Published by CXJ Labs, an automation company. Not independent. Read our disclosure.
The Automation Index

Can purchase orders be automated?

Largely automatable: 69% of task time

The answer

Largely. Raising purchase orders, sending them, routing approvals and matching deliveries and invoices to them are highly automatable. In our model about 70% of PO time can go to software. What stays human is approving larger spends, chasing late suppliers by phone and sorting out deliveries that do not match the order.

The tasks, and how much of each is automatable today

Six tasks make up purchase orders. Raising the PO is about 85% automatable. Sending it and matching delivery notes and invoices are about 75% automatable. Collecting what is needed from job sheets is about 60% automatable. Approvals are half automatable. Chasing late deliveries stays mostly human, because suppliers respond to people.

TaskActivity typeMin / weekAutomatable
Collect what is needed from job sheets and engineersA structured request form against a job removes most phone calls. Vague requests like 'the usual' still need a conversation.Collecting data
base rate 64%		16060%
Raise the purchase orderSupplier, items and job code come from the request. New items without a catalogue code need a person.Processing data
base rate 69%		16085%
Approve orders above the thresholdRouting and one-tap approval are automatic. The decision about whether to spend stays with a manager.Applying expertise
base rate 18%		3050%
Send the PO and confirm with the supplierEmail sending and confirmation tracking are automatic. Some suppliers still want a phone call.Interacting with people
base rate 20%		8075%
Match delivery notes and invoices to the POAI reads invoices and delivery notes well. Part substitutions and split deliveries still need a person to decide.Processing data
base rate 69%		12075%
Chase late deliveriesSoftware spots late orders and sends a first chase. Getting a supplier to prioritise you is a phone call.Interacting with people
base rate 20%		6040%
Whole process61069%

Base rates are McKinsey Global Institute's technical automation potential for each activity type. The share for each task is our adjusted estimate. Method.

Hours saved per week

For an electrical contractor raising about 40 purchase orders a week, PO work takes around 10 hours a week by hand. Automating the tasks above returns about 7 hours a week. The time left is approvals, supplier calls and the mismatches that genuinely need someone to decide.

Modelled on: An electrical contracting business raising about 40 purchase orders a week to a handful of regular suppliers, handled by an office administrator.

  • 40 POs a week, each needing about 4 minutes to gather details from job sheets or engineers.
  • About 4 minutes per PO to raise it in the accounting system and 2 minutes to send and confirm.
  • About 30 minutes a week on approvals for orders above a spending threshold.
  • About 2 hours a week matching delivery notes and invoices, and an hour chasing late deliveries.

10 hours a week by hand, 7back with automation. Formula: the sum of each task's minutes multiplied by its automatable share, divided by 60, rounded to the nearest half hour.

Tools that do the job

You need a structured way for staff to request materials, an accounting system that holds POs, and something that matches documents. A Typeform or Airtable request form feeds n8n or Make, which raises the PO in Xero or QuickBooks. An AI agent reads supplier invoices and delivery notes for matching.

Airtable
Materials request form tied to job numbers, plus a supplier and item catalogue.
Xero
Holds purchase orders and receives supplier bills for matching.
QuickBooks
Alternative accounting system with purchase orders and bill matching.
n8n
Raises POs from requests, routes approvals by amount and runs the matching step.
AI agent
Reads supplier invoices and delivery notes and compares them line by line with the PO.
Slack
One-tap approval messages for managers on site.

A simple workflow

The workflow starts when an engineer submits a materials request against a job. Orders under a threshold become POs and go to the supplier straight away. Orders above it go to a manager to approve. When the invoice arrives, an AI step reads it and matches it to the PO and delivery note.

  1. TriggerEngineer submits a materials request against a job number.
  2. DecideUnder the threshold: raise the PO. Over it: send to a manager to approve in Slack.
  3. ActPO is created in the accounting system and emailed to the supplier with the delivery address and date.
  4. MatchInvoice arrives; AI matches it to the PO and delivery note. Matches are approved for payment.
  5. EscalateMismatches and deliveries more than a day late go to the administrator with the documents attached.

Keep these human

  • Approving larger spends. The threshold decision belongs to whoever owns the budget.
  • Resolving mismatches. Deciding whether to accept a substitute part or dispute an invoice line needs judgement.

Where it goes wrong

  • No item catalogue. Free-text item names defeat matching. Build a short catalogue of the 50 items you order most.
  • Approval bottlenecks. If one manager approves everything, POs wait. Set thresholds so routine orders never need approval.
  • Auto-approving invoices that match the PO but not the delivery. Always match all three documents before payment.

What most guides miss

Most PO guides focus on raising orders faster. Raising is the quick part. The time and money leak at matching: invoices paid for items that never arrived or arrived short. Automate the three-way match between PO, delivery note and invoice, and send only the mismatches to a person. That is where the savings are.

The evidence