How many invoices can one AP clerk handle?

Published benchmarks put a manual AP clerk somewhere between 1,000 and 2,000 invoices a month. The range is that wide because the number depends almost entirely on what arrives, not on who is doing the keying.

Two clerks with identical skill and identical volume can differ by 40% based on document mix alone.

130 productive hours is the real monthly ceiling

Start with time available, because most staffing models start with headcount instead.

A full-time month is about 170 hours. Subtract PTO accrual, meetings, month-end close support, vendor calls, training, and the ordinary interruptions of an office, and productive keying time lands nearer 130.

At 10 minutes an invoice that is 780 invoices. Not 2,000.

If a benchmark says 2,000, it is either counting a highly templated population like utility bills, or counting invoices that arrived already coded, or it is aspirational.

An exception costs 2.5x a clean invoice, so the mix sets the number

This is the part that makes single-number benchmarks misleading.

A clean invoice is roughly 10 minutes: open, save, find the vendor, key the header and lines, pick the GL account and property, submit. An exception is roughly 25, because it involves a second person and usually an email to a vendor who will not reply today.

Run the same clerk against three different document mixes:

Same person, same hours, a 31% swing in output. Nobody got better or worse at their job.

That is why "invoices per clerk" is a bad staffing metric on its own and "invoices per clerk at our exception rate" is a usable one.

Property management runs a worse mix than the benchmarks assume

The published figures generally come from environments where a vendor has a billing department and sends one clean PDF.

Multifamily does not work that way. The population arriving in the AP inbox at a 24,000 unit portfolio includes:

Single-page vendor invoices. Multi-page statements with one invoice inside. Batch PDFs where a vendor stapled nine invoices for nine properties into one file. Handwritten invoices photographed on a truck dashboard. W-9s. Certificates of insurance. Month-end statements summarizing invoices already paid.

Only the first group is what a benchmark means by "an invoice." The rest still consumes a clerk's time, and some of it is not payable at all.

Adding a clerk buys less capacity than the second clerk's salary suggests

Two structural reasons.

Coordination cost. Two clerks working the same supplier master develop different habits, and coding consistency degrades. That shows up at close, not in the productivity number.

The exception tail does not split evenly. Exceptions cluster around specific vendors and specific properties. A second clerk absorbs volume but the hard cases still route to whoever knows that vendor, so the bottleneck moves rather than widening.

The honest version: the second clerk gets you maybe 70% of the first clerk's throughput on total volume, and less than that on the population that actually causes delay.

At 98% automation the clerk's job is the 2%

ResProp runs a 98% automation rate across 24,000-plus units, meaning 98% of invoices require no human touch. The 2 in 100 that reach a person are the clerk's job now.

Volume grew and ResProp added zero AP hires. The team stopped re-keying and started reviewing exceptions, which is both a better use of an experienced AP person and the work that actually needs judgment.

The measured outcome was 40 hours a month saved per team and more than $150,000 a year in cost. Roughly $120,000 of that is the replaced per-page tool, which was tracking toward nearly $10,000 a month.

Do not use a vendor's FTE-equivalence tile as your staffing plan, ours included. Capacity equivalent is a different thing from headcount removed, and the clerk you keep is the one who knows your vendors.

What automation does not change about staffing

Being straight about the limits, because this is where AP vendors oversell.

You still need someone who knows your vendors. Extraction reads what is on the page. Deciding whether a $4,200 plumbing invoice at a 1966 property is reasonable is judgment, and no model has your context.

You still need someone to chase. When an exception is genuinely the vendor's error, somebody has to call. That call takes as long as it always did.

You still need a Controller who owns the chart of accounts. Coding automation learns from your history, so if two properties code the same expense differently the model learns both, faithfully. Automating an inconsistent GL produces a faster inconsistent GL.

Work out your own number this week

Three inputs, none of which requires a project.

Count last month's posted invoices. Count how many required a second touch, which you can usually approximate from however you track "things AP had to chase." Divide to get your exception rate.

Then: productive hours, times 60, divided by (clean minutes times clean share plus exception minutes times exception share).

If the answer is well below what your team is actually producing, they are working past 130 productive hours and you have a retention problem forming rather than a productivity win.

What is your exception rate?