There are two kinds of automation. Choosing the right one for each part of a job makes a real difference to cost, reliability and how well it holds up over time.

AI automation uses an AI model each time it runs. It can read an invoice, understand an email, and work out what a messy document is about. It is flexible, copes with inputs nobody planned for, and can be set up quickly.

Coded automation uses fixed rules written in code. It runs exactly the same way every time, costs very little to run, and can stay entirely on your own systems. It takes more work to build.

Three questions that decide it

Does the input vary? Handwriting, free-text emails, PDFs from forty different suppliers: this is where AI shines. It is also why many older automation projects struggled, because nobody could write a rule for every case.

How often does it run? A task that runs a few times a month can comfortably go through an AI model each time. A task that runs thousands of times a day is usually better in code, where each run costs next to nothing.

Must the answer be identical every time? Payroll and statutory submissions must be. A first-draft reply needn’t be. Where an auditor or a customer expects the same answer twice, code is the right tool.

Where the data question fits

Owners often ask about data and AI, and this choice is part of the answer. With AI automation, the data is sent to an AI service to be processed, so decide in advance what may be sent and what stays out. With coded automation, the data can stay on your own systems from start to finish, which matters for the most sensitive records.

The pattern that works

Most real workflows use both. AI handles the edges, where the variety is: reading the document, sorting the request, pulling out the fields. Code handles the middle, where the rules are: the calculation, the checks, saving to your main system, the audit trail.

Get that split right and the workflow is economical to run, stands up to an audit, and keeps working when a supplier changes their invoice layout.

Before you approve a project

Ask which kind each part will be. A clear answer is one of the best signs of a well-planned project.

In your department

  • Accounts & Finance: Reading supplier invoices in forty layouts suits AI. The calculation, the tax and the posting to your accounting system belong in fixed rules an auditor can follow.
  • Operations & Delivery: Let AI read incoming requests and pull out the details. The schedule updates that run hundreds of times a day are cheaper and steadier in code.
  • HR & Admin: Payroll and statutory submissions must give the same answer every time, so they run on rules. AI helps with the varied part, such as claims and leave requests written in many different ways.