Process automation: the guide

Someone who spends four hours a week copying orders from email into the invoicing system spends 16 hours a month. At €25 an hour, that is €400 a month, every month. Multiplied by the similar tasks spread across the company, this is the cost nobody accounts for.

Automation doesn't create a new expense: it swaps a recurring, growing expense for a one-off investment plus maintenance. The first step is putting a number on that cost.

The best candidates share four traits: high volume, clear rules, little human judgement and an error with a contained cost. Repeated communication, moving data between systems, generating documents, qualifying leads and answering frequent questions are the classic examples.

In sales, for instance, you can automate lead enrichment and qualification, meeting summaries, proposal drafts and follow-ups; negotiation and strategic accounts stay with people. In finance, reconciliation and invoice processing are common candidates.

Zapier suits non-technical teams and simple automations of one to three steps, but gets expensive as volume grows. Make offers better value for visual flows with conditional logic, with a steeper learning curve. n8n gives full control, can be self-hosted and integrates with any AI model through an API, but needs someone technical to set it up.

For SMEs orchestrating critical operations, our default choice is self-hosted n8n: the data stays in-house and the cost doesn't rise with volume.

A classic automation follows fixed rules. An AI agent receives a goal, decides the steps and uses the company's tools to achieve it: triaging emails and tickets, qualifying leads, reconciling invoices. When an agent takes over a role end to end, it is called a digital worker.

The difference in cost structure is significant: handling more volume no longer requires more people for mechanical tasks. Start with a person approving what the agent proposes, and only grant autonomy once consistency has been proven.

Trying to automate everything in the first project is an expensive mistake. What works: pick the process with the best return, automate it, measure against the initial numbers, stabilise and only then move on to the next. Each project funds the next, in savings and in confidence.

Also account for what usually gets left out: cleaning data, training the team and giving the change time. And set an owner and a metric from day one: most failures don't come from the technology, they come from automations nobody owns.