untouched in the first place. That’s the kind of scoping work covered under process optimization: mapping where invoices actually get stuck before deciding what to automate.

How the Agent Handles the Reminder Cycle, Step by Step

The agent’s job is narrow by design. It watches, drafts, sends, and reconciles. It doesn’t negotiate and it doesn’t decide who gets sued.

1. It watches open invoices in your accounting system. Connected to platforms like Bexio, Abacus, or Sage via API, the agent tracks due dates and payment status continuously instead of waiting for someone to run a report. The moment an invoice crosses a threshold (due today, 10 days overdue, 30 days overdue), it acts.

2. It runs a tiered reminder sequence. A friendly nudge goes out the day payment is due, in the client’s own language, restating the QR-bill reference and IBAN. If the invoice is still open ten days later, the tone shifts to a formal reminder with clearer payment terms stated. Past that, typically around 30 days, the agent stops sending and flags the account for a person to decide the next step.

3. It answers the routine replies. A large share of what comes back is predictable: “can you resend the invoice,” “what’s your IBAN,” “can we split this into two payments.” The agent handles the first two automatically. Installment requests get routed straight to a person, since agreeing to a payment plan is a judgment call that depends on the relationship and the amount involved.

4. It reconciles automatically once payment lands. By matching incoming bank statements (camt.053 files) or QR-bill reference numbers against open invoices, the agent closes the loop without anyone touching a spreadsheet. The reminder sequence for that invoice stops the moment the match is confirmed.

5. It sends a weekly receivables digest to the owner. A short summary: total outstanding, which accounts are aging past 30 or 60 days, and which reminders went out that week. It’s the kind of visibility most SMEs currently get only when someone remembers to pull it together, usually right before a cash crunch makes it urgent.

This overlaps with the exception-handling work covered in AI agent invoice processing on the payables side: the same agent architecture that reads incoming supplier invoices can watch outgoing ones, just pointed in the other direction.

Where This Differs From a Scheduled Email Tool

Plenty of accounting platforms already offer basic reminder scheduling: send an email X days after the due date, done. That covers step two above and nothing else. It has no memory of what the client replied last time, no way to route a reply to the right person, and no idea whether the invoice was actually paid until someone manually marks it closed.

The gap shows up fast. A scheduled email tool will happily send a formal overdue notice to a client who paid yesterday, because it has no reconciliation step. It will also send the exact same wording to a first-time late payer and a client who has been late nine times in a row, because it has no memory of history.

An agent connected to both the accounting system and the bank feed avoids both problems, but it also needs clean underlying data to work from. If invoice records are inconsistent, or client contact details are out of date, that gets fixed first. No agent compensates for a messy source system; it mainly makes the mess visible faster.

What Stays Human

The agent drafts and flags. It does not decide.

• Escalation to formal reminders or a collection referral. The agent flags an account as chronically late; a person decides whether to escalate, wait another cycle, or call the client directly.
• Payment plans and any discount or write-off. These are relationship and cash-flow calls that require someone who knows the client and the numbers behind the business.
• Anything with legal weight. Formal debt collection in Switzerland runs through a defined legal procedure at the local debt collection office (Betreibungsamt). Starting that process is a decision only a person should make, with the full context of the relationship in view.

If your firm already works with a Treuhand or fiduciary on receivables and VAT, the reminder agent sits upstream of that relationship rather than replacing it. Our piece on how Swiss fiduciaries use AI agents in accounting work covers where that boundary typically sits.

What This Costs and When It Pays Back

A pilot integration, connecting the agent to your accounting system and an email or SMS channel, typically runs CHF 10,000 to 25,000, depending on how many systems it reads from, how many languages and reminder tiers you need, and how much of your invoice data is already structured versus scattered across PDFs. Running costs after that are modest: platform hosting, messaging fees, and periodic tuning as your client mix changes. Expect the low end of that range if you’re on a single accounting platform with one or two languages, and the higher end if you’re reconciling across multiple entities or legacy systems that need custom connectors.

The bulk of the payback case rests on DSO. Shaving even 10 days off a 48-day DSO on CHF 2.4 million in revenue frees close to CHF 65,000 in working capital, money that no longer needs to sit on a bank overdraft or come out of a credit line. The 6 to 8 hours a month reclaimed from manual chasing adds a smaller, secondary return, usually redirected toward client work rather than counted as a hard saving.

That math holds up better the more invoices you issue and the more consistent your client base is. It gets shakier for a handful of invoices a month or for a business whose late payers are a small, fixed set of accounts that reminders won’t change anyway. Run the numbers on your own DSO and invoice volume before committing to a build; it takes an afternoon with your accounting export and tells you more than any generic ROI table.

Who This Fits, and Where It Doesn’t

A good fit if:

• You issue 30 or more invoices a month to a recurring base of B2B clients
• Your accounting platform (Bexio, Abacus, Sage, or similar) exposes an API
• DSO drag is visibly affecting cash flow or your credit line usage
• Clients span more than one language and the current reminder tone is inconsistent between them

Probably not worth building yet if:

• You issue a small number of invoices a month; a shared reminder template and a calendar entry cover this fine
• Most of your clients are large accounts on fixed, centralized payment runs (retail chains, public sector, big corporates paying on scheduled batch cycles), where a reminder email doesn’t move their internal process at all
• Your invoicing is still largely paper-based or lives in spreadsheets with no system to connect to

The bookkeeping side of this problem, matching payments and keeping the ledger current once invoices are paid, is covered in more depth in automating bookkeeping with AI agents, which is worth reading alongside this if reconciliation itself is the bigger pain point for your team.

Where to Start

The fastest way to know if this pays back for your business is to look at your actual numbers: current DSO, invoice volume, and how many hours someone is really spending chasing payment each month. Those three figures tell you most of what you need before committing to a build.

Book a 30-minute call with Orange ITS and we’ll walk through your receivables data together, tell you honestly whether the payback case holds up at your volume, and scope what a pilot would look like if it does.

Book a call with Orange ITS