It is the third working day of the close. The controller is still chasing two bank reconciliations, a supplier statement that does not agree, and an accrual nobody owns. Ask the team where their week went and you will hear "the close", "invoices" and "emails". All true, and not much use. Finance team capacity gets discussed as a headcount question, when the first question is where the existing hours go. Many teams have never measured it.
A time audit measures it. It takes four weeks and costs very little. It belongs before any conversation about AI, because it shows which work is high in volume and rule-bound enough to hand over, and which work only looks that way from a distance.
Why finance team capacity is hard to see
The ERP counts transactions. It does not count the second and third touch on the same transaction, and that is where the hours hide.
An invoice keyed twice, a payment returned for wrong bank details, a reconciliation redone because a journal went to the wrong period: each looks like one unit of work in the system and costs two or three. Interruptions are the other blind spot. Sales asking about a customer balance, an approver who needs chasing, an auditor wanting a sample. That is real work, and nobody plans for it.
What to record, and for how long
Run the audit for four weeks, including one full month-end close. Shorter than that and you either miss the close or see only the close.
Process, not task. Give people a fixed list of 15 to 20 processes: supplier invoices, payment runs, cash application, collections, bank reconciliations, intercompany, journals and accruals, management reporting, tax and statutory work, audit requests. Free-text logging produces data nobody can add up.
Half-hour blocks, logged daily. Memory at the end of the week flatters first-pass work and forgets interruptions.
Rework as its own tag. Each block is tagged first-pass, rework or exception, or interruption. This tag matters more than any other column.
Volume from the system. Pull counts for the same four weeks from the ERP or workflow tool: invoices received, payments made, receipts applied, reconciliations completed, journals posted. Do not ask people to estimate them.
Tell the team plainly that this measures the process and will not be used to rate individuals. If they suspect otherwise, the rework tag goes quiet first.
A template for the finance team capacity audit
| Column | What to enter | What it shows |
|---|---|---|
| Process | One of the fixed list | Where hours sit |
| Monthly volume | Count from the system | Scale of the work |
| Total hours | Sum of logged blocks | Cost in time |
| Rework and exception hours | Blocks tagged rework or exception | Hours spent fixing, not doing |
| Hours per unit | Total hours ÷ volume | Effort you can compare month to month |
| Rework share | Rework hours ÷ total hours | How much of the process corrects itself |
| Rules written down? | Yes, partly, no | Whether the work can be specified |
| People involved | Names or roles | Dependence on one person |
| Upstream cause of rework | Free text, from the team | Where the fix really sits |
For a rough outside reference, APQC's Open Standards Benchmarking puts the cross-industry median total cost of processing accounts payable at $6.50 per invoice (6,546 companies), and the median time from receipt of invoice to payment at 15.0 days (2,226 companies). Ardent Partners' payables research, sponsor-underwritten and without a stated sample, reports an 18.4% average invoice exception rate. Definitions differ between sources, so use these for orientation rather than as targets.
Reading the result: where AI and better process fit
Sort each process by three questions. Is the volume high? Are the rules written down, or could they be? How much of the time is rework, and where does it start?
| Pattern | Typical examples | What to do |
|---|---|---|
| High volume, written rules, low rework | Bank matching, standard invoice coding, recurring journals | Candidates for automation or AI assistance now |
| High volume, written rules, high rework | Three-way match, cash application without remittances | Fix the upstream cause, then automate |
| High volume, rules unclear | Supplier queries, collections follow-up | Write the rules first; AI drafts, a person decides |
| Low volume, judgment-heavy | Unusual contracts, provisions, board commentary | Keep with experienced people |
The second row surprises people most. Payables rework often starts in purchasing, with purchase orders raised after the invoice arrives or goods receipts never recorded. Cash application rework often starts with customers who pay without a remittance. Automating the downstream step clears the exception queue faster without making it any shorter.
Gartner's November 2025 finance AI release put payables automation among the three most common uses, cited by 37% of the finance leaders it surveyed, while 91% said the impact was low or moderate at first. A time audit usually explains that gap. The volume sits in transactional processes, and the gain depends on how clean the process was before any tool arrived.
Decide where the hours go before you free them. Released time is quietly absorbed unless it has a destination. Name the work that will receive it, such as analysis the team never reaches or collections calls that are currently skipped, before the change goes live.
What a time audit will not tell you
It records what people did, not what they should have done. A team that produces a report nobody reads will log those hours faithfully, and someone still has to ask whether the report is needed. It is also self-reported, and the first week is usually tidier than the rest.
It will not show the work that is not happening. A maintained cash forecast, margin analysis by customer, a review of supplier terms: these are often the reason to free capacity at all, and they leave no trace in a log of the current week.
Frequently asked questions
How do you measure finance team capacity?
Log hours by process in half-hour blocks for four weeks, tag each block as first-pass, rework or interruption, and pull transaction volumes for the same period from the system. Hours per unit and rework share by process then show where capacity goes.
How long should a finance time audit run?
Four weeks, including one full month-end close. That captures the routine weeks and the close peak without wearing the team out.
Which finance processes are best suited to AI?
Processes with high volume, written rules and low rework come first, such as bank matching and standard invoice coding. Where rework is high, fix its upstream cause first, or automation only moves exceptions faster.
What does it typically cost to process a supplier invoice?
APQC's Open Standards Benchmarking puts the cross-industry median at $6.50 per invoice processed, across 6,546 companies. Your figure depends on how you allocate people and system costs, so track your own trend before comparing.
Where CFOLogic fits
We suggest CFOs run a time audit like this before deciding where AI should go. Our work is designing and building AI-assisted tools for finance on top of existing systems such as SAP, Microsoft Dynamics and QuickBooks Online, reading data rather than changing it, and with a named person signing off what goes out. How that sits within our CFO Success Partners work is set out on the service page.