--- title: "When is process automation worth it? Criteria and calculation – Riccardo Nentwich" description: "When automating a process pays off: five test questions, a traceable calculation and the costs after go-live." url: "https://riccardonentwich.com/en/insights/when-is-process-automation-worth-it/" language: "en" translation: "https://riccardonentwich.com/wissen/wann-lohnt-sich-prozessautomatisierung/" updated: "2026-08-31" --- [Insights](https://riccardonentwich.com/en/insights/)[Insights](https://riccardonentwich.com/en/insights/?type=insights) # When is process automation worth it? Whether automating a process will pay off can be estimated fairly well in advance. It takes five test questions, an honest calculation and a look at the costs after go-live. - Author [Riccardo Nentwich](https://riccardonentwich.com/en/about/) - Published 31 August 2026 - Reading time 7 min Process automation is worth it when a process **comes up frequently**, mostly follows **clear rules**, relies on **information that is available digitally** and has **few exceptions**, and when the effort saved over the period of use is greater than building, running and maintaining the solution. If one of these conditions is missing, simplifying the process or making better use of existing software is often the cheaper answer. This article helps with the question of whether automating one particular process is worth it. Which of several processes is the right place to start is a different question. That one is answered in the article [Which process should you automate first?](https://riccardonentwich.com/en/insights/which-process-to-automate-first/) First ## Simplify first, then automate Before doing any calculations, a simpler question is worth asking: does this step need to happen at all? As early as 1990, Michael Hammer warned against simply pouring outdated processes into software instead of fundamentally rethinking them. The idea still holds. Automate a cumbersome process and you end up with a faster cumbersome process. That is why three checks come first: - **Can the step be dropped?** Some transfers, copies or approvals exist only because things have always been done that way. - **Can the existing software already do it?** Many systems have import, export or rule functions that nobody has set up. - **Can the process be standardised?** If every employee works differently, there is nothing uniform to automate yet. Only what remains after that is a candidate for automation. Suitability ## Five test questions for a specific process The following questions clarify whether a process is technically and organisationally suitable for automation. They do not replace a calculation, but they decide whether a calculation is worth doing. ### 1. Does the process come up often enough? Automation has one-off costs. They are spread across the number of cases. A process that occurs dozens of times a day carries these costs more easily than one that comes up twice a month. **Good sign:** The task is part of day-to-day business and keeps coming back in a similar form. **Warning sign:** Nobody can say how often it actually occurs. Then measure first. ### 2. Does it follow rules that can be written down? What can be described can be automated: if the order number is missing, ask for it. If the amount is below a threshold, approve it. As long as the rules only exist in the head of an experienced employee, they cannot be automated. Clarification is needed first. **Good sign:** Two employees would handle the same case in the same way. **Warning sign:** “It depends” is the most common answer. ### 3. Is the information available digitally? An email, a form, a record in the system: inputs like these are easy to process further. With PDFs and scans, the details first have to be recognised. That is often possible today, but it brings uncertainty and checking effort. Paper, phone calls and things shouted across the office are the most difficult inputs. **Good sign:** The information arrives in a recurring form, for example always the same form or the same fields. **Warning sign:** Every sender provides the details differently, often incompletely. ### 4. How many exceptions are there, and who decides them? Hardly any process is completely uniform. What matters is how large the share of standard cases is and whether exceptions can be reliably detected. A good solution handles the standard cases and puts the exceptions in front of a person. A model from research on human–machine interaction is helpful here: Parasuraman, Sheridan and Wickens (2000) distinguish four functions that can be automated independently of each other: acquiring information, analysing it, making a decision and carrying it out. The model comes from complex technical systems, but the distinction helps in the office too. It is often worth automating the capturing and preparation and leaving the decision with a person. **Good sign:** Most cases run the same way, and deviations can be spotted. **Warning sign:** Almost every case needs an individual decision. ### 5. Will the process stay stable for a while? An automation is tied to systems, formats and rules. If these change frequently, the savings turn into maintenance effort. **Good sign:** The process and the systems involved have run in a similar way for years. **Warning sign:** A system change is planned, or responsibilities are currently being reorganised. Calculation ## What does the process cost today? The basis of any cost-benefit assessment is today’s effort. It can be roughly determined with three values: **Time per case** × **cases per year** × **internal hourly rate** = current process costs per year A worked example with assumed values: a case takes 6 minutes and comes up 40 times a day, on 220 working days. That is 52,800 minutes or 880 hours a year. There is a public reference point for the hourly rate: according to the German Federal Statistical Office (Destatis), an hour worked in industry and services in Germany cost an average of €45.00 in 2025, including non-wage labour costs. Your own rate may differ considerably, depending on industry, region and type of work. With this average, the example comes to around €39,600 a year. More important than the precision of the hourly rate are time and volume. Both tend to be estimated, and are often misjudged in the process. If you want to base the decision on firmer ground, record them for two weeks, for example with the [“Measure First” tracking sheet](https://riccardonentwich.com/en/insights/measure-first-tracking-sheet/). ## What realistically goes away Today’s process costs are not the savings. The picture only becomes realistic once standard cases and exceptions are separated: **Standard cases per year** × **minutes saved per standard case** ÷ 60 = hours saved per year In the example: assume 70 per cent of cases are standard cases, and for those, 1 minute for a quick check remains instead of 6 minutes. That is 6,160 cases with 5 minutes