Accounts receivable automation: which steps to automate first

Accounts receivable automation comes down to putting four weekly recurring steps on fixed rules: bucketing open invoices by age, sending reminders on a set schedule, following up on payment arrangements, and tracking payment behaviour per customer. The conversation with a customer who does not pay stays human work. The administration around it should not be.

For a controller this is a liquidity project rather than a collections project. In a portfolio of € 3 million in open receivables, 15 to 25 percent is typically past due. That is half a million already earned, already invoiced and simply not yet collected: cash you release without adding a single euro of margin. The bottleneck is rarely the willingness to call. It is the hours that go into building the list of who to call.

What accounts receivable automation takes over, and what it does not

Break the AR cycle apart and six steps remain. How much of each can run without judgement differs sharply.

Step	Can be automated	What stays human work
Pulling open items and sorting them into age buckets	Fully	Checking that the source connection still runs
Filtering out credit notes, prepayments and intercompany	Fully, with rules	Assigning new relations to the right group
Sending reminders at fixed moments	Fully	The exception: a customer with an open dispute
Recording and following up payment arrangements	Largely	Making the arrangement itself
Monitoring payment behaviour and DSO per customer	Fully	The conclusion: credit limit, prepayment or delivery stop
Escalation to collections or write-off	Flagging only	The decision, and its effect on the relationship

Four of the six steps are pulling, filtering and sorting data. That ratio matches the rest of the finance administration, for the same reason: work without judgement is the work you can automate away. How the ratio plays out across the full record-to-report chain is covered in accounting automation.

How to approach accounts receivable automation

1. Start with the aging analysis, not with the reminder. Until you can see how the open balance is spread across the buckets current, 1 to 30, 31 to 60, 61 to 90 and over 90 days overdue, you are automating a dunning process without knowing where the delay sits. Smartbooks shows that breakdown live per entity and consolidated, with the amount and the invoice count in every cell, and lets you click through to the underlying invoices and the relation behind them.
2. Clean the list before you act on it. Intercompany receivables, credit notes and prepayments distort every aging analysis, because there is no real payment term behind them. Filter intercompany relations out structurally instead of by hand every month, and group branches of the same customer into one relation. Otherwise three people call the same purchasing department.
3. Fix the reminder schedule in the system that issues the invoice. Three moments are enough: a friendly reminder two days after the due date, a second one on day fourteen with the full open statement attached, and on day thirty a message that goes to both the contact person and the purchasing department. The schedule belongs in the package that sent the invoice; the reporting layer around it measures whether the schedule works.
4. Drive planned receivables from a DSO metric. The planned receivables balance then moves along with your revenue forecast, so a lengthening payment term shows up in the cash flow forecast instead of only in the actuals. For the ratio itself, see DSO calculation.
5. Let AI build the weekly action list. This step holds the most hours and the least judgement. A question such as which customers are more than 30 days past due, sorted by amount, excluding intercompany returns that list straight away in the AI Controller chat or through the MCP connection with Claude, with the journal entries underneath so every line can be checked. What used to be an export and an afternoon of pivot tables is now a question.
6. Measure on one fixed basis. Put DSO, the percentage past due and the amount in the over-90-days bucket side by side in the same monthly report. Without a baseline you will not know, six months later, whether the schedule worked or whether one large customer simply settled up.

Worked example: a portfolio of 1,400 open invoices

A neutral example. A group with three operating companies has € 3.1 million outstanding across 1,400 invoices. After filtering out intercompany, the aging analysis looks like this.

Bucket	Amount	Invoices
Current	€ 2,310,000	1,010
1 to 30 days overdue	€ 480,000	250
31 to 60 days overdue	€ 185,000	82
61 to 90 days overdue	€ 71,000	34
Over 90 days overdue	€ 54,000	24
Total	€ 3,100,000	1,400

Watch the ratio between amount and count. The 24 invoices over 90 days are 1.7 percent of the invoice count and 1.7 percent of the amount: small, old items that usually hide a dispute. The liquidity sits in the 1 to 30 days bucket, at € 480,000 across 250 invoices, averaging € 1,920 each. Those are not problem customers but a payment rhythm running a few days behind, and that is exactly where a fixed schedule works and a phone call does not.

