Home › Blog › Complete process automation guide

The complete process automation guide — assess, design, build

Choose the right process and building the automation is easy. Choose wrong and no tool keeps it alive. Which is why half of this guide is about choosing.

Selection criteria Design principles Automation that survives

The short answer

Automation survives when three things hold: the rules are stable, the input format is consistent, and a person notices when it fails. Miss any one and it is abandoned within months.

So automating "the most painful process" first is usually the wrong order. The most painful process is usually painful precisely because its rules are unstable. Start where the rules are stable.

Stage 1 — Selection: what makes work automatable

Condition	Good fit	Poor fit
Rule stability	The procedure has been the same for months	Every case needs an exception judgement
Input format	Arrives on the same form	Arrives as free text in an email
Frequency	Weekly or more often	Once a quarter
Verifiability	You notice a wrong result immediately	It only surfaces months later
Reversibility	Mistakes can be undone	Cannot be undone (sending, payment)
Ownership	A name is attached	Spread across several teams

Stage 2 — Design: four principles

Never touch the original — Automation output always goes somewhere new. Overwrite the source and it cannot be undone — and what cannot be undone cannot be verified.

Never fail silently — A failure has to raise an alert. Silently failing automation is the most dangerous kind: you discover weeks of unprocessed work afterwards.

Put a person in front of irreversible steps — An approval gate before external sending, payment and deletion. This is accountability design, not a technical preference.

Add one step at a time — When extending working automation, add one step. Add two and finding which broke costs more than the extension saved.

Stage 3 — Build: what to build it with

Features already in your tools — Check here first. More is possible without a new tool than people expect, and the maintenance burden is lowest.

A workflow automation tool — Right when the job spans several tools. The flow is visible on screen, so it can be handed over when the owner changes.

Scripts — Flexible but hard to hand over. It easily becomes something that stops when its author leaves, so fix the documentation and where it lives.

Where AI attaches — Only to judgements rules cannot express: classification, unstructured extraction, summarisation. Arithmetic and branching are exact and reproducible as rules.

Stage 4 — Operation: why automation dies

The most common cause of abandonment is a connection breaking quietly — an expired API key, a permission change, a changed form. Which is why failure alerts belong in the design.

The second is the owner moving on. Automation only its author understands becomes untouchable when they leave. Make sure at least two people hold the flow documentation and the access.

The third is the process changing. When the business rule moves and the automation does not, it produces wrong results automatically. That is the most dangerous failure, and only a periodic sample check catches it.

The rest of this cluster

Getting started — Where to start with automating repetitive work

Concrete patterns — Seven patterns for replacing Excel work · How meeting-note time was reduced · Weekly report automation

Internal documents — A practical guide to internal document search AI (RAG)

Tool comparison — Workflow automation tools compared — n8n, Make, Zapier, custom

Agents — What is an AI agent, and how does it differ from a chatbot?

Cost — The costs that are not on the AI tool price page

Frequently asked questions

Should we not automate the most painful process first?

Usually not. If it is painful because it needs constant exception judgement, it does not meet the criteria. Build a success on stable rules first, and use that credibility to approach the hard one.

Is managing several automation tools difficult?

It is, so consolidating is better. But leave alone what already works inside a tool you pay for — migrating is itself a cost.

How do we report the benefit?

Against the numbers from before you started: monthly volume, minutes per item, and review time. Including review time is what makes the report credible.

What do we do with automation that failed?

Turn it off explicitly rather than leaving it. Half-working automation quietly producing wrong results is the most expensive state there is. Record why you turned it off — that is material for the next attempt.

Find the target

Start with which process fits

The automation assessment shows how many steps your repetitive work takes and the time it could save. Free, no sign-up.

Run the assessment