The CFO's highest-ROI AI investment is sitting in accounts receivable

Boards are demanding AI ROI. CFOs are writing business cases for AI tools that cost seven figures but produce benefits that are hard to attribute. Meanwhile, most finance teams are still chasing overdue invoices manually—sending the same email template on day 60, day 75, and day 90, and calling it a process.

That gap is worth examining. Before a software company acquires another AI tool for demand generation or product recommendations, it's worth asking whether the finance operation itself has been automated. Accounts receivable is one of the clearest places in a technology company where AI produces measurable, attributable outcomes—and most companies haven't gotten there yet.

What the ROI argument actually looks like

A company with $40M in ARR typically carries $4M–$7M in outstanding receivables at any given moment. If 6% of that becomes delinquent (a conservative rate for software companies with enterprise customers), that's $240K–$420K in accounts at risk. Industry data shows the average B2B company recovers less than 50 cents on the dollar for accounts that age past 12 months. The gap between "placed within 60 days" and "placed at 14 months" is not 15%—it's often 60–70 cents per dollar.

AI applied to the early stages of that pipeline—intelligent dunning sequencing, payment intent prediction, anomaly detection on accounts that change behavior—doesn't just speed up collections. It changes which accounts ever become delinquent in the first place. That's a compounding return, not a one-time efficiency gain.

What "AI in AR" actually means (versus marketing language)

There's a lot of vendor noise in this space. When a software platform says it uses "AI-powered collections," that usually means one of three things:

• Intelligent dunning sequencing: The system decides when to send a reminder, which channel to use (email, SMS, in-app), and what tone to take based on customer behavior. A customer with a spotless 4-year payment history who is 12 days late gets a different message than one who was 45 days late twice last year.
• Payment prediction scoring: ML models trained on payment behavior, company signals (job postings, funding announcements, press releases), and AR history to score likelihood of collection. High-risk accounts get escalated earlier; low-risk ones get a lighter touch that preserves the relationship.
• Dispute classification: Natural language processing on dispute responses to identify which objections are legitimate (billing errors, genuine product issues) vs. which are stall tactics ("we need approval from procurement") that should be escalated immediately.

All three are real. All three work. None of them is magic—they require clean data, a consistent process upstream, and a human or agency to handle escalated accounts that the automation can't resolve.

Where AR automation ends and collection agencies begin

CFOs sometimes ask whether AI-powered AR software replaces the need for a collection agency. The honest answer: it replaces the need for mass-blast dunning emails and manual spreadsheet tracking. It does not replace the need for a firm that can negotiate with a debtor's AP team, document a claim for escalation, or recover accounts from companies that have gone silent.

The practical line is around 90 days. Automated AR tools are optimized for the 0–90 day window—nudging customers who have the money but haven't prioritized your invoice. Once an account has crossed 90 days with no response and no payment plan, you're dealing with a different kind of problem: a company that either can't pay, has decided not to, or is using delay as a negotiation tactic. That's where an external agency takes over, and AI doesn't change that calculus.

What good AR automation does is ensure you're placing accounts at 90 days instead of 180—which, given the decay curve on receivables, is a significant recovery improvement before the agency ever gets involved.

The build-vs-buy mistake CFOs are making

Many mid-market software companies ($20M–$200M ARR) are building AR automation in-house using workflow tools and LLM APIs. The CFO sponsors the project, the engineering team builds dunning sequences in Zapier or an internal tool, and six months later it's half-working and owned by no one.

This is a version of the same mistake that plagued CRM implementations in the 2010s. The technology works—the process and ownership don't. A collections process that an engineer built and a finance analyst is supposed to maintain will be abandoned within 18 months. Off-the-shelf AR automation (Tesorio, Invoiced, YayPay, Gaviti) exists specifically because this problem is worth solving at the product level, not the Zapier level.

Where to start if your board is asking for AI ROI

If your company needs to show AI ROI in FY2026 and you haven't yet automated your AR process, that's the easiest case to make. The math is attributable. Before: DSO of X days, bad debt rate of Y%, manual collections cost of Z hours per week. After: measurable improvement in each. Unlike AI investments in product or marketing, the output of AR automation is a dollar amount—and it compounds with your revenue growth.

The other advantage: AR automation doesn't require a six-month implementation. Most platforms are live in 30–60 days. The ROI is visible in the first quarter. For a CFO trying to build internal confidence in AI investment discipline, that's a useful first win.

Start with your aged receivables report. If anything in that report is over 90 days, you have a specific, solvable problem—and the path to solving it doesn't start with another AI pilot. It starts with a placement decision.