Business process automation: where to start

"We need to automate everything" is a phrase you hear a lot, and it's almost always wrong.

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

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

What automation is, and what it isn't

Automation is moving repetitive work with clear rules into a system. A person makes the decision, the system executes it.

What it isn't:

• It's not replacing an employee with software. Headcount usually doesn't shrink — the person moves to different work. And as the company grows, the need to hire someone new simply arrives later.
• It's not digitizing every process in a row. Some processes are faster and cheaper done by hand.
• It's not a one-time project. It's a sequence of steps: one process first, then the next.

Where to start: three criteria

Score each process against three questions.

1. How often does it repeat? Several times a day is high frequency. Automating something that happens once a year almost never pays off.

2. How clear are its rules? Can you write the process as "if this, then that"? If yes, the rules are clear. If a person has to think it through and decide every single time, the process isn't ready for automation yet.

3. What does a mistake cost? What happens when it goes wrong: does a customer leave, does a fine land, does stock go missing? If the mistake is cheap, automating it isn't urgent.

The top-right quadrant goes first. The bottom-left quadrant may never get automated, and that's fine.

Turn it into numbers

Criteria are useful, but the decision gets made on numbers. Fill out this table for every candidate process.

Example. Document approval: 200 times a month, 15 minutes each, three people involved, average hourly cost $3.

200 × 0.25 hours × 3 people × $3 = $450 a month, or $5,400 a year.

On top of that come errors and delays. If automation saves 70% of that time, that's $3,800 a year. At a project cost of $8,000, it pays for itself in roughly 25 months. That's a borderline case, which is exactly why it needs to be compared against other candidates.

Five steps

1. Document the process as it actually is

Not the ideal version, the real one. Who starts it, who approves it, where it stalls. This step alone is often where you discover that part of the process isn't needed at all.

2. Cut what's unnecessary

This is the cheapest step, and the one most often skipped. If the 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'll end up digitizing an unnecessary step too.

3. Write the rules down

"If the amount exceeds 5 million sum, the director approves it." Every exception gets written down too. A rule that can't be written down can't be automated either.

4. Launch and run in parallel

For the first month, the old and new processes run side by side. It's extra load, but it surfaces mistakes before the full switchover.

5. Measure

A month after launch, re-measure the numbers you recorded at the start: frequency, time, error count. Those numbers become the basis for deciding on the next process.

Four mistakes that come up most often

Digitizing chaos. If the process is a mess, putting it into a system just gets you a digitized mess. Order first, system second.

Too much at once. Automating seven processes at the same time is too big a change for employees to absorb. Resistance builds, and none of the processes actually gets finished.

Employees left out of the project. The system gets handed down from above, and one day an employee just sees a new screen. The fix: have one or two people from each department involved throughout the project.

No baseline measurements. If the numbers weren't recorded at the start, there's nothing to point to later when someone asks "did this actually get better." And asking for budget for the next project gets harder.

When not to automate

• The process repeats less than a few times a month
• The rules haven't settled yet and change every month
• The mistake is cheap and easy to fix
• The process is planned to change completely within six months
• Both the process and the person running it will be gone within a few months

Two matches, and waiting is the cheaper option. That's not inaction, it's savings.

What it costs

Depending on scale, automation comes in three formats:

The single biggest driver of price is the number of integrations. Connecting to each existing system takes one to three weeks.

Bottom line

Automation doesn't start with "everything." It starts with one process, and that process gets chosen using three criteria.

The practical order:

1. List five candidate processes
2. Calculate the monthly cost for each — frequency × time × people × hourly cost
3. Score them against the three criteria and start with the highest scorer
4. Cut unnecessary steps before automating, not after
5. Record the baseline numbers — a month later they'll show the result