What is business process automation? From Excel macros to AI agents

1. What is business process automation?

⚡ In one sentence

Business automation, or Business Process Automation (BPA), means replacing manual, repetitive tasks with technology, so that a process runs faster, with fewer errors and without constant human supervision.

💡 In plain terms

Let's look at something every company does: responding to customer inquiries.

A customer asks about a price, a deadline or availability. Someone on the team (or the owner themselves) checks the information, prepares a reply and sends it by email. The more inquiries there are, the more time is consumed by repeating the same actions: checking, copying, writing.

Automating this step means the system itself logs the inquiry, suggests a reply based on current data, and lets you send it with a single click. Minutes instead of hours. And instead of the risk that an inquiry gets lost in the inbox, every one has its own status and deadline.

Automation does not eliminate the human. It eliminates the manual searching for data, the re-keying and the email coordination. People gain time for what requires their skills: talking to the customer, negotiating, building relationships.

🔧 Deep dive

Automation is a broad concept that, in practice, covers several approaches. RPA (Robotic Process Automation) is a technology in which "software robots" mimic a person's actions on a computer: they log into systems, copy data, fill in forms. RPA works particularly well where a company has many systems that do not communicate with one another (we cover this problem in the article What is systems integration?). At the other end of the spectrum is intelligent automation (Intelligent Process Automation, IPA), a combination of automation and AI, where the system not only performs tasks but "understands" documents, interprets text and makes decisions.

The key is not choosing a single technology, but matching the right level of automation to a specific process. It always starts with a diagnosis. Only then comes the choice of tool (see What is process optimization?).

2. From an Excel macro to an AI agent: the spectrum of automation

⚡ In one sentence

Automation is not "all or nothing". It is a spectrum from the simplest macro to an intelligent system, and most companies should start at the bottom, not the top.

💡 In plain terms

Let's stay with customer inquiries, but show two different approaches:

Simple automation: A customer sends an inquiry. The system automatically logs it, assigns a number and sends the customer a confirmation: "We've received your inquiry and will reply within 24 hours." Simple. But if the customer wrote "hey, I need something similar to what I ordered last year, but in a different color" - a rule-based system won't be able to handle that.

Intelligent automation: The system "reads" the content of the message, recognizes that it's a request for a quote (and not a complaint or an invoice query), finds the customer's previous order and prepares a draft reply that takes their history into account. A person checks and sends it, but doesn't start from scratch.

Both approaches have their place. It all depends on the process and the stage the company is at. And between them lies a whole spectrum of solutions:

• Macros and rules - a report generated every Monday without manually copying data, a rule in the inbox that routes customer inquiries to the right folder. The simplest solutions, yet they can save hours every week.
• Workflow - a fixed sequence: "when a customer inquiry comes in, it first goes to Anna; if Anna doesn't respond within 4 hours, the system sends a reminder; after 24 hours it escalates to a supervisor." The way of working is written down once, and the system enforces it every time.
• RPA (Robotic Process Automation) - a program that does what a person does manually on a computer: it opens a system, copies data into a second system, fills in a form, generates a document. Except it doesn't forget, doesn't make mistakes, and works 24 hours a day.
• Automation with AI elements - the system sorts correspondence: it recognizes orders, complaints and general questions and routes them to the right person. In manufacturing, AI takes over visual quality control.
• AI agent - a customer writes "when will I get my order?", and the system itself checks the status, generates a reply and sends it. It only hands over to a person what it can't handle on its own (more in the article What is RAG and an AI agent?).

🔧 Deep dive

In practice, it helps to distinguish three stages a company may be at, because that determines what kind of automation makes sense:

• Digitization - documents are digital (scans, PDFs, Excel), but processes are still manual. Someone still opens the file and re-keys the data. At this stage it's worth starting with a simple workflow.
• Digitalization - processes run digitally: the system generates a quote, tracks orders, sends notifications. At this stage the company is ready to automate repetitive processes (RPA, business rules).
• Digital transformation - a change in the operating model: the customer configures an order themselves on a portal, the system suggests a discount based on purchase history. At this stage intelligent automation and AI come into play.

"We often meet companies that want to deploy advanced solutions straight away, while the core information - price lists, terms, customer data - is scattered and duplicated across several places. That's why we always start with the fundamentals. If they are solid, every further level of automation is deployed faster and more cheaply."

