AI Agents for Accounts Receivable: Automate Collections and Accelerate Cash Flow

Accounts receivable is where cash flow lives or dies. A company can book record revenue and still run out of cash if customers don't pay on time. The average B2B company has a Days Sales Outstanding (DSO) of 45–60 days, and 10–15% of invoices go past due. For a $50M revenue company, that's $5–7.5M tied up in late payments at any given time—capital that could fund hiring, inventory, or product development.

The AR problem isn't that companies don't follow up. It's that following up at scale, with the right message, at the right time, for the right invoices, requires more bandwidth than most finance teams have. A controller managing $20M in outstanding receivables might have 500+ open invoices across 200 customers, each with different payment terms, dispute histories, and communication preferences. Manual follow-up means spreadsheets, email templates, and a lot of copying and pasting.

AI agents are changing this equation by automating the high-volume, repetitive parts of collections while keeping humans in the loop for relationship-sensitive situations.

What AR agents automate

Payment reminder sequences. The agent monitors invoice aging and sends personalized reminders based on configurable rules. A friendly reminder 5 days before due date. A firm follow-up 3 days after. An escalation notice at 15 days past due. Each message references the specific invoice number, amount, and payment options—not a generic "you have an outstanding balance" email. The agent adapts tone and frequency based on the customer's payment history: reliable payers get a gentle nudge; chronic late payers get earlier, more direct outreach.

Cash application. When payments arrive, they need to be matched to the correct invoices. This sounds simple until you're processing 500 payments a day where the remittance info says "Invoice 2024-1234" but the actual invoice number is "INV-2024-01234," or the customer paid three invoices with one check and no remittance advice. AI agents match payments to invoices using fuzzy matching on reference numbers, amount combinations, and customer history—automating 80–90% of cash application that would otherwise require manual matching.

Dispute detection and routing. When a customer replies to a payment reminder with "we never received these goods" or "the invoice amount is wrong," the agent classifies the dispute type, creates a case record, pulls the relevant documentation (purchase order, delivery confirmation, contract terms), and routes it to the appropriate person for resolution. Instead of a collector spending 20 minutes gathering context for a dispute, the agent presents a complete case file in seconds.

Dunning escalation. The agent manages the entire dunning lifecycle: from soft reminders to demand letters to collections agency referral. At each escalation stage, it evaluates whether the situation warrants the next step based on the amount, customer relationship value, payment history, and any open disputes. A $200 invoice from a $2M/year customer gets a different escalation path than a $200 invoice from a one-time buyer.

Customer payment portal notifications. The agent monitors whether customers have viewed invoices in your billing portal, opened payment reminder emails, or started (but not completed) a payment. These behavioral signals help prioritize follow-up: a customer who opened the reminder three times but hasn't paid likely has a question or issue, while one who never opened it might just need a different channel (phone call, text, postal mail).

Promise-to-pay tracking. When a customer commits to paying by a specific date, the agent records the promise, sets a follow-up trigger if the payment doesn't arrive by that date, and maintains a history of kept vs. broken promises per customer. This data informs credit decisions and collection priority.

Implementation approach

Phase 1: Invoice aging analysis (week 1). Connect the agent to your ERP or accounting system. Analyze the current AR aging: how much is current, 30-day, 60-day, 90-day+? Identify the top 20 customers by outstanding balance and the most common dispute types. This baseline tells you where the agent will have the most impact.

Phase 2: Reminder automation (weeks 2–3). Configure reminder sequences for each aging bucket. Start with email-only outreach. Use your existing email templates as a starting point, then let the agent personalize based on customer data (name, invoice details, payment history). Run in shadow mode for one week—the agent drafts reminders but a human approves before sending. Review for accuracy and tone.

Phase 3: Cash application (weeks 3–4). Connect the agent to your bank feed or payment processor. Train it on your historical payment-to-invoice matching patterns. Start with high-confidence matches (exact amount + exact reference number) and gradually expand to fuzzy matches. Human review queue for matches below the confidence threshold.

Phase 4: Dispute handling and escalation (weeks 4–6). Configure dispute classification rules and routing logic. Connect to your order management and shipping systems so the agent can pull supporting documentation. Define escalation rules based on your credit policy. Go live with full automation, monitoring agent decisions daily for the first month.

Measuring ROI

The primary metric is DSO reduction. Most companies see a 15–30 day improvement within 90 days of deploying an AR agent, driven by:

• Faster first contact: Reminders go out on time, every time—no more invoices falling through the cracks because the collector was busy with other accounts
• Higher contact rate: The agent sends 100% of scheduled reminders; human collectors typically manage 40–60% of their assigned portfolio each cycle
• Faster dispute resolution: Disputes are identified, documented, and routed in minutes instead of days
• Better cash application: Payments are matched same-day instead of sitting in a suspense account for a week

Secondary metrics include:

• Collection effectiveness index (CEI): Percentage of receivables collected versus total available for collection. Target 80%+ with an AI agent.
• Bad debt write-off rate: Typically drops 20–40% because more invoices are followed up before reaching uncollectable status.
• Cost per collection: Drops 50–70% as the agent handles volume while collectors focus on high-value negotiations.

Integration considerations

ERP connectivity. The agent needs read access to invoice data (amounts, due dates, payment terms, customer contacts) and write access to update collection status and log activities. Common integrations include NetSuite, QuickBooks, Xero, SAP, and Sage.

Email and communication channels. The agent sends reminders via email (using your domain for deliverability), and optionally SMS or portal notifications. It should track opens, clicks, and bounces to optimize timing and channel selection.

Bank feeds. For cash application, the agent needs access to incoming payment data—either through direct bank feeds, payment processor APIs (Stripe, PayPal), or lockbox services.

CRM. Optionally connect to your CRM so the sales team has visibility into collection status for their accounts. A sales rep should know if their largest customer is 60 days past due before their next QBR.

When to keep humans in the loop

AR automation works best as augmentation, not full replacement. Keep humans involved for:

• Key accounts where the relationship is more valuable than any single invoice—a $5M/year customer gets a phone call from the CFO, not an automated email
• Complex disputes involving contract interpretation, service level disagreements, or legal implications
• Final escalation decisions—sending an account to collections or initiating legal action should always require human approval
• Credit hold decisions that affect the customer's ability to place new orders
• Negotiations on payment plans, early payment discounts, or settlement offers

The agent handles the 80% of AR work that is process-driven and repetitive. Humans handle the 20% that requires judgment, empathy, and relationship management.