How to spot a task worth automating
Before the list, a rule of thumb so you can look at your own operation. A task is a good candidate for automation when it ticks most of these boxes:
- It repeats: every day, every week or every month, always the same way.
- It has clear rules: if you can explain it as "when this happens, do that", a computer can follow it too.
- It moves data from one place to another: from an email to a spreadsheet, from a spreadsheet to the system, from the system to a PDF.
- Mistakes are expensive: a mistyped number on an invoice or an order lost in the inbox.
- It depends on one person: when that person goes on vacation, the task stalls or gets done wrong.
The core idea: you don't automate a person's job; you automate the mechanical steps that keep them from doing the work that actually takes judgment.
With that filter, these are the ten tasks I run into most often in small and mid-sized companies. None of them needs a new system: they're usually solved by connecting the tools you already use.
Data that's copied by hand today
1. Entering orders that arrive by email. Orders come in by email, sometimes as an attachment, and someone types them into the system or a spreadsheet. An automation reads the email, extracts the data (customer, products, quantities) and creates the order. If something doesn't add up, say a product that doesn't exist, it flags it for a person to review. No more typing, and no more orders forgotten in the inbox.
2. Keeping spreadsheets in sync with the system. Prices in an Excel file, inventory in the system, customers in another shared sheet: every change has to be made in two or three places, and sooner or later they don't match. Automatic sync makes a change in one place show up everywhere else, with a log of what was updated and when. This is usually solved with an integration between systems.
3. Onboarding new customers. When a customer signs up, they have to be created in the management system, in invoicing, on the mailing list, and maybe sent a welcome email. With a single form, the data is entered once and the remaining steps run on their own. Less duplicated data and no forgotten steps.
Money: payments and invoicing
4. Payment reminders. Checking who owes what, writing the email, remembering to follow up a week later. An automation checks due dates and sends the reminder at the right time, with the amount and payment method, and alerts your team when someone has ignored several reminders. You get paid on time more often, and nobody has to play the awkward role of chasing payments every day.
5. Invoicing. Issuing invoices one by one on a tax authority's website is slow and error-prone. Once it's integrated with your system, the electronic invoice is generated from the order or sale and sent to the customer. In the management system I built for a plant nursery, for example, an overnight process batch-invoices everything pending from the day, without anyone having to sit down and do it.
Reports and alerts
6. Monthly reports. Every month someone exports data from two or three places, pastes it into a spreadsheet, builds charts and emails them out. It's the classic example: the report can be generated and sent on its own, on the day and at the time you choose, always in the same format. And it no longer depends on the one person who knows how to build it being available.
7. Low-stock alerts. You find out a product is out when a customer asks for it. With an automatic alert, when an item's stock drops below a minimum you set, an email or message goes to whoever handles purchasing. The next step, if it makes sense, is for the alert to draft a purchase order for the supplier.
Email, documents and files
8. Sorting emails and inquiries. A shared inbox where everything lands: sales inquiries, complaints, supplier invoices, résumés. An automation can classify each message and route it to the right person, or tag it by urgency. With artificial intelligence this works even when emails are written any which way, not just based on keywords.
9. Generating documents. Quotes, contracts, delivery notes and certificates built by copying a template and changing names, dates and amounts. From the data that's already in the system, the document is generated as a correctly formatted PDF and, if you want, sent straight to the customer. No more quotes with the previous customer's name on them.
10. Backing up files. Backups that depend on someone remembering to run them will eventually not happen. An automatic backup copies your important files and databases somewhere else every day, keeps several versions and alerts you if something failed. And it's worth testing a restore every now and then: a backup that's never been tested is a hope, not a backup.
Where to start
You don't need to automate all ten. To choose the first one, I look at three things:
- How much time it takes today. Add up the hours spent on it each month, counting everyone involved.
- How much a mistake costs. A wrongly issued invoice or a lost order weighs more than a report that goes out a day late.
- How simple it is. A task with clear rules and few exceptions is the best place to start: it ships fast and builds confidence for the next one.
Whichever wins on all three goes first. It's often not the flashiest one but the most boring one, and that's exactly the one that brings the most relief when it disappears.
What not to automate (yet)
- Processes that change every week. Get them in order first; automating a messy process just makes the mess faster.
- Decisions that take judgment. Approving a special discount or handling a sensitive complaint. What you can automate is gathering the information so that decision gets made quickly.
- Tasks done twice a year. The savings rarely justify the work, unless a mistake there is very expensive.
How I approach it
In a process automation project I start by understanding how the task is done today, step by step, with the person who does it. From that we estimate how much time it takes and what automating it costs, so you can decide with numbers. Then I build it by connecting the tools you already use, with alerts for when something doesn't go as expected, and we run it alongside the manual process until you're confident it works.