- Karol Jurewicz, process optimization, cm-opti

3. Where to start? The Pareto principle in automation

⚡ In one sentence

The Pareto principle - known as the 80/20 rule, where 20% of causes generate 80% of effects - helps answer the most important question in automation: where to start?

💡 In plain terms

In a company, this principle means something very practical: a handful of tasks consume most of the team's time. And it's precisely those tasks that are worth automating first.

Interestingly, the principle can be applied twice, and then it becomes truly precise. Let's see it with an example:

A service company with 20 employees. Across the customer service process, from inquiry to after-sales support, there are many repetitive tasks. But when the owner looks closely, it turns out that the vast majority of the team's time goes into three things: preparing quotes, recording orders and handling complaints. That's the first application of Pareto: a few areas consume most of the time.

Now let's take one of them: preparing quotes. Within that area there are many steps, but two eat up the most hours: manually looking up current prices in spreadsheets and copying customer data from an email into a template. That's the second application of Pareto: a few tasks within a single area account for most of the problem.

Automating those two tasks - a price list connected to the quoting system, customer data pulled in automatically - cuts the time to prepare a quote from two hours to fifteen minutes. No revolution. No new system. Just connecting what the company already has. In practice, the handful of tasks that this double Pareto leads to are often no more than 4-5% of all tasks in the company, yet they account for over 60% of lost time.

🔧 Deep dive

The Pareto principle is a prioritization tool used in Six Sigma (the Pareto chart is one of the seven basic quality tools), Lean Management and ABC analysis in logistics. In the context of automation, the double Pareto guards against two traps: "let's automate everything" (expensive, disproportionate effort) and "let's automate the easiest things" (minimal impact on the business). The double Pareto points to a third path: automate what is simultaneously simple to implement and has the greatest impact. This is exactly what we described as the impact/effort matrix in the article on process optimization.

The double Pareto is simply applying the same principle twice: first at the level of the whole company, then within the chosen area. In large organizations with hundreds of repetitive tasks it can even be applied three times (20% × 20% × 20% = 0.8% of tasks → ~51% of the effect). For an SME, two levels are plenty.

"When a client says 'we want to automate our processes', the first question isn't 'which tool?', but 'where in your customer service process are you losing the most time on tasks that don't require thinking?'. The answer usually points to a few specific steps in two or three processes. Automating those steps, even with the simplest tool, can free up a dozen or more hours a week."

- Michael Jan Rogocki, AI solution architecture, cm-opti

4. What does it look like in practice? From a manual process to an intelligent system

⚡ In one sentence

Automation is not a one-off project. It's a step-by-step process in which each stage delivers measurable value in its own right.

💡 In plain terms

Whatever area of the company you automate, the path looks similar:

Stage 0 - Manual. Everything works, but it rests on people, their memory and their commitment. At greater scale, problems appear: information gets lost, response times grow, and continuity of work depends on specific individuals.

Stage 1 - Putting things in order. The team writes down how a given area should work: who is responsible for what, what the rules are, what follows what. The result is a clear process that doesn't depend on who happens to be at work. That alone speeds things up, and the only investment is the time spent analyzing and organizing what already exists.

Stage 2 - Simple automation. The system takes over repetitive tasks: it records, assigns, generates documents from templates, sends notifications, keeps an eye on deadlines.

Stage 3 - Intelligent automation. The system doesn't just execute, it "understands": it interprets the content of a message, recognizes the type of case, suggests a reply. The human approves rather than starting from scratch.

You don't have to aim straight for stage 3. It's enough to start with the first one, because the further levels of automation are only possible once the foundation is working.

🔧 Deep dive

The staged model above reflects the approach cm-opti applies in practice. In one project for an insurance company, the move from manually handling claims correspondence, through writing down classification rules, to deploying an NLP agent delivered measurable results: a 30% reduction in the document-processing backlog, around 25% less correspondence requiring manual handling, and 90% accuracy in automatically recognizing the risk category. A full description of the technology is in the article What are OCR and NLP, and how does AI read documents?.

