Business Process Automation: Where to Start

The first process to automate is one that repeats at least weekly, follows the same rules every time, takes several hours a week in total, and costs money when it is done wrong by hand. Score candidates on these four measures from 1 to 5, for a maximum of 20. A process scoring 15 or more is a good first target. Narrow it to a single step and run a pilot in two weeks.

Most growing companies have a process everyone agrees should be automated. It comes up in leadership meetings, people complain about it by the coffee machine, and yet it's the same process it was a year ago.

The reason is rarely laziness or a lack of skill. Getting started requires someone to stop and make a choice, while everything else on the calendar is shouting for more urgent attention.

Why Does Getting Started Feel Hard When Everything Is Urgent?

The most common reason automation gets delayed isn't doubt about its value. It's that there are too many candidate processes and too little time to choose between them. When any process could be the next target, the easiest option is to avoid choosing at all.

A week in a growing company fills up fast. The sales team needs support, customers expect answers, and the weekly report has to be on the leadership team's desk by Friday. A single automation project loses this competition for attention because its payoff shows up weeks later, while today's problems demand a fix today.

The result is familiar: a process that would take minutes if automated keeps taking hours every week, year after year. Nobody stops to calculate the cost. Over 45 working weeks, three hours a week adds up to 135 hours a year. At roughly €45 an hour including employer costs, that one step costs about €6,000 a year. According to Statistics Finland, an hour worked already cost employers an average of €34.4 in 2020, and the figure for white-collar work is higher today.

How Do You Identify the Right Process to Automate First?

A good first automation target is a repetitive, rule-based, time-consuming step where information is moved between systems by hand and where mistakes are costly. Those qualities can be measured with four measures. Score each candidate from 1 to 5 and add up the points.

• Frequency: how often does the process repeat? A task that runs daily or weekly delivers a visible payoff much sooner than one that happens once a quarter.
• Time spent: how much time does it take per week in total? Add up the hours of everyone involved. A task that takes one hour from five people is as big as a five-hour task for one person.
• Cost of error: what does a manual mistake cost? If an error shows up in billing, payments or with the customer, the value grows fast, often beyond the time saved alone.
• Rule-based: does the process follow the same rules every time? Clear, stable rules make the implementation fast and inexpensive. If the rules keep changing, aim the first automation at a steadier process.

Points are given against these anchors. Scores of 2 and 4 fall between them.

Measure	5 points	3 points	1 point
Frequency	Daily	Weekly	Monthly or less often
Total time spent	Over 10 hours a week	3–10 hours a week	Under 3 hours a week
Cost of error	Shows up in billing, payments or with the customer	Shows up in internal reporting, fixed after the fact	Noticed and fixed immediately
Rule-based	Same rules, exceptions are rare	Some cases need judgment	Almost every case needs interpretation

How to read the total: 15–20 points is a good first target, 10–14 points goes in the queue for later, and a process scoring under 10 points is not worth automating now.

Here's what the scoring looks like with four typical candidates side by side:

Process	Frequency	Time spent	Cost of error	Rule-based	Total
Processing purchase invoices	5	5	4	4	18
Moving billing data out of the CRM	3	3	5	5	16
Compiling the weekly report	3	3	3	5	14
Quarterly budget round	1	3	3	1	8

The winner isn't the one that sounds the most technically interesting, it's the one with the most points. The budget round touches the most people, but it happens four times a year and requires judgment, so the savings accumulate slowly and the implementation is difficult.

A low rule-based score doesn't mean a process can never be automated. It means rule-based workflow automation won't be enough, and the process needs an AI-assisted implementation, which costs more. That's why the first target should be a process with clear rules.

The same scoring works when the data comes from a device rather than a system. If someone reviews sensor or meter readings every day, compares them against thresholds and reacts to deviations, that is the same repetitive checking work as compiling a report. The basics, and what to do with the sensor data you collect, are covered in a separate article.

Which Processes Are Typical First Automation Targets?

The most typical first automation targets are data transfers in finance and sales: processing purchase invoices, logging leads into the CRM, moving billing data and compiling reports from several sources. They combine two things that raise the score: the work repeats every week with the same logic, and a mistake shows up in money or with the customer.

Process	Why it fits as a first target	Typical technology
Purchase invoices from email into the finance system	High volume, repeats daily, errors show up in payments	AI-assisted (invoices arrive in different formats)
Leads from a web form into the CRM	Fully rule-based, delays cost deals	Workflow automation
Billing data from the CRM into invoicing	Rule-based, errors show up on the customer's invoice	Workflow automation
Compiling a report from several sources	Repeats weekly, takes several people's time	Workflow automation, AI for the summary
Classifying and routing incoming customer messages	Daily, speeds up response times	AI-assisted (messages are free-form text)
Syncing data between systems	Removes double entry, for example between CRM and accounting	Workflow automation

When the target is a plain data transfer between two cloud services, an off-the-shelf tool is often enough. We covered when Zapier, Make or Power Automate is enough and when you need a custom API integration.

What Should You Not Automate First?

A process that repeats rarely, changes constantly or needs case-by-case judgment almost every time is a poor first target. Nor should the first automation be a business-critical process that has no manual fallback in case the automation stops.

• Infrequent processes, such as the budget round or annual planning. The savings accumulate too slowly for a pilot to show a result.
• Constantly changing processes. If the rules change every month, maintaining the automation eats up the savings.
• Cases that need interpretation, such as contract negotiations or unusual customer complaints. They suit a later, AI-assisted phase, once the first automation has proven its value.
• The whole process at once. The broadest process isn't the best starting point. The best one is a narrow, repetitive step within it, where information is copied by hand from one place to another.

