Most businesses that try automation start with the most exciting task rather than the right one — usually an AI chatbot — and end up with something impressive that nobody trusts. The better first automation is almost always boring: a small, frequent, repetitive job that currently eats someone’s time every single day.
Here is how to find it.
The four-question test
A task is a good candidate for automation when the answer to all four of these is yes:
- Does it happen often? Daily or weekly, not twice a year. Automating a rare task costs more than doing it by hand forever.
- Does it follow rules? If you could write the steps on a sticky note — “when this arrives, put it here, then tell this person” — software can follow them.
- Does it happen on a computer? Forms, emails, spreadsheets, WhatsApp, your store, your accounting tool. If the data is already digital, it can be moved.
- Is a mistake recoverable? Start where an error is an inconvenience, not a disaster. Leave money movements and anything legal until you trust the system.
A useful exercise: for one week, have everyone jot down each repetitive task they do and roughly how long it took. By Friday the best candidate is usually obvious, and it is rarely the one anyone guessed.
Good first automations
These come up again and again, in almost every kind of business:
- Enquiries that file themselves. A website form or WhatsApp enquiry lands in a spreadsheet or CRM, the customer gets an instant acknowledgement, and the right person is alerted. Nothing sits in an inbox overnight.
- Payment and renewal reminders. Polite, on schedule, and stopped automatically once the payment arrives.
- The weekly report. Numbers pulled from your store, ad accounts and sheets into one summary that arrives on Monday morning without anyone building it.
- Order and booking updates. Confirmations, reminders and status changes sent the moment something happens.
- Review requests. A message a few days after delivery or a visit, asking a happy customer to leave a review.
None of these needs AI. That is a feature: plain rules are cheaper to run, easier to understand and far more predictable.
Where AI genuinely helps
AI earns its place when the input is messy — when a person would have to read something to decide what to do with it:
- Sorting free-text messages. Is this enquiry about pricing, a complaint, or a job application? Tagging it and routing it saves someone reading every one.
- Reading documents. Pulling the supplier, date and amount out of an invoice, or the key details out of a form someone filled in badly.
- Drafting replies for a person to approve. The assistant writes the first version; a human checks it and presses send.
- Answering the same questions all day. An assistant that answers from your own products, prices and policies, and hands anything unusual to a person.
Where AI should stop
Be wary of anything where being slightly wrong is expensive:
- Amounts and calculations. Totals, taxes and discounts should be computed by plain code, never generated by a language model.
- Decisions with consequences. Refunds, cancellations, anything a customer could hold you to. Let AI suggest; let a person decide.
- Talking to customers unsupervised, on day one. Start with the assistant drafting and a person approving. Loosen the reins once you have read a few hundred of its answers.
No-code tool or custom code?
Tools like n8n, Zapier and Make are excellent for connecting common services quickly, and a self-hosted n8n can run a lot of workflows very cheaply. They are the right place to start for most businesses.
Custom code makes sense when a workflow becomes business-critical, handles a lot of volume, needs logic that is awkward to express in boxes and arrows, or has to live inside software you already run. Plenty of setups use both: n8n for the glue, code for the heavy lifting.
Build it to fail loudly
The worst automation is not one that breaks. It is one that breaks quietly and nobody notices for a month. Whatever you build, make sure it has:
- A log of what it did and when.
- An alert that reaches a person when a step fails.
- An owner — one named person who gets that alert.
- An off switch that anyone on the team can find.
Outside services change their rules and APIs without asking you. An automation with these four things survives that; one without them just stops.
A first project, in practice
A typical first build looks like this: website and WhatsApp enquiries land in one sheet with the date, source and what the person asked about; the enquirer gets an instant, friendly reply; the team gets a message with the details; and anything not answered within a few hours gets flagged. It takes days rather than months to build.
We built a version of this for a Chandigarh marketing agency, whose enquiry form now files every lead straight into their own spreadsheet. If you describe the repetitive task that eats your week, we will tell you honestly whether it is worth automating — and what we would leave to a person.