Measure Your Invoice Bottleneck Before AP Automation

Supplier invoices arrive in several inboxes. Someone downloads each attachment, types the same fields into accounts, looks for a purchase order and chases an approver. The queue feels slow, but nobody can say whether capture, entry, matching, approval or exceptions cause most of the delay.

Do not choose an AP tool from that feeling alone. Start with the invoice processing bottleneck diagnosis and measure a representative set of invoices from arrival to payment preparation.

The aim is not to create a perfect time-and-motion study. It is to collect enough evidence to separate stable work that can be automated from judgement, exceptions and payment controls that need people.

Define the start and finish

Teams often quote “invoice processing time” while measuring different things.

Use a clear boundary:

• Start: the invoice first reaches an inbox, portal or intake channel controlled by the business.
• Finish: the invoice is approved and posted, ready for inclusion in an authorised payment run.

Payment settlement is a separate measure. Mixing bank clearance time into internal invoice handling hides the bottleneck.

Record both:

• Elapsed time: the total time between start and finish.
• Touch time: the time people actively spend handling the invoice.

An invoice can need four minutes of entry but wait four days for an approval. Automating entry alone will not remove the larger delay.

Choose a representative sample

Take a real operating period, not only the easiest invoices. A working week can provide a first map, but include month-end or seasonal work if it behaves differently.

Segment the sample by factors that change the route:

• Purchase-order and non-purchase-order invoices.
• Existing and new suppliers.
• Clean documents and poor scans.
• Single and multiple VAT treatments.
• Routine coding and project or cost-centre coding.
• Domestic, overseas and foreign-currency suppliers.
• Ordinary invoices, credit notes and duplicates.

State the period and sample size. Do not call it an average for the whole business if it misses busy periods or important exceptions.

Map every hand-off

Follow each sampled invoice through the real process. Include unofficial steps, such as saving a PDF to a folder or messaging an approver outside the accounts system.

A typical map contains:

1. Arrival in email, portal, post or upload.
2. Download or central capture.
3. Supplier and invoice identification.
4. Field entry or extraction.
5. Duplicate check.
6. Purchase-order or receipt matching.
7. Nominal, project and VAT coding.
8. Approval routing and chasing.
9. Exception resolution.
10. Posting to the accounts system.
11. Preparation for the payment run.

For every hand-off, record the system, owner, active time, waiting time and reason it moved to another person.

Capture the baseline fields

Use a spreadsheet if that is the quickest controlled way to begin. One row per invoice is enough.

Field	Why it matters
Arrival and ready timestamps	Separates total elapsed time from opinion
Active handling minutes	Shows where staff effort sits
Intake channel	Reveals fragmented capture
Supplier and document type	Groups predictable and unusual work
PO or no PO	Shows the effect of matching discipline
Number of approval touches	Exposes unclear routing and chasing
Exception category	Identifies work automation must contain
Re-entry count	Shows where systems do not connect
Correction or rework	Exposes extraction, coding and mapping quality
Duplicate flagged	Confirms a key financial control
Final status and owner	Prevents lost items from disappearing from the sample

Do not put bank details or unnecessary personal data in the measurement sheet.

Separate waiting from work

Create a simple timeline for several ordinary and exceptional invoices. Label each period as:

• Active handling.
• Waiting for a person.
• Waiting for data or a purchase order.
• Waiting for a system.
• Exception investigation.

This distinction changes the prescription.

• High entry time may justify extraction and direct accounting integration.
• Long approval queues may need ownership rules and reminders.
• Repeated missing POs may need a purchasing-process change.
• High exception time may need better supplier or master data before AI.
• Delays between tools may need an API or workflow repair.

Buying document extraction when approval ownership is the main bottleneck produces a faster arrival into the same queue.

Build an exception register

The happy path is easy to demonstrate. The exception queue decides whether an automation is safe and useful.

Group every exception into a cause:

• Supplier not recognised.
• Invoice number missing or duplicated.
• Purchase order absent or does not match.
• Goods not received.
• Nominal, project or cost centre unclear.
• VAT treatment needs review.
• Bank details changed.
• Approval owner absent or disputed.
• Document unreadable.
• Accounts system rejected the record.

For each category, record frequency in the sample, the person qualified to decide it and the evidence they need. The future workflow should route the exception with that context, not silently guess.

Protect the controls while measuring

The current manual process may be slow because it contains necessary controls. Do not remove them by accident.

Record who can:

• Create or amend a supplier.
• Change supplier bank details.
• Code and post an invoice.
• Approve the commercial spend.
• Prepare a payment run.
• Release payment.

The UK National Cyber Security Centre recommends verifying important payment instructions through another method, using known contact details rather than relying only on the message that requested the change. That boundary belongs in the target workflow as well as the baseline.

Automation can collect, extract, propose, match, route and prepare. Supplier changes, material exceptions, approval and payment release need controls appropriate to the business.

Turn the evidence into a pilot hypothesis

Choose one bounded part of the process and state what should change.

For example:

For existing UK suppliers with readable PDF invoices and a valid purchase order, the pilot will capture the document, propose the required fields, match the PO and route exceptions. People will approve the invoice and release payment.

Then select measures from the baseline:

• Median elapsed time for the selected segment.
• Median active handling time.
• First-pass extraction and matching rate.
• Exception rate by cause.
• Rework and duplicate count.
• Approval waiting time.
• Tool and implementation cost.

The pilot is successful only if the measured result improves without weakening the agreed controls. Do not substitute a vendor's generic savings percentage for this comparison.

Use UK evidence as context, not your result

HMRC-commissioned research confirms that UK SMEs use a mixture of electronic and document-led invoicing processes, and that adoption concerns include cost, integration and implementation. That supports measuring the existing process before choosing a route. It does not provide an APIwise customer result or predict your saving.

The appropriate answer may be a feature already present in Xero, QuickBooks or Sage, an established capture product, a workflow configuration, or a thin integration between existing tools. A custom build needs a measured reason.

When APIwise fits

APIwise fits when invoices cross several systems or inboxes, the exception path is unclear, or standard tools cannot cover the required matching and routing. Our AI automation work starts with the workflow and controls, then scopes the smallest useful pilot.

Bring the process map, baseline sheet and three representative exceptions to an AI Readiness Assessment or Integration Health Check. The output can compare off-the-shelf, configured and custom routes against the work you actually measured.

If invoice entry and approval chasing are holding up the team, use the invoice bottleneck checklist or tell APIwise where the queue builds up.

Sources used

• HMRC: electronic invoicing, SME usage and attitudes
• NCSC: business email compromise guidance