How Much Does the First Automation Cost and When Does It Pay for Itself?

Based on Empirica's implementations, automating one target typically costs €5,000–28,000, and a well-chosen target pays for itself in under a year. The price depends above all on whether the automation has to interpret content or only move data. That's why the rule-based measure is part of the scoring: a high score there usually means a less expensive implementation.

The payback period is the implementation price divided by the annual saving. For example, processing purchase invoices that takes 25 hours a week pays for itself in 4–9 months. We walk through the calculations and what drives the price in how much does automation cost. The method for calculating the monthly running cost is in what running AI costs per month.

You can estimate the implementation price and payback period for your own candidates with the automation potential quick estimate, which takes up to four candidates at once.

Workflow Automation or AI-Assisted Automation: Which One Do You Need?

Pick the simplest technology that solves the problem. Rule-based workflow automation handles more cases than leadership teams usually expect, and it's faster and cheaper to set up than an AI-based implementation. The dividing line is whether the task requires judgment.

	Workflow automation	AI-assisted automation
Fits when	Information moves the same way every time	The task requires judgment or reasoning
Exceptions	Each case is coded separately	Adapts without a new rule for every case
Data format	Structured, predictable	Variable: text, attachments, free-form fields
Time to deploy	Days to weeks	Weeks, and it needs test data
Maintenance	Grows with each special case	Stays steadier, needs monitoring
Typical example	Order from CRM into billing	Classifying incoming invoices or messages

Rule-based automation hits its limit without anyone noticing: every new exception adds one more rule, until maintaining it costs as much time as the manual work did. That's the point to switch technology. We covered the distinction in more depth in our guide on autonomous AI agents and how they differ from traditional automation.

Once the level of technology is clear, the next choice is how to acquire it. We cover when an off-the-shelf AI tool is enough and when a custom solution pays off in a separate article.

What Does Automating One Process in Two Weeks Look Like?

A narrow target doesn't need a big project. Two weeks is enough when the first week goes to mapping and the second to a pilot.

The starting point: the sales team logs closed deals in the CRM, but billing details are copied into the invoicing system by hand once a week. The task takes the controller about three hours a week and regularly produces small data entry errors that aren't caught until the following month. The process scores 16 points, so it clears the threshold. This is a textbook example of the hidden cost of manual sales operations.

Week 1: map it out. Pin down exactly what data moves, where it comes from, and in what format. Check whether both systems support an API, or whether a simple integration tool needs to sit in between.

Week 2: pilot it. Implement the automation that moves the data directly, and run it alongside the manual process for one week. Compare results: do the numbers match, and how much time is actually saved?

Once the pilot works, the manual step disappears and the controller's three hours a week go toward other work. Running both in parallel is the single most important step: it proves with numbers that the automation produces the same result, so adoption doesn't stall on doubt.

What Should You Do Tomorrow?

Start by listing three to five processes that took up the most manual time last week and score them on the four measures. These five steps move things forward without a separate project:

1. List three to five processes that took up the most manual time last week.
2. Score them on frequency, time spent, cost of error and how rule-based they are.
3. Pick the highest scorer, provided it reaches 15 points, and narrow it to one clear step.
4. Set a two-week checkpoint to confirm the first version works.
5. Measure the time saved and use it to decide what to automate next.

The same principle applies more broadly, which we covered in 5 signs your company is ready for data-driven decision-making: a small measurable first step matters more than a perfect starting moment.

If the list won't fall into order on its own, a fixed-price Automation Assessment walks through your processes and tells you which targets to take first and what they're worth in euros. Implementations run on Microsoft Azure, and we're a Claude Partner Network member, so where data is processed and who can access it is known from the start.

Automation doesn't need a perfect moment or a big project. It needs one narrow process, a simple way to score and pick it, and two weeks to see a result.

Frequently Asked Questions

Where should business process automation start?

Start by scoring three to five candidate processes on four measures: how often the process repeats, how much time it takes per week in total, what an error in it costs, and whether it follows the same rules every time. Give each 1 to 5 points and pick the highest total. Then narrow that process to one clear step as your first target rather than automating the whole thing at once.

Which processes should be automated first?

The best first targets are repetitive, rule-based, error-prone steps where information moves between systems by hand. Typical examples include processing purchase invoices, logging leads from a web form into the CRM, moving billing details from a CRM into an invoicing system and compiling a report from several sources. Infrequent processes are better left for later, even when they touch a lot of people.

How many points does a good first automation target score?

With four measures on a scale of 1 to 5, the maximum is 20. A process scoring 15 or more is a good first target, one scoring 10–14 should go in the queue, and one scoring under 10 is not worth automating now.

Which processes should not be automated first?

Don't pick a process that repeats rarely, changes constantly or needs case-by-case judgment almost every time. Nor should the first automation be a business-critical process without a manual fallback.

What is the difference between workflow automation and AI-assisted automation?

Workflow automation follows a fixed rule in the form of "when X happens, do Y" and suits processes where information moves the same way every time. AI-assisted automation can interpret context and adapt to exceptions, so it suits tasks where the data varies in format or requires reasoning. The rule of thumb is to pick the simplest technology that solves the problem, because workflow automation is faster and cheaper to deploy.

How long does the first automation take to implement?

A narrow first target typically takes two weeks: the first week goes to mapping (what data moves, from where, in what format) and the second to a pilot run alongside the existing manual process. That parallel run proves with numbers that the automation produces the same result before the manual step is removed.

---

Empirica Finland is a Finnish provider of operational AI and automation that builds automations from the data produced by a company's systems as well as its devices and sensors, and is responsible for keeping those automations running. The Automation Assessment is a fixed-price way to find out where to start.

← Back to home