You've heard "AI saves time" often enough now. The more exciting question is a different one: At what specific point in your business is automation already paying off this quarter, and not just in a strategy paper for 2028? That's exactly what we're talking about here. Not about the big transformation, but about the two or three processes that cost time today and won't from next month.
You don't need a data science team or a six-figure budget for this. What you do need is a clear view of what kind of work pays off if you automate it and what doesn't (yet). We'll give you that in this article. If you're looking for specific tools, it's worth taking a look at n8n as an automation tool; here we deliberately stay one level above, with the question "where is it even worth it?".
AI automation or workflow automation: what you really need
Before we talk about money, a clarification that will save you a lot of frustration later. Not all useful automation is "AI". A large part of what eats up time in SMEs are clear if-then processes: an order comes in, a data record moves from A to B, a reminder is sent. This is classic workflow automation, rule-based, predictable and therefore cost-effective and stable. You don't need AI for this, and you shouldn't use AI where a simple rule is enough.
AI comes into play precisely when the input is unstructured and a human previously had to "take a quick look and categorize" it. An invoice as a PDF from which the supplier, amount and service must be extracted. A customer email that should be sorted into "complaint", "request for quotation" or "invoice enquiry". A photo of the incoming goods that is compared with the order. At these points, an AI workflow replaces what previously required judgment and experience. This is where the greater leverage arises, because it is precisely these "short categorization" tasks that are difficult to squeeze into rigid rules.
The honest rule of thumb: first ask whether a rule is enough. If so, use the rule. However, if the step requires an understanding of language, image or context, this is the point at which AI automation pays off instead of just being more expensive.
Four places where process automation pays off immediately
If you want to automate business processes and the return should be visible quickly, it's worth sorting by lever type rather than by department. Four patterns almost always pay off the fastest in SMEs
Firstly: high volumes, small steps Everything that happens hundreds of times a month and only costs a few minutes per process. Entering incoming invoices, transferring orders to the right system, maintaining master data. The individual tasks seem too small to be taken seriously, and that's exactly what makes them expensive, because nobody ever tackles them specifically. Five minutes times four hundred a month equals over four full working days. This is where the return is easiest to calculate.
Secondly: unstructured input that someone has to sort out. The classic AI lever. Routing emails to the right mailbox, reading PDFs, pre-qualifying enquiries. The time saved is real, but the real value is often the uniformity: the machine sorts the 200th email at 5 pm just as carefully as the first one at 8 am.
Thirdly: expensive omissions. Some processes hardly take any time at all, but once they slip through, it gets expensive. A missed cancellation deadline extends an annual contract that nobody wanted for a long time by another twelve months. A stock level that runs to zero unnoticed, in the week with the highest demand of all weeks, and every order that you cannot deliver now, the customer places with a competitor. Here, automation doesn't pay off in terms of hours saved, but in terms of a single loss avoided per year, which often exceeds the tool costs many times over.
Fourthly: response time that determines turnover. Every minute counts when it comes to sales enquiries. The much-quoted lead response study by James Oldroyd (MIT, data up to 2007; the Harvard Business Review summarized it in 2011 under "The Short Life of Online Sales Leads") shows: The chance of qualifying an online lead is around 21 times higher if the initial contact is made within five minutes rather than 30 minutes. Automation that forwards an enquiry to the right person in seconds does not sell itself. But it ensures that the person can still call in the right time window.
How to calculate the return in five minutes
You don't need to commission an ROI study to know whether a process is worthwhile. A beer mat calculation is enough for the initial decision. You need four numbers, and you already know three of them.
Take a specific process. Estimate how many minutes a process takes manually and how often it occurs per month. Multiply this by an honest internal hourly rate to calculate your monthly expenditure. Compare this with two items: the one-off setup costs and the ongoing tool costs per month.
A conservative example calculation for invoice entry, purely as a model: 5 minutes per document, 400 documents per month, that's a good 33 hours. With an internal rate of 45 euros per hour (model assumption), the manual effort is around 1,500 euros per month. This compares to perhaps two days of setup and around 60 to 70 euros a month in running costs for infrastructure and readout service. Even if the automation only processes 70 to 80 per cent of the documents cleanly and the rest continues to be processed manually, the break-even point is often reached in the first month. From month two onwards, the savings are full.
It's important to be honest about two things. The hourly rate must be real, not embellished. And the set-up costs must be fully disclosed, including the internal time required for set-up. If you calculate both properly, you don't need to do any fancy maths: the cases that are worthwhile are also clearly worthwhile when calculated conservatively.
What you start with - the first candidate counts the most
The most common mistake is not the wrong tool, but the wrong first process. If you start with the most complex, politically sensitive process, you will burn out for weeks, fight resistance and end up with no sense of achievement to sustain the next automation. Consciously look for something else to start with.
The ideal first candidate fulfils three things: it costs a noticeable amount of time today, it follows a clear and stable process, and it belongs to a person who is on board. This last condition is underestimated. Automation whose process changes every month or over which two departments fight is a bad start, no matter how much it theoretically saves. Go for the quiet, unloved routine job that nobody argues about.
And plan the second step at the same time. As soon as the first automation is up and running, the infrastructure is in place and the next processes almost fall into place by themselves. Your team will suddenly take a different look at the week and discover little time wasters everywhere that previously passed as "just the way it is".
Where AI automation is not (yet) the right lever
In order for the decision to bear fruit, the limits also need to be put on the table. AI automation is not an end in itself. In some places, it is simply the wrong tool.
For processes that occur infrequently or constantly change form, setting them up takes more time than the automation ever gives back. Leave it alone. For decisions with real scope for discretion (personnel issues, individual goodwill, anything with legal implications), the machine should at most prepare, never make the final decision. And in highly regulated areas, where every step must be fully verifiable, standard on-board resources are often not sufficient for complete audit trails. This is not a reason not to do it, but a reason to set it up properly from the outset.
The most honest sentence at the end of this section: AI automates a bad process just as reliably as a good one, only faster. If you don't understand the underlying process, you are likely to make the problem worse. That's why in every good project, understanding the process comes before automating it.
Conclusion: start small, calculate honestly, stay tuned
AI automation in SMEs is not a question of a grand vision, but of the right sequence. Sort your processes according to whether someone needs to "read and categorize" something; these are your AI candidates. Prioritize according to the four levers: high volume, unstructured input, costly omissions, critical response time. Calculate the first case conservatively and look for a first candidate that nobody will argue about. If that works, the rest will almost take care of itself.
And that's the real point: it's not about the big strategy paper for 2028, but about the one process that will take less time next month. That's where you start. The best next step costs half an hour: Take the process that cost you the most time this week and do a clean calculation. If you would like an external perspective, we at nextlevels can help you with AI consulting and process automation and specifically with the automation of AI workflows. The article on five SME projects with ROI in 12 months is the perfect neighbouring read for strategic placement in the first digitalisation projects.