The accounts receivable function at most mid-market companies runs on a workflow that has not changed meaningfully in a decade. AR staff generate aging reports each Monday, manually sort open invoices by amount or days overdue, send reminders in batches using templates with a customer name inserted at the top, and move down the list until time runs out. The 40 invoices they did not reach will wait until next Monday.
This works well enough when payment behavior is predictable and volumes are manageable. Neither condition holds reliably in 2026. Average payment cycles in B2B commerce have lengthened. Finance teams have not grown in proportion to the number of accounts they manage. And customers increasingly expect communication that reflects an understanding of their specific account relationship, not a form letter with their company name in the salutation.
The Playbook That Worked Until It Didn't
The batch-reminder model has three structural weaknesses that become more visible as invoice volumes scale. First, it treats all customers the same. A customer who reliably pays on day 38 but never responds to email receives the same reminders as a customer who processes invoices on Tuesday mornings and reads everything. The message lands at the same time, in the same channel, in the same tone. One of those customers needed a reminder three days earlier. The other never needed it at all.
Second, the batch model is reactive, not predictive. AR staff work from what has already happened (a payment is overdue) rather than from signals about when action is most likely to produce a result. The timing of a reminder is determined by the AR team's weekly schedule, not by when the customer is most likely to respond.
Third, manual prioritization selects on the wrong variable. Sorting by invoice amount captures the highest-value outstanding items, but it does not account for collectability. A 500,000 yen invoice from a customer with a dispute in progress is not the same priority as a 500,000 yen invoice from a customer who simply forgot to submit the payment request internally.
What AI Changes in the Collections Equation
When we talk about AI in AR collections at AccordX, we mean three specific capabilities working together on each invoice. Timing prediction reads each payer's historical response patterns and selects the moment a follow-up is most likely to produce action. Channel routing determines whether email, SMS, or a phone-coaching prompt is the right medium for that particular customer at that particular aging stage. Message generation drafts the specific text, calibrating tone and urgency level to match the relationship history and the invoice status.
None of these capabilities are new individually. The meaningful change is that they operate per customer, not per campaign. A customer who consistently responds to Tuesday morning email gets a Tuesday morning email. A customer who has ignored three emails but opened an SMS immediately after the last late payment gets routed to SMS. This is not segmentation by industry or company size. It is behavior-driven routing on the individual account level.
A Realistic Picture from Current Use
Consider the kind of scenario we see in early pilots: a finance team at a mid-size distribution company managing roughly 180 active accounts across multiple product lines. Two AR staff split responsibility by customer segment. Each Monday they spend about an hour and a half generating reports, categorizing overdue accounts, and preparing the week's reminders. By Thursday, the lower-priority accounts have not been contacted. By the following Monday, some of those have moved into the 31-60 day bucket.
When the team connected their invoice data to AccordX, the first change they noticed was not a dramatic reduction in DSO. It was that the Thursday backlog disappeared. Every account in the queue received an outreach at an appropriate time regardless of where it fell on the manual priority list. The outcome visible within the first billing cycle was more consistency, not necessarily faster average payment. That came later, as the timing model accumulated enough data to sharpen its predictions for each account.
Where Finance Teams Reasonably Push Back
The most common concern we hear is about customer relationships. AR managers at companies with long-standing accounts are cautious about introducing automation precisely because those relationships are built on personal familiarity, and they do not want a system to send something that damages trust or feels impersonal. That caution is reasonable. It is not a reason to avoid automation, but it is a reason to implement it carefully.
The rule we have found useful: automate the contact moments that do not require human judgment, and keep humans in the loop for the moments that do. A day-7 reminder to a current account with no dispute history requires almost no judgment. A day-60 follow-up to a key account that has raised a pricing concern requires a human response. An automated system should identify the second situation and route it out of the queue, not process it the same way it processes the first.
We are not saying that full automation is the goal. We are saying that the specific judgment calls worth preserving are the ones involving relationship-sensitive escalation, active disputes, and accounts that are showing early signs of financial difficulty. The volume work, routine reminders, channel switching, and follow-up scheduling, is where automation creates the most value with the least risk.
The Starting Point for Most Finance Teams
The teams that have adopted AR automation most successfully did not start by automating everything. They started by identifying the segment of their portfolio where manual follow-up was most mechanical: current-due accounts with no open disputes and no special relationship handling requirements. For most companies, that is 60 to 70 percent of open invoices by count, even if it represents a smaller share by value.
Automating that segment first freed up AR staff time for the accounts that genuinely required human involvement. Once the baseline behavior data accumulated, the timing and channel recommendations became more useful, and the team was in a better position to decide which higher-complexity accounts were ready for partial automation as well.
The shift toward AI in AR collections is not about replacing the finance team's judgment. It is about ensuring that judgment is applied to the accounts that actually need it, rather than being diluted across 150 routine reminders that a well-configured system can handle accurately and at scale.
