How I decide what to automate

Typing by hand can be the right answer. Three steps tell me when it isn't.

Before I suggest building anything, I find what holds the business back, mark what can be judged without numbers, and measure one ordinary week. Then I check the build against four questions. They're all here, so you can run them on your own workflow.

One rule runs through all of it: a saved hour only counts if you can cash it in.

1Find what holds the business back

Before I look at what to automate, I ask what holds the business back. The answer decides what a saved hour is worth.

Sometimes it's one team, like an order desk where orders wait. Sometimes it's the whole business: not enough orders, or people who are hard to find. That's the idea behind the Theory of Constraints.

Say a wholesaler's order desk types every order into the ERP by hand. If orders wait at the desk, hours saved there help straight away. If something else holds the business back, they help only if they reach it.

So I write it down in one line before I count anything.

2Mark what can be judged without numbers

Some steps can be judged before anyone counts anything. I give each step of the workflow one of three marks.

Leave aloneThe step needs judgement, a conversation with a customer, or the company's own know-how. A person does it better, whatever the numbers say.
Fix without softwareBefore building anything, I check whether the webshop or ERP you already have could do it. Some steps also go away once customers send what's always missing.
A candidateThe step is copying, not deciding: the same few layouts, typed the same way each time. Whether it's worth automating waits for step 3.Automate, if the numbers say so

At the wholesaler's desk, calling a customer about an unclear order is left alone. A quote request typed into both the CRM and the ERP is a fix without software, if the two can share it. Typing the order in is the candidate.

3Measure one ordinary week

The candidates need numbers. I count one ordinary week, and three numbers are enough.

How many come in.
How many minutes one takes to type.
How many need sorting out by hand, and the minutes each one takes. Some are unclear, so someone chases them before they can be typed: a delivery date is missing, or a part number. Others come back wrong and get corrected after they're saved.

The third number can turn the answer around. Say the desk gets 60 orders a week, at 6 minutes each to type. That's about 276 hours a year. If 1 in 4 is unclear and takes 40 minutes to sort out, that's another 460. Software that types would save at most the 276. The 460 would stay.

One ordinary week

Come in

Minutes to type one

Need sorting out

Minutes to sort one out

Then the size. If it all comes to under 3 hours a week, I'd leave it alone. At that size, typing by hand is the right answer.

Three hours a week is about 140 hours a year. Even if software took four fifths of it, about 110 hours come back. At your own cost per hour, that has to pay for building it and keeping it running.

The 3-hour line

3 h a week × 46 weeks

≈ 140 h a year

Four fifths of that

≈ 110 h back a year

Under it: leave alone

What changes the answer

Five things push it one way or the other. These are the rules I mark by.

What I find outTowards automatingTowards leaving it alone

How many come in a weekAutomate ifManyLeave ifFew: under 3 hours a week in all

How long one takes to typeAutomate ifLong, with many lines and fieldsLeave ifA few minutes

How many need sorting outAutomate ifFew, and quick to sort outLeave ifMany: sort those out first

How alike they areAutomate ifThe same few layouts every timeLeave ifEvery one looks different

What holds the business backAutomate ifThis team: saved hours help straight awayLeave ifSomething else, and the hours can't reach it

How it comes out

Three ways those five can combine. The numbers are examples, and a year is 46 working weeks.

Don't build anything: under 3 hours a week

Say 20 orders come in a week, at 7 minutes each. That's 2 hours 20 minutes, about 107 hours a year. Typing by hand is the right answer.

20 orders × 7 min

= 2 h 20 min a week

≈ 107 h a year

Don't build anything

Wait: the hours are in the orders that need sorting out

That's the desk from step 3: 460 hours sorting out against 276 typing. I'd ask customers for what's always missing before they order, such as the article number or the delivery date. That needs nothing built. Then I'd measure the typing again.

15 unclear × 40 min

≈ 460 h a year

60 orders × 6 min typing

≈ 276 h a year

Wait: sort those out first

Go ahead: many, alike, and somewhere for the hours to go

Say 150 orders come in a week, at 8 minutes each. That's 20 hours a week, about 920 a year. Most come from a few customers, in the same layout every time. Software can fill those into the ERP, with a person checking each one.

At your own cost per hour, four fifths of those hours, about 736, is the most building and running it could be worth in a year. And only if those hours are cashable, which is the next part.

150 orders × 8 min

= 20 h a week

≈ 920 h a year

Go ahead, for the regular ones

Typing by hand stops being the right answer when there's more than 3 hours of it a week, most of it looks alike, and the saved hours have somewhere to go.

Somewhere for the hours to go

A saved hour only counts if you can cash it in. I call those cashable hours. There are two ways to get them: you pay for fewer hours, or the business does more of what it needs.

Say software saves the order desk 10 hours a week. Those hours can go four places.

Fewer paid hours. The overtime stops, a temp's hours go, or a vacancy stays unfilled. At your own cost per hour, that's money you no longer spend.
More orders through the desk. If the desk is what holds the business back, orders wait there. With 10 hours back, it gets through more of them.
Winning more orders. If what holds the business back is too few orders, a quicker desk adds nothing by itself. The hours count only if they go to winning orders, such as following up quotes.
Absorbed into the day. Nobody is paid for fewer hours, and nothing more gets done. The build cost money and gave back nothing you can spend.

Where 10 saved hours can go

10 hours saved a week

Fewer paid hoursCashable
More orders through the deskCashable if orders wait there
Winning more ordersCashable if orders are short
Absorbed into the dayNot cashable

Only cashable hours count. So when I do the sums below, I leave the rest out.

Before you build, answer four questions

I use this check before I suggest building anything.

How many hours does it cost you?Turn the price into hours using your own cost per hour. Add your people's time to set it up.
How many hours come back each month?Hours now, minus hours after, minus checking and upkeep.
How many months until those hours come back?The hours it costs, divided by the hours back each month.
Will those hours still matter then?It must run that long. The hours must go to what holds the business back, or to fewer paid hours.

If you can't answer one, measure first.

Most of what companies automate shouldn't be.

Before anyone quotes you a build, count one ordinary week: how many orders come in, and the minutes to type one. The calculator turns them into hours a week, at your own cost per hour.

Work out your hours