Autonomous accounts receivable: the self-driving AR function

Autonomous accounts receivable is an AI agent that runs the collections cycle end to end and is accountable for the cash result. Instead of a person chasing invoices, applying payments, and chasing down disputes, the agent does the work across the whole ledger, continuously, and escalates only the cases that need a human decision. It is measured the way you would measure a collections team: cash recovered, DSO, and the share of receivables it clears without help.

The word that matters is autonomous. A lot of software calls itself AI but waits for a person to drive it. Autonomous AR is the opposite. You give it the ledger and the rules of engagement, and it decides what each account needs and acts on it. The team's job changes from doing the work to supervising the agent that does it.

What autonomous AR looks like day to day

Picture the morning of an AR manager running an autonomous function. There is no queue of reminders to send, because the agent sent them overnight, each one timed and worded to the specific account. There is no stack of unapplied payments, because the agent matched the cash that landed and posted it. The aging report is already current.

What is left is a short list of decisions. A customer asked for a 60-day extension on a six-figure balance. A dispute came in that needs a credit memo only finance can approve. A strategic account went quiet and the agent thinks sales should step in. The manager works that list, makes the calls, and the agent carries out what they decide.

The volume that used to fill the day is gone. The agent handled it. What reaches a person is the handful of judgment calls that genuinely need one. That is the shape of an autonomous function: the machine runs the routine, the human runs the exceptions.

The full AR cycle an agent can run

Autonomous AR is not one task done well. It is the whole post-sale cycle, with the steps connected so nothing falls between them.

• Collections and dunning. The agent decides which accounts to contact, when, and how. It writes the message to the account's situation, escalates tone as an invoice ages, and pauses outreach the moment a dispute or a promise to pay is open, so it never chases a customer who already engaged.
• Cash application. Incoming payments get matched to open invoices and posted, including the messy ones: partial payments, lump sums covering many invoices, and remittance that arrives separately from the funds. The ledger reflects reality without anyone keying it in.
• Dispute and deduction handling. When a customer pushes back, the agent recognizes it as a dispute, gathers the proof of delivery or the contract or the pricing, routes it to whoever can resolve it, and tracks it to closure.
• Credit and risk. The agent watches each account's payment behavior and adjusts how hard it pushes, leaning on shaky accounts sooner and giving reliable ones room.
• Reporting. DSO, aging, and the collections forecast stay current on their own, because the agent is the one taking the actions that move those numbers.

The leverage comes from the connection. The same agent that read the dispute is the one that paused the dunning, so the handoff that used to drop never happens.

Levels of AR autonomy explained

Not every team is ready to hand over the whole ledger on day one, and they should not have to. Autonomy comes in levels, and you move up them as trust builds.

• Level 0, manual. A person does the work. Software just records it.
• Level 1, assisted. The system suggests an action, drafts the email, proposes the match, and a person approves each one. This is where most "AI" AR tools stop. It still needs a human in the loop on every item, so it does not actually reduce the work much.
• Level 2, supervised autonomy. The agent acts on its own inside set thresholds and escalates anything beyond them. It sends the reminders, applies the clean and the messy cash, and routes disputes, while a person approves the high-value or unusual moves. This is where autonomous AR earns its keep.
• Level 3, broad autonomy. The agent's mandate widens as its track record proves out. Thresholds rise, more segments come under its control, and the exceptions that reach a person get rarer.

The point of naming the levels is that you choose where to start and you can see where you are. You do not flip a switch from manual to fully autonomous. You move up as the agent earns it.

Human oversight and approval thresholds

Autonomous does not mean unsupervised. The model that works is supervised autonomy: the agent runs inside guardrails you set, and crosses a line back to a person whenever a case falls outside them.

You define the thresholds. A payment plan over a certain length needs approval. Any write-off goes to finance. Balances above a dollar amount get a human review before the tone escalates. Named strategic accounts always loop in their owner. Inside those lines the agent acts freely; outside them it stops and asks.

Two things make this safe. First, every action is logged with the reason behind it, so you can see exactly what the agent did and why, and audit it after the fact. Second, the thresholds are yours to tighten or loosen. You start conservative, watch what the agent does, and give it more room as it proves it deserves it. Oversight is the control surface, not a bottleneck on every item.

Results: DSO, recovery, and team capacity

The reason to run AR autonomously is the numbers, and three of them move.

DSO comes down because the work happens on time, every time. Reminders go out the day they should, not when someone gets to them. Cash gets applied the day it lands, so the ledger is not overstating what is still owed. A few days off DSO on a large book is a material amount of cash freed up. See how AI drives DSO reduction for the mechanics.

Recovery improves because nothing slips. Every account gets the right follow-up at the right time, and the disputes that used to sit and silently age get worked to closure instead. The receivables that would have aged into bad debt get caught while they are still collectible.

Capacity opens up because the routine volume is gone. The same team can run a book several times larger, because they are no longer the ones sending reminders and keying in matches. They supervise the agent and handle the exceptions. Headcount stops being the thing that caps how much you can collect.

How to phase into autonomous AR

Do not try to automate everything at once. Phase it in where the value is clear and the risk is low.

Start with the work that is highest volume and most rule-bound, usually dunning and cash application. Those consume most of an AR team's hours and follow patterns an agent learns fast. Set conservative thresholds, let the agent run a defined segment, and watch what it does against what your team would have done.

Measure the change. Track DSO, recovery, and hours reclaimed on that segment against the rest of the book. When the agent is matching or beating the manual baseline, widen its mandate: more accounts, higher thresholds, the next workflow. This is supervised autonomy in the language of autonomous finance more broadly, and the same playbook applies. Prove it small, then scale it. The agent earns the ledger one segment at a time, and the work it owns only grows.

How Rex runs autonomous AR

Rex is an autonomous AR agent. It is not a dashboard you read or a copilot that drafts emails for you to send. It runs the collections cycle across your whole ledger, on its own, and is accountable for the cash it recovers and the DSO it brings down. It chases the invoices, applies the payments, works the disputes, and adjusts to each account, escalating only the cases that need a human decision.

You stay in control through the guardrails. Rex acts inside the approval thresholds you set, surfaces the exceptions that cross them, and logs every action with the reasoning behind it, so you can audit what it did and trust it with more over time. The team stops doing the routine work and starts supervising the agent that does it. To go deeper on what an AR agent is and how it decides, read what an AI AR agent is.

See how Rex runs accounts receivable end to end.