Activity	Before	After
Building the aging overview across three entities	3 hours per week	0 − always current
Stripping out intercompany and credit notes	1 hour per week	0 − set as a rule
Building and distributing the action list	1.5 hours per week	15 minutes
Sending reminders	2 hours per week	0 − on schedule
Calling and agreeing payment arrangements	1 hour per week	3 hours per week
Total	8.5 hours	3.25 hours

These figures illustrate the proportions; they are not a customer result. Note the line above the total: calling triples. That is the point of accounts receivable automation. Not less contact with customers, but more contact and less administration around it.

Limits and risks

• An unmatched payment becomes an unjustified dunning letter. Automation speeds up the mistake as well. A customer who has paid, but whose bank transaction has not been matched yet, receives a reminder that costs the trust ten phone calls built up. Run the schedule only after the day's bank transactions have been processed.
• Age says nothing about collectability. A 100-day invoice at a solid customer waiting for a credit note is a different case from a 35-day invoice at a customer in payment trouble. The buckets sort on time, not on risk; the weighting stays yours.
• Auditability. Every automated step has to trace back to the underlying invoice and journal entry, not to a balance. That is more than an auditor requirement: it is exactly what you need the moment a customer disputes an amount.
• Receivables balances are customer data. For an AI connection, use one that runs under your own login and access rights and that only reads. The Smartbooks MCP connection does that: it reads data and posts nothing, and a user sees only the companies they can already open in the application.
• AI presents an error with the same confidence as a fact. A model summarising a receivables list can add up a relation group incorrectly without any change in tone. Always trace an answer back to the invoices underneath before you call a customer.

Pitfalls in automating the AR process

• Starting to dun before the aging analysis is clean, so the first round mostly chases intercompany items and credit notes.
• Steering on average DSO alone. One large customer paying on day 75 hides a portfolio that otherwise sits neatly on day 28.
• Trying to build the reminder schedule in the reporting layer instead of in the package that issues the invoice. Measuring and sending are two different functions.
• Leaving branches of the same customer as separate relations, so three people call the same purchasing department about three invoices.
• Measuring success in reminders sent instead of in DSO days and the amount past due.

Frequently asked questions about accounts receivable automation

What should you automate first in the AR process?

The aging analysis, including the exclusion of intercompany and credit notes. It is the list every other step depends on, and as long as it is built by hand it costs hours every week and is already out of date by the next review meeting.

Can you use AI for accounts receivable automation?

Partly, and not in the part most people expect. AI is strong at building and summarising the action list: which customers, which amounts, which movement against last month. A decision on a credit limit or a delivery stop is a judgement with commercial consequences and belongs with a person. Give AI read access to your receivables position, not decision authority.

How many DSO days can this win?

That depends entirely on the starting point. If the overrun sits mainly in the 1 to 30 days bucket, a fixed reminder schedule is usually enough for a few days. If it sits beyond 60 days, there is an agreement or invoice quality problem that no schedule solves. Measure for three months before setting a target, or you will book a gain that came from the revenue mix.

Does Smartbooks send the dunning letters?

No. Smartbooks is the analysis and reporting layer: the aging analysis per entity and consolidated, payment behaviour per relation, DSO in your plan, and the option to ask questions about all of it through AI. Sending invoices and reminders stays in your accounting or invoicing package, and Smartbooks reads from there.

Treat this as a liquidity measure with a baseline rather than a software purchase: first a clean aging analysis, then the schedule, then the weekly list out of AI, and the freed-up hours into the conversations that actually matter. What a shorter payment term does to the rest of your current assets is covered in working capital; how to put the effect into your cash forecast, in the 13-week cashflow planning.