"We need to automate everything" comes up often, and it is almost always wrong.

Automating everything at once produces three things: the budget grows, staff resist, and the system ends up half-working.

The right question is different: which single process to start with.

What automation is, and what it is not

Automation is moving repeatable, clearly defined work into a system. A person makes the decision, the system carries it out.

What it is not:

• Not replacing a person with software. Headcount usually does not shrink — the person moves to other work. During growth, it simply delays when you need to hire the next one.
• Not digitising every process. Some processes stay faster and cheaper done by hand.
• Not a one-off project. It is a sequence: one process goes live, then the next.

Where to start: three criteria

Score each process against these three questions.

1. How often does it repeat? Several times a day is high. Something done once a year almost never justifies automating.

2. How clear is the rule? Can the process be written as "if this, then that"? If so, the rule is clear. If a person has to think it through every time, it is not ready for automation yet.

3. What does a mistake cost? What happens if it goes wrong — does a client leave, does a fine follow, does stock go missing? If a mistake is cheap, automation is not urgent.

The upper-right quadrant comes first. The lower-left quadrant may never need automating, and that is fine.

Turn it into a number

The criteria help, but the decision gets made with a number. Fill in this table for every candidate process.

A worked example. Document approval: 200 times a month, 15 minutes each time, three people involved, average hourly cost $8.

200 × 0.25 hours × 3 people × $8 = $1,200 a month, or $14,400 a year.

Add errors and delays on top. If automation saves 70% of that time, that is roughly $10,000 a year. If the project costs $9,000, it pays for itself in about eleven months. That is a good return, though every candidate is still worth comparing against the others.

Five steps

1. Document the process as it actually runs

Write down the real process, not the ideal one. Who starts it, who approves it, where it stalls. This step alone often reveals that part of the process is unnecessary.

2. Remove what is unnecessary

The cheapest step, and the one most often skipped. If an approval chain has four people and two of them have never once rejected anything, they come out of the chain. Do this before automating, or you end up digitising a step that never needed to exist.

3. Write the rule down explicitly

"If the amount exceeds $2,000, the director approves it." Every exception gets written down too. A rule that cannot be written down cannot be automated.

4. Launch and run in parallel

For the first month, the old and new methods run side by side. It is extra work, but it surfaces errors before the old method is retired.

5. Measure

Re-measure the same figures recorded at the start — frequency, time, error count — a month later. Those numbers become the basis for deciding on the next process.

The four most common mistakes

Digitising disorder. If the process is a mess, putting it into a system produces a digitised mess. Order comes first, the system comes second.

Starting too many at once. Automating seven processes simultaneously is too much change for staff at once. Resistance rises and none of them lands fully.

Leaving staff out of the process. The system arrives from above and one day an employee sees a new screen. The fix: bring one or two people from each department into the project as it happens.

Not measuring the result. If no baseline was recorded, there is no answer to "did it actually help" afterwards — and asking for budget on the next project becomes harder.

When not to automate

• The process repeats fewer than a few times a month
• The rule has not settled yet and changes every month
• A mistake is cheap and easily corrected
• The process is planned to change entirely within six months
• The person who runs it, and the process itself, will not exist in the near future

Two of these are enough to wait. That is not inefficiency — it is being economical.

What it costs

Automation comes in three sizes depending on scope:

The factor that affects price most is the number of integrations. Each connection to an existing system takes one to three weeks.

In summary

Automation does not start with "everything." It starts with one process, and that process is chosen against three criteria.

Practical steps:

1. Write down five candidate processes
2. Calculate the monthly value for each — frequency × time × people × hourly cost
3. Score them against the three criteria and start with the highest
4. Remove unnecessary steps before automating
5. Record baseline numbers — a month later they show the result