A key observation: stage 1 (putting things in order) is critical to the success of stage 3 (AI). Without clearly defined categories and rules, there's no way to judge whether the automation system is working correctly. There's nothing to compare against. The effects of technology are always proportional to the quality of the process that feeds it (cf. What is artificial intelligence?). Measuring those effects - KPIs, dashboards, comparing the "before" and "after" - is the domain of data analysis and BI.

5. Business process automation examples, department by department

⚡ In one sentence

Business automation delivers quick results in repetitive processes based on documents and rules, such as invoice workflows, order handling, logistics, HR matters and reporting.

💡 In plain terms

Areas where business process automation can be introduced in stages and its effect measured, and the technologies that work well in them:

Automating accounting processes

A classic example is the workflow for supplier invoices. The invoice enters the system, the data is read automatically, the software suggests account coding based on the supplier's history and passes the document to the right person for approval. The accountant checks and approves instead of retyping. Mandatory structured e-invoicing makes this even easier: in Poland the National e-Invoicing System (KSeF) has been mandatory since 2026, and in Germany companies have had to be able to receive e-invoices in formats such as XRechnung or ZUGFeRD since 2025. Structured invoices arrive as XML, so the data can be pulled in directly, without reading scans. OCR is still useful for foreign invoices, receipts and other documents.

Automating logistics processes

In logistics, most time goes into retyping shipping documents and statuses between systems. In one of our projects, data from scans and emails is read and entered into the TMS in the background, and the employee only checks and approves it. Handling a single document went from minutes to seconds (case study: TMS integration).

Automating HR processes

In HR, automation mainly covers request workflows and deadline tracking: preparing documents for a new employee, creating accounts, reminders about medical check-ups and health and safety training, leave approvals. Recurring employee questions ("how many days of leave do I have left?", "how do I claim travel expenses?") can be answered by an AI assistant based on the company's policies.

Automating reporting and office processes

A monthly management report put together by hand from several sources can take several days of work. At a company where Karol Jurewicz led such a project, standardizing the KPIs and automating data retrieval cut its preparation from 2.5 days to 2 hours (more details).

🔧 Deep dive

A process is suitable for automation when it meets most of five conditions:

• Volume - the task recurs often (daily or weekly, in dozens or hundreds of cases).
• Rules - decisions can be described ("if the amount is above X, the director approves").
• Digital data - the information is in systems, emails or files, not only in people's heads.
• Stability - the process does not change every month.
• Cost of error - manual mistakes are expensive or risky.

A simple estimate of the potential: number of cases per month × manual handling time per case. With 400 invoices a month and 4 minutes each, that is about 27 hours of work that can be won back, before counting the cost of errors and delays. The best starting point for this calculation is a process map with the times written down.

6. RPA bots or AI agents: which to choose?

⚡ In one sentence

An RPA robot performs programmed steps on structured data and always does the same thing, while an AI agent plans its next actions itself, understands unstructured content and uses tools, which makes it more flexible but requires oversight.

💡 In plain terms

The two approaches can be combined. An AI agent reads a customer's email, recognizes that it is an order and extracts the data. Writing it into the ERP is then done by a deterministic integration or an RPA robot that always does it the same way.

If the system has an API, a direct integration is more stable and cheaper to maintain than a robot that "clicks" around the screen, because it does not break when the application's layout changes. RPA is a good solution where there is no API or opening one up does not pay off. More on how agents work in the article What is RAG and an AI agent?

🔧 Deep dive

With AI agents, oversight has to be designed in (human-in-the-loop). The system acts on its own only when its confidence exceeds a set threshold, and doubtful cases go to a person. This is how our implementation at an insurer works: the risk category is recognized correctly in 90% of cases, so clear cases are routed automatically and doubtful ones are checked by an employee (case study: claims correspondence classification).

The second element is data. The agent has access to the company's systems and documents, so you need to define clearly which tools it may use, which operations it may perform and where the data is processed. Data should stay in a secure, agreed environment compliant with the GDPR, and every step of the agent should be logged so that it can be reconstructed and demonstrated in an audit.

7. What not to automate, and how to recognize it

⚡ In one sentence

Don't automate processes you don't understand yourself, ones that change every week, or ones that require empathy and judgment. There, people are irreplaceable.

