AI ROI in accounts receivable: modeling the business case

The ROI of AI in accounts receivable comes from outcomes an agent is accountable for, not from cutting software seats. The levers are recovered cash, lower DSO, reduced bad debt, and capacity freed from manual work. Model those four against the annual cost of the system and you have a defensible business case. For most companies, the cash freed by a DSO reduction alone covers the cost several times over.

This guide gives you the formulas and worked examples to build that case for any AI AR vendor. The numbers below use illustrative figures. Swap in your own balance, DSO, and bad-debt rate to size the case for your company.

The real costs of manual AR

Before you model the upside, name the cost of the status quo, because it is rarely a line item anyone sees. Manual AR is expensive in three ways.

First, slow collections trap cash. Every extra day of DSO is working capital sitting in your customers' bank accounts instead of yours, which you either do without or borrow to cover. Second, slow follow-up turns recoverable invoices into bad debt. Accounts that get chased late, or not at all because the team is buried, slide past the point of recovery. Third, skilled people spend their days on repetitive work: sending reminders, applying cash, chasing remittances. That is salary spent on tasks a system handles better.

There is a fourth cost that compounds the others: inconsistency. A manual team works the squeaky wheels and lets quieter accounts age, chases hard at month-end and lightly the rest of the cycle, and skips steps when someone is on leave. That unevenness is invisible in any report, but it is where DSO creep and avoidable write-offs come from. The cost is not just the hours spent, it is the cash lost because the work was not done consistently.

None of these show up neatly in a budget, which is why manual AR looks cheaper than it is. The business case for AI AR starts by making these hidden costs visible, then sizing the upside of removing them.

ROI levers: DSO, recovery, bad debt

Three outcomes drive most of the return. Model each one separately, then add them up.

DSO reduction. This is usually the largest lever. The cash freed by cutting DSO is your average daily sales times the number of days you remove.

Cash freed = (Annual revenue / 365) x days of DSO reduction

For a company with $200M in revenue cutting DSO by 8 days:

($200,000,000 / 365) x 8 = $4,383,562 freed

That is one-time cash unlocked. Its ongoing value is what that cash earns or saves you, for example at a 10% cost of capital:

$4,383,562 x 10% = $438,356 per year

Cash recovery. Working the whole ledger consistently recovers invoices that manual follow-up misses. If steadier collections lift your recovery rate by even 1 point on a large balance, multiply that point against the dollars at risk in your aging report.

Bad-debt reduction. Earlier, consistent action keeps more invoices from ever going bad. If you write off 0.8% of $200M ($1.6M) and cut that by a third:

$1,600,000 x 33% = $528,000 saved per year

These three lines, the annual value of freed cash plus recovery gains plus bad-debt savings, are the core of the case. For our example, that is roughly $438K plus recovery plus $528K, already approaching seven figures before touching labor.

Capacity and headcount reallocation

The fourth lever is people, and the honest framing matters. The strongest case is not firing collectors. It is letting the same team manage a much larger book while moving to higher-value work like disputes, strategic accounts, and analysis.

Size it by the hours reclaimed. If an agent takes over the repetitive volume, estimate the share of your AR team's time spent on reminders, cash application, and chasing remittances, then value those hours.

Capacity value = hours freed per week x 52 x fully loaded hourly cost

For a 4-person team where the agent absorbs 60% of a $90,000 fully loaded role:

4 x 0.60 x $90,000 = $216,000 of capacity redeployed per year

Whether you bank that as avoided future hiring or as more output from today's team is a business choice. Either way it belongs in the model as redeployed capacity, not as a layoff line, which is both the more credible and the more durable story.

The durability point is worth dwelling on. A headcount-savings case dies the first time the team grows back, and it makes the project a target. A capacity case survives growth, because the value is that the same team now manages a far larger book without the cost scaling with it. That is the structural win of automating the repetitive volume: your AR cost stops growing in lockstep with your customer count. For a company adding accounts every quarter, the avoided future hiring alone can outweigh every other line in the model over a few years.

Building a defensible ROI model

Put the levers together and net out the cost. The structure is simple.

Annual gain = DSO cash value + recovery gain + bad-debt savings + capacity value
Net gain = Annual gain - annual system cost
ROI = Net gain / annual system cost

Using our example, with a conservative $150,000 recovery gain and a $250,000 annual system cost:

Annual gain = $438,356 + $150,000 + $528,000 + $216,000 = $1,332,356
Net gain   = $1,332,356 - $250,000 = $1,082,356
ROI        = $1,082,356 / $250,000 = 4.3x

To keep it defensible, be conservative on every input, separate one-time cash unlocked from recurring annual value, and state your assumptions plainly so a CFO can challenge each one. A model that survives scrutiny is worth more than an aggressive number that does not. When you compare vendors on these numbers, demand the same rigor from each, which ties directly into how to evaluate AI AR vendors on outcomes rather than feature lists.

Payback timelines to expect

Payback is the cost divided by the monthly gain.

Payback (months) = annual system cost / (annual gain / 12)

In our example:

$250,000 / ($1,332,356 / 12) = 2.3 months

Real timelines are usually a few months to a year, because the agent ramps. You start supervised on a slice of the ledger, then widen its authority as trust builds, so the full DSO benefit lands a quarter or two in, not on day one. Model a ramp rather than instant full value, and the payback figure you present will hold up when the actual cash arrives.

Tracking ROI after go-live

A business case is a forecast. Prove it with the same metrics after launch. Baseline DSO, bad-debt rate, recovery rate, and team hours before the agent starts, then track them on a fixed cadence against that baseline.

Watch one number above the rest: DSO trend against where you started. It is the cleanest single proxy for whether the cash case is real. Pair it with the share of the ledger the agent works autonomously and the volume it escalates, so you can see both the result and how it was produced. Reporting actuals against the model also builds organizational trust, because the finance team sees the agent delivering the outcomes it was bought to own.

One caution on attribution. Some of a DSO improvement may come from factors outside the agent, like a strong cash quarter or a one-time clean-up of old accounts, so do not credit every gain to the system. The honest way to track ROI is to hold the agent to the metrics it directly controls, the percentage of the ledger worked, the consistency of follow-up, the recovery rate on accounts it touched, and treat the headline DSO drop as the result those actions produce. A model that survives this scrutiny after go-live is the one that earns the next phase of investment, because finance leaders trust numbers they can trace back to specific actions.

How Rex delivers and proves the ROI

Rex is an agentic AI AR agent that runs collections, cash application, and dispute resolution across your whole ledger on its own, accountable for the outcomes this model is built on: cash recovered and DSO down. Because Rex takes the actions itself and logs each one, the gains in your business case are measurable in production, not just projected on a slide. It works every account consistently, recovers invoices manual follow-up misses, and escalates the cases that need a person, so the DSO and bad-debt lines in your model are the ones Rex is measured against.

Model the case on your own numbers, then see how Rex moves them.