A wider view of automation.
Automation is often treated as a task-level problem: take a repetitive job and make a computer do it. That's the narrowest version of the work. We take a wider view — reviewing whole processes and redesigning them with the right combination of human steps, software automation, and AI, informed by what the data shows.
Start with the process, not the tool.
Every business process — from sales handover to invoice approval to customer onboarding — is a sequence of steps, decisions, and handoffs. Many were designed under old constraints and never revisited. Some exist only because someone once built a workaround and no one removed it.
Our job is to review processes end-to-end and redesign them with the tools now available. Sometimes that means automation. Sometimes AI. Often the biggest win is cutting a step entirely, or putting a person in a place where they weren't before.
The point is to use the right tool for each step — and combine them so the process is faster, clearer, and less prone to things falling through the cracks. With good data behind it, the right change usually becomes obvious.
Three tools, one intersection.
Every process of any complexity is built from some combination of three things: people using their judgement, software doing predictable work on rails, and AI handling language and pattern. The interesting part isn't any one of them — it's where they overlap.
Use the right tool — per step.
The question isn't "which tool is best?" — it's which tool is best here. Plot any task against how much judgement it needs and how often it happens, and the answer becomes obvious. Treating it any other way is how you end up with AI where a rule would do, or a person doing what a script should.
Four tools. One process.
Process review.
Before automating anything, understand what a process is for, where it breaks, which parts earn their cost, and which parts exist only out of habit. The review itself often uncovers the biggest improvements — before any technology is introduced.
Automation.
Software doing the mechanical work: moving data between systems, triggering actions, applying rules, generating documents, chasing the things that need chasing. Predictable, auditable, and boring — which is exactly what makes it reliable.
AI.
Used where judgement, language, or pattern recognition is involved — summarising, classifying, drafting, extracting from unstructured input, answering questions over your own data. Not used where a simple rule would do the job more cheaply and more reliably.
Human in the loop.
Kept deliberately. Most processes are better with human judgement at specific moments — approvals, exceptions, nuance, relationships. The question is which moments, and why, so the people involved focus on the parts that genuinely need them.
A process, using all three.
One complete process. Eight stages. Each step uses whichever tool is right for it, then hands back to the next. The lanes below show exactly where each tool takes over and where it lets go.
Use the right tool for the step. Combine them so fewer things fall through the cracks.
The same outcome — fewer steps.
Most processes don't need adding to. They need taking apart and putting back together with everything unnecessary removed. Automation and AI aren't the point — the outcome is. And the outcome is usually a shorter, clearer process with fewer things to watch and fewer places to fall down.
A practical sequence.
Map the process as it really runs.
Not how the org chart says it runs. The real version, including the spreadsheets, workarounds, and informal handoffs.
Simplify before you automate.
Remove steps that don't earn their place. Automating a bad process just means you get a bad outcome faster.
Pick the right tool per step.
Rule-based automation, AI, or a person — whichever is most reliable, accountable, and cost-appropriate for that step.
Build for change.
Processes shift. We design so tomorrow's change to the business doesn't mean starting the whole thing over.