💡 In plain terms

Automation makes sense where tasks are frequent, repetitive and based on clear rules. Where those conditions are missing, it's better to leave the work to people:

• Situations that require empathy and judgment - a frustrated customer after a third faulty delivery needs a conversation, not an automated reply. AI can prepare the data, but a person handles the conversation.
• Areas that are still taking shape - if the company is testing different approaches, automation will freeze one variant.
• Tasks no one can describe - if handling something requires a "feel" for it, you first have to uncover the rules (cf. process mapping).
• Low-volume tasks - if something is done once a week for 10 minutes, the cost of automation will outweigh the savings.

🔧 Deep dive

An important note: automation doesn't have to cover an entire area of a company's operations. Even where judgment and relationships dominate, e.g. negotiations with a key client, some tasks are purely mechanical: preparing a quote from the price list, checking order history, calculating the margin. Often it's enough to automate just one such step to unblock the whole flow and give people room for work that requires their judgment and experience.

8. The first step

You don't have to automate the whole company. It's enough to choose one area - the one that, by the Pareto principle, consumes the most time on repetitive, mechanical tasks, and start with the simplest solution.

The first automation doesn't have to be spectacular. It can be a rule in the inbox, a document template that fills itself in, or a simple notification. One thing matters: measure the effect. How much time was saved? How many fewer mistakes? How much faster did the customer get a reply? Those numbers are the best argument for the next step.

We work with companies in Poland and Germany, from small businesses to large enterprises. Whatever the scale, automating a company starts the same way: with a diagnosis, not with a tool. Together with the client's team we look for those few tasks that consume time without delivering proportionate value. Only then do we choose a solution, one suited to the problem and to the stage the company is at. Automating an SME doesn't require big budgets. It requires the right starting point.

- The cm-opti perspective

Frequently asked questions

What is business automation?

Business automation, i.e. business process automation, means moving repetitive, rule-based tasks from people to systems: from rules in the email inbox and workflows, through RPA robots, to intelligent automation with elements of AI. The aim is a faster process, fewer errors and team time won back for work that requires judgment.

Which processes in a company are best suited to automation?

Frequent, repetitive processes based on clear rules and digital data: invoice workflows, order entry, logistics documents, HR requests, periodic reporting. It is best to start with the one that takes up most of the team's time.

What is the difference between an RPA robot and an AI agent?

An RPA robot replays a programmed script on structured data and always works the same way. An AI agent understands unstructured content such as emails or documents, plans its next steps itself and uses tools, but needs confidence thresholds and human oversight. In practice the two are often combined.

Does process automation require a large budget?

No. The first automations can be built with tools the company already has: rules in the email inbox, document templates, simple functions in existing systems.

Will automation replace my employees?

No. Automation takes over repetitive tasks, and people gain time for work that requires their skills and judgment.

What is the difference between automation and artificial intelligence?

Automation is a broad concept, from simple rules to intelligent systems. AI is one of the tools of automation, useful where text, images or unstructured data need to be interpreted. More about AI in the article What is artificial intelligence?.

Where should a small company start with automation?

With a single area that is repetitive, based on clear rules and consumes a disproportionate amount of time. The Pareto principle helps you find it.

Want to find that first area to automate? Let's talk - together we'll find the process worth starting with.

Related articles in the cm-opti knowledge base

• What is artificial intelligence?
• What is process optimization?
• Process mapping - how to create a process map step by step
• What are OCR and NLP, and how does AI read documents?
• What is RAG and an AI agent?
• What is computer vision?
• What is systems integration?
• What is data analysis and BI?

Concepts explained in this article → Glossary

Automation, Business Process Automation (BPA), Robotic Process Automation (RPA), Intelligent Process Automation (IPA), workflow, digitization, digitalization, digital transformation, NLP, OCR, e-invoicing (KSeF, XRechnung), API, AI agent, human-in-the-loop, the Pareto principle

Sources and references

• The Pareto principle (80/20) - Vilfredo Pareto, "Cours d'Économie Politique" (1906); application in quality management: Joseph M. Juran (1940s).
• The distinction between BPA, RPA and IPA - based on publicly available industry knowledge.
• Poland's National e-Invoicing System (KSeF), mandatory from 2026: Crowe (in Polish)
• E-invoicing in Germany (XRechnung, ZUGFeRD): IHK Gießen-Friedberg (in German)
• cm-opti project results: described in the Case studies section