How to automate repetitive tasks in a small business
Log a week of tasks, score each by frequency, minutes and error risk, automate the safest first, keep a person approving and measure the hours you save.
6 min read
Start business automation with a repeated task whose result you can check. Measure the time it takes, remove unnecessary steps, and test a small batch before allowing actions on customer messages or stock. This guide shows how to compare potential savings with the work and risk that remain.
Step 1: a one-week time log
Guessing what takes time rarely works; the tasks people complain about are not always the ones that eat the most hours. For one normal week, everyone who does routine work keeps a simple list with five columns:
- Task, in a few words (“copy web order to stock sheet”).
- How many times it happened that day.
- Minutes each time, roughly.
- Where the data comes from and where it goes (email, web shop, spreadsheet, accounting tool).
- What goes wrong when it is done badly.
Avoid an unusual week such as a holiday peak or a stocktake. At the end, add up each task and convert it to a monthly figure. Many owners find that a few small tasks done dozens of times add up to more than the big jobs they had in mind.
Step 2: score the candidates
Give each task a score: times per month × minutes each time × error risk. Error risk is a simple scale: 1 if a mistake is harmless and easy to spot, 2 if it causes extra work, 3 if it reaches a customer or the stock figures.
| Task | Times per month | Minutes each | Error risk (1–3) | Score |
|---|---|---|---|---|
| Copy web orders into the stock sheet | 120 | 3 | 3 | 1,080 |
| Write order confirmation emails by hand | 120 | 2 | 3 | 720 |
| Build the weekly sales report | 4 | 45 | 2 | 360 |
| Send payment reminders | 15 | 5 | 3 | 225 |
The score highlights work worth investigating; it is not approval to automate it. Review potential time savings separately from consequences of an error. Start with reversible internal work, and keep customer messages and stock changes subject to review.
Worked example: hours saved per month
Illustrative figures for a small online shop. The four tasks above take 360 + 240 + 180 + 75 = 855 minutes a month, about 14.25 hours.
Automation never removes the work entirely; someone still has to check the result. Suppose the shop automates the top three:
- Stock update from orders: a person spends 30 seconds per order checking it, so 120 × 0.5 = 60 minutes instead of 360. Saving: 300 minutes.
- Order confirmations: drafts are generated and a person reviews one daily batch in 5 minutes over 22 working days, 110 minutes instead of 240. Saving: 130 minutes.
- Weekly report: produced automatically and checked in 10 minutes, 40 minutes a month instead of 180. Saving: 140 minutes.
Total saving: 300 + 130 + 140 = 570 minutes, or 9.5 hours a month. Check: 855 minutes before, 60 + 110 + 40 + 75 = 285 minutes after, and 855 − 285 = 570. Payment reminders stay manual in this pilot because an incorrect message reaches a customer; their priority score of 225 is not permission to send automatically. Set any saving against the cost of the tool and the time to set it up and maintain it.
What to automate first
These tasks usually score high and carry little risk when a person checks the output:
- Data typed into two systems. The same order, product or customer entered twice is the classic candidate. Remove the second entry rather than speed it up.
- Stock updates after a sale. Each sale should lower the stock count without anyone retyping it. This matters most if you sell in several channels; see how to sell on several marketplaces.
- Order confirmations and shipping notices. Standard messages built from order data, sent from a template.
- Recurring reports. Weekly sales, slow-moving stock, purchases due. Same data, same layout, every time.
- Reminders. Reorder points reached, invoices due, follow-ups you promised a customer.
Our guide to stock control in a small business explains how to set reorder points worth automating.
What not to automate
Some steps should keep a person in the loop, at least until you have months of evidence that the automated version is reliable:
- Anything that moves money: payments to suppliers, refunds, price changes, discounts.
- Messages that decide something for a customer: refusing a return, answering a complaint, changing a delivery promise.
- One-off or rare tasks. If it happens twice a year, writing a checklist is cheaper than building an automation.
- Processes nobody understands yet. Automating a messy process makes the mess faster.
Before changing prices or discounts in bulk, check the effect with the margin calculator.
Keep a human approval step
The safest pattern is draft, review, send: the system prepares the order update, the email or the report, and a named person approves it before it reaches a customer, a supplier or the books. Review in batches (for example, once in the morning and once after lunch) so the check takes minutes, not the whole day. If the same reviewer approves the same kind of draft for weeks without changes, you can decide to sample instead of checking every item, but make that a deliberate decision, not a habit that crept in. The same principle applies to AI-written text; see AI for shops.
Test with a small batch
Run the new process in parallel with the old one for a week, or on a small slice: one product category, one sales channel, the first 20 orders. Compare the results line by line. Write down what to do if it fails: who notices, how you switch back and who fixes the data.
Measure before and after
Use the same time log from step 1. After a month, log the same tasks again for one week and compare:
- minutes per task, including the time spent reviewing;
- number of errors found (wrong stock, wrong address, missing order);
- complaints or questions from customers about the affected step.
If the time saved is small or the errors went up, change the process or switch it off. Automation that nobody measures tends to stay forever, whether it helps or not.
Common mistakes
- Automating before cleaning the data. Duplicate products or customers get copied faster, not fixed.
- No owner. Every automation needs a person who checks it and knows how to stop it.
- Removing the human check too early, especially on messages to customers.
- Counting only the time saved, not the setup, the subscription or the hours spent fixing it.
- Too many tools. Each new tool is another place where data can disagree; see managing your business in one system.
Many of these tasks exist because the same data lives in several places.