Process
Automating a bad process just makes it fail faster
Most automation projects start with a tool and a task: “let’s automate invoice entry”. The task gets faster. The process around it (the approvals, the exceptions, the chasing) stays exactly as slow as before.
The better starting question is what the process is for. Which steps exist for control, which for habit, and which only because two systems never talked to each other? Once that is clear, a third of the steps often do not need automating at all. They need removing.
Our rule of thumb: simplify first, standardise second, automate third. Technology applied in that order compounds. Applied in reverse, it preserves the complexity you were trying to escape, only now it is harder to see and more expensive to change.
AI
Where AI agents earn their place in operations
AI agents are at their best where work is high in volume, heavy in language or documents, and governed by judgement that can be described. Reading an invoice, classifying a request, drafting a reply, checking a contract against a policy.
They are at their weakest where the stakes are high and the rules are unwritten. That is not a reason to avoid them; it is a reason to design the process so that people review the exceptions and the agent handles the rest.
Three questions help decide. Can we say what a good outcome looks like? Can we check it cheaply? And what happens when the answer is wrong? If those have clear answers, an agent probably belongs in the process. If not, start with decision support and keep a person in the loop.
Automation
n8n, Zapier or custom code? Choose by the process
The question we hear most is which platform is best. The honest answer is that it depends on the process, and on who will own it once we leave.
Simple, linear flows between well-known applications are quick to build and easy for a business team to maintain in a tool like Zapier. Flows with branching logic, custom data handling or self-hosting requirements tend to suit n8n. Processes that are core to the business, or that need an interface of their own, can justify custom development.
What matters more than the platform is that the workflow is documented, monitored and owned. A modest tool with a clear owner outperforms a powerful one nobody dares to touch.