saved each, around 513 hours or roughly €23,100 a year in total. The remaining 30 per cent are still handled by a person, with the same effort as before. The calculation shows two things. The savings are well below today’s process costs. And the share of standard cases has a big influence on the result. With 50 per cent standard cases, the same example would only come to around 367 hours. ## What the solution costs, including after go-live The second side of the calculation is often underestimated, because it doesn’t end with implementation. - **Build:** analysis, implementation, connecting the systems, testing with real cases, introducing it to employees. - **Running costs:** licences, hosting and, where applicable, usage-based costs, for example for AI services. - **Maintenance:** adjustments when forms, interfaces or rules change. - **Exceptions and checks:** the time employees still need for checks and special cases. This is already taken into account in the savings above and must not disappear twice. This gives a simple metric: **One-off effort** ÷ (**annual savings** − **annual running costs**) = years to payback If the payback period is well below the time the process is likely to remain stable, the calculation speaks in favour of automation. If it is of the same order of magnitude, there is a high risk that systems or rules will change before the investment has paid for itself. ## What cannot be expressed in hours Some effects are real but hard to put in euros. They still belong in the decision, as separate points alongside the calculation. - **Consequences of errors.** Anyone who transfers data by hand makes mistakes, and checking by eye is less reliable than people assume. In an experiment by Barchard and Pace (2011) with 195 students, visually checking entered data was not significantly better than entering it without any check at all. The study took place in a laboratory, not in a company. But it shows that the transfer itself is the source of errors. Where a wrong value has costly consequences, such as a wrong delivery date in planning, that counts for more than the minute saved. - **Response time.** A quote that reaches the customer a day earlier, or an enquiry that doesn’t sit untouched over the weekend, affects revenue, not costs. - **Capacity.** In most businesses, hours saved don’t mean a job saved, but time for work that is currently left undone. That is a more honest calculation, and usually a more accurate one. - **Cover.** A process that only one person has mastered is a risk. Clear rules and a traceable solution reduce it. Decision ## Four possible results The assessment doesn’t always end with “automate”. Four results are equally valid, because each describes a sensible decision: Possible results - Leave it The process is too rare, too inconsistent or too unstable. The effort for a solution is out of all proportion to the benefit. - Simplify Steps are dropped, responsibilities become clear, forms become consistent. Often that alone achieves a lot. - Use what is there The software already in use can take over the step if it is set up accordingly. - Automate Frequency, rules, data and exceptions fit, and the calculation also covers the costs after go-live. ## When you are better off leaving it Some situations almost always speak against automation, at least for now: - The process comes up rarely and is different every time. - The rules are disputed or not settled. Automation would only hide an open conflict. - A system change is coming up. The solution would have to be rebuilt shortly afterwards. - The real problem lies elsewhere, for example in missing information from the customer. Then a better enquiry helps more than faster processing. ## The next step For the decision, two things are usually enough: a clear description of the process and a few measured values. There are templates for both: the [Process Profile](https://riccardonentwich.com/en/insights/process-profile/), for recording the process on one page, and the [“Measure First” tracking sheet](https://riccardonentwich.com/en/insights/measure-first-tracking-sheet/), for recording frequency, time, queries and exceptions over two weeks. Together, they let you base the calculation in this article on real figures. ## Sources 1. [Michael Hammer: Reengineering Work: Don’t Automate, Obliterate. Harvard Business Review, July/August 1990.](https://hbr.org/1990/07/reengineering-work-dont-automate-obliterate) Basis for the advice not to transfer existing processes into software unchanged, but to simplify them first. 2. [Raja Parasuraman, Thomas B. Sheridan, Christopher D. Wickens: A model for types and levels of human interaction with automation. IEEE Transactions on Systems, Man, and Cybernetics – Part A, 30(3), 2000, pp. 286–297.](https://doi.org/10.1109/3468.844354) Model according to which automation can take over individual functions (acquiring information, analysing it, deciding, acting) to different degrees. Developed in research on complex technical systems. 3. [Kimberly A. Barchard, Larry A. Pace: Preventing human error: The impact of data entry methods on data accuracy and statistical results. Computers in Human Behavior, 27(5), 2011, pp. 1834–1839.](https://doi.org/10.1016/j.chb.2011.04.004) Experiment with 195 students on the accuracy of data entry: visual checking was not significantly better than simple single entry. Laboratory setting, not a business context. 4. [Statistisches Bundesamt (Federal Statistical Office of Germany): „Eine Arbeitsstunde kostete im Jahr 2025 durchschnittlich 45,00 Euro“ \[An hour of work cost an average of €45.00 in 2025\]. Press release no. 148 of 29 April 2026 (in German).](https://www.destatis.de/DE/Presse/Pressemitteilungen/2026/04/PD26_148_624.html) Average labour costs per hour worked in industry and services in Germany, used as a guide for the hourly rate in the worked example. Would you like to look at a specific process together? [Arrange an initial call](https://riccardonentwich.com/en/initial-call/) ## Read next Tool ### [Measure first: a tracking sheet for manual processes](https://riccardonentwich.com/en/insights/measure-first-tracking-sheet/) How often, how long, how many queries and exceptions? A tracking sheet for recording a manual process for two weeks and then evaluating it simply. Tracking sheet with evaluationPDF 10 September 2026 Insights ### [Which process should you automate first?](https://riccardonentwich.com/en/insights/which-process-to-automate-first/) Several candidates, one first project: prioritising with six criteria, a simple comparison grid and the most common mistakes when choosing. 6 min read 20 August 2026 Tool ### [Process Profile: capture a business process on one page](https://riccardonentwich.com/en/insights/process-profile/) Trigger, result, steps, friction points, volume, exceptions and decisions: one process on one page. For printing, with instructions and examples. Fill-in templatePDF 21 September 2026