Automation · 6 min read
What Should Actually Be Automated?
The conversation about automation usually starts in the wrong place — with a technology, and specifically with AI. The better starting point is a working week: where does skilled time go into work that follows rules a machine could apply?
Automation is not the goal
Nobody's business improved because it “had automation.” It improved because quotes went out the same day, because invoices matched orders without checking, because customers were told their delivery had shipped without anyone remembering to tell them. Automation is a means; the goal is a process that is faster, more consistent, and less dependent on memory.
That framing matters because it changes what you automate. You stop looking for impressive use cases and start looking for repetitive ones.
Three tests for a good candidate
A task is a strong automation candidate when it passes three tests. It is repetitive: the same shape of work happens many times a week. It is rule-based: a person could write down how the decision is made, even if the rules are long. And it is currently done by hand: someone is copying, checking, chasing, or re-entering.
Quotation follow-ups, invoice reminders, order confirmations, data synchronisation between systems, weekly report assembly — these pass all three tests in most businesses.
Where AI fits — and where it doesn't
Traditional automation moves structured data according to rules: when an order is approved, create the invoice. AI adds the ability to handle unstructured input: reading a customer's rambling email and extracting a structured order, summarising a long thread, classifying an incoming request.
The dividing line is judgment. If the step requires accountability, relationships, or decisions under ambiguity, a person should make it — with the system preparing everything they need. If the step is extraction, routing, or confirmation, software can usually take it.
Keep a human in the loop
The reliable pattern is automation with a checkpoint. The system drafts, the person approves. The system flags the unusual invoice instead of silently paying it; it prepares the quote and a salesperson sends it. This is not a half-measure — it is how you get the speed of automation without its worst failure mode, which is confident mistakes at scale.
Starting small
The best first automation is boring and frequent: something that happens daily, follows clear rules, and annoys someone every time it is done manually. Prove value there, measure the hours returned, and let the results fund the next step. Automation programs built one honest win at a time tend to survive; ones launched as grand transformations tend to become case studies in someone else's slide deck.
Facing this decision in your own business? A short conversation is usually enough to clarify the options.
Talk to HESEDNext step
Have a business process that should be easier?
Tell us what you're trying to improve. We'll help determine whether the right solution is a website, a custom application, automation, or something simpler.