Automation

Automation is the use of machines or software to carry out defined work without a person in the loop. The work has to be definable before it can be automated — which is why process clarity, not technology, is usually the binding constraint.

The word covers a very wide range. At one end sits a rule that moves a file; at the other, a stack that reads a document, decides what to do and does it.

Scope is the only useful way to tell the levels apart. Each step up adds capability and adds something to maintain.

Adoption keeps climbing. The European Central Bank’s Q4 2025 survey found 38% of euro area firms at an advanced stage of AI adoption, with a further 33% experimenting.

Key takeaways

• Automation removes the person from defined work, not from judgement work.
• The levels differ by scope: one task, one process, or a whole stack.
• An automated process still needs an owner, a monitor and a failure plan.
• Most automation failures trace to an unclear process, not to the tooling.

How it works

Automation starts by pinning a process down: the trigger, the steps, the decision points and the exceptions. Anything that cannot be written down that precisely is not yet automatable, however good the software.

Rule-based tools take it from there. Robotic process automation drives existing applications the way a person would, clicking and typing through screens, which is why it works on systems with no usable interface for machines.

The levels stack by scope.

Exceptions decide whether any of it works. Every automated process needs a defined route for the cases it cannot complete, and a named person at the end of that route.

A process with a 5% exception rate is usually fine. One at 30% is a process that was not ready — and automating it just moves the mess downstream faster.

The business case is usually framed as cost removed, which overstates it. Bots break when the underlying screen changes, and someone has to notice, diagnose and fix them.

That maintenance load is what the automation savings ratio exists to capture: net cost avoided divided by the total cost of building and running the automation.

Evidence on motive is more mixed than the sales pitch suggests. The US Census Bureau’s 2023 Annual Business Survey found 50.8% of businesses used robotics to automate work previously done by people, with quality and reliability cited almost as often.

Examples

Automation lands differently depending on whether the work is a keystroke, a document or a decision. These four cases show the levels doing what each is actually good at.

A shared services centre in Manila posts supplier invoices into an ERP system that has no modern interface. A rule-based bot keys them overnight, and the finance team reviews only the exceptions it could not match.

An insurer routes incoming claim correspondence by reading it. Letters arrive in no fixed format, so rules alone fail and the work needs a model that can interpret language rather than match a template.

A telecoms operator chains the two together for service activation: a model reads the order, rules provision the line, and a workflow engine escalates anything that stalls. That chaining — rather than any single tool — is what the term hyperautomation describes.

A recruitment firm automates candidate screening against fixed criteria, then stops. Everything past the shortlist is judgement, and the firm treats the cut-off as a design decision rather than a limitation it hopes to remove later.

That last case is the useful one. Knowing where to stop is a more reliable predictor of a successful programme than how much of the stack a buyer eventually adopts.

Buyers who want the capability run rather than built usually reach automation outsourcing, which contracts discovery, development and ongoing bot support to a provider.

Related terms

The automation family is a scope ladder, and most confusion comes from treating its rungs as synonyms. Each entry below owns one rung or one way of measuring the result.

• Robotic Process Automation: rule-based software driving existing applications screen by screen.
• Cognitive Automation: software that reads unstructured input and makes a judgement call on it.
• Intelligent Automation: rules, models and process management fused into one platform.
• Hyperautomation: the business-led chaining of a whole automation stack at once.
• Automation Outsourcing: contracting discovery, build and ongoing bot support to an external provider.
• Automation Savings Ratio: net cost avoided divided by the cost of building and running the automation.

FAQ

What is the difference between automation and artificial intelligence?

Automation is about executing defined work without a person. Artificial intelligence is about handling work that was not defined in advance, and the two are increasingly combined rather than chosen between.

Which processes should be automated first?

High-volume, rule-stable, low-exception ones. A process that changes every quarter will cost more to maintain as an automation than it saves.

Does automation remove the need for staff?

It removes defined tasks, not oversight. Someone still owns the process, watches for failures and handles the exceptions the automation refuses.

Why do automation programmes stall?

Usually because the process was never documented properly. The tooling exposes the gap rather than causing it.

How is an automation’s return actually measured?

Net cost avoided against the full cost of building and maintaining it, over a defined period. Counting only the hours removed, and ignoring the engineers who keep the thing alive, is how programmes report savings nobody can find in the budget.

Can automation be outsourced?

Yes, and discovery, build and ongoing support are commonly contracted as one package.

Outsource Accelerator covers the operations side if you are scoping a first automation.