What to automate first: a 4-factor scoring method

Automate first the task that happens often, takes real time, costs you when it goes wrong, and follows the same rules every time. Often that is data entry between two systems, such as copying orders or invoices into accounting. It is not always the task people complain about most.

To check this for your own business, use a simple scoring method with four factors: how often the task happens, how long it takes, what a mistake costs, and whether it follows clear rules. You can run it on a whiteboard in under an hour.

What most advice on this topic says

Search for “what to automate first” and you will find the same short lists: lead follow-up, invoicing, scheduling, reminders, reporting. Those lists are not wrong. But they are generic, and many come from software vendors with a product to sell.

Three pieces of advice are widely repeated, and we agree with all of them:

• Fix the process before you automate it. Automating a task that two people do in three different ways locks in one way and breaks the others.
• Keep a person in the loop for risky work. Where money, contracts or key relationships are involved, let the automation prepare and flag, and let a human approve.
• Start with one task, not everything. Build one, check it works, then move on.

What the generic lists miss is your business. Your first candidate depends on your volumes, your systems and your error rate. That is what scoring is for.

The scoring method

Give each task a score from 1 to 5 on four factors. Multiply the four numbers. The highest product goes first.

Score = Frequency x Time x Error cost x Rule-based

Why multiply instead of add? Because a task that scores 1 on any factor is a poor candidate, and multiplication shows that. A task that happens once a month and needs judgment every time will never win, however long it takes.

Use 2 and 4 for cases in between. You can adjust the bands to your business. A 15-minute task is small in a firm of 200 people and large in a firm of five. The point is to use the same bands for every task.

A note on the “Time” factor: a task that takes 2 minutes but happens 200 times a day is a big job. Frequency and time together give you the total hours. If you track hours directly, you can score that instead.

How to run it

1. List the candidates. Ask each team: “What do you copy, paste, re-type, chase or check every week?” Aim for 15 to 25 tasks.
2. Estimate, do not measure. Rough numbers are fine at this stage. If two tasks are close, measure them later.
3. Score each task on the four factors. Do it as a group, so people can challenge each other.
4. Multiply and rank. Look at the top five.
5. Sense-check the top three. Is the process stable? Is the data in one place? Does anyone own it? If not, fix that first.
6. Pick one. Build it, measure it, then move to the next.

For a rough view of what your manual work costs per year, try the time leak calculator.

Worked example

Take a store with a team of eight handling about 40 orders a day. The owner lists five tasks. The numbers below are illustrative, not from a real business.

The accounting copy-paste wins with 300. It happens all day, takes a few minutes each time, and errors hit the books. It follows fixed rules: same fields, same destination.

The custom order approval scores lowest, and that is correct. It needs judgment, happens rarely, and should stay with a person.

Now the numbers behind the winner. Say copying one order takes 3 minutes and there are 40 orders a day, on 250 working days:

• 40 orders x 3 minutes = 120 minutes a day, or 2 hours
• 2 hours x 250 days = 500 hours a year

At an illustrative cost of 25 per hour, that is 12,500 a year of work that adds nothing. Add the cost of the occasional mis-keyed order and the case is clear. If building and testing the automation took 40 hours, the hours saved would cover that in about a month. Your own numbers will differ. The method does not.

Delete or simplify before you automate

Some tasks should not be automated at all. Ask three questions first:

• Does anyone read this report, or use this approval step?
• Is the same data typed into two places that could be one?
• Is there a step that exists only because of an old system?

If the answer shows the task is not needed, remove it. A deleted task costs nothing to run and nothing to maintain.

Rules or AI?

Rule-based automation does the same thing every time: if the order is over a set amount, tag it and email the team. It is predictable and easy to test. AI tools can read free text, such as an email, and draft a reply or pull out an order number. They can also get things wrong, so keep a person checking anything that touches money or customers. Score both kinds of task the same way. A task that is “needs judgment every time” scores 1 on the rule factor, with or without AI.

Choosing the right tool

Once you know what to automate, choose the lightest tool that works:

• Built into your tools. Many systems have rules or flows built in. Check these first.
• Connector tools such as Make or Zapier. Good for moving data between apps when the rules are clear.
• A custom internal tool. Worth it when the process is central to your business, involves several systems, or needs screens your team will use daily.

Our automation and software service covers all three. If you run on Shopify, the guide to automating the Shopify back office shows common flows.

When NOT to automate

Automation is not always the answer. Hold off when:

• The process is unstable. If the steps change every month, you will rebuild the automation every month. Stabilize and document it first. Our guide on how to write an SOP helps.
• Each case needs judgment. Complaints, negotiations and exceptions need a person.
• Volume is low. A task done twice a month for five minutes is not worth building and maintaining anything for.
• The data is messy. Automation moves bad data faster. Clean the source first.
• Nobody owns the process. An automation without an owner breaks silently.
• The mistake is expensive and hard to reverse. Paying suppliers or refunding customers can be prepared by a machine, but should have a human approval step.
• You do not understand the process yet. Map it first. See process mapping for small businesses.

A good question to ask: “If this automation fails at 2 a.m., who finds out, and how?” If the answer is “nobody”, add monitoring before you go live.

Test before you trust it

Before an automation touches real customers or money:

1. Dry run. Run it on test data or a copy, and compare the result with what a person would have done.
2. Notify-only mode. Let it send a message saying what it would do, without doing it, for a week or two.
3. Try to break it. Feed it a blank field, a duplicate, an odd character and an out-of-range number.
4. Add an off switch and an owner. One named person can pause it and gets the failure alerts.

What to do after the first win

Once the first automation runs, check it after one week and again after one month. Note the time saved and the errors avoided. Then take the next item on your ranked list. Re-score every six months, because volumes and tools change.

Key takeaways

• Score each task from 1 to 5 on frequency, time, error cost and rule-based, then multiply.
• Multiplication pushes down tasks that fail on any factor.
• Estimate first, measure only the top few candidates.
• Fix and document the process before you automate it, and delete steps nobody needs.
• Keep a human approving anything that moves money or affects key customers.
• Do not automate unstable, low-volume or judgment-heavy work.

If you would like help scoring your own list, we offer a free audit.