The first process you automate in a ten-person company is usually not the most annoying one. It is the one that repeats often, has clear rules and goes through the hands of a single person. If the process changes every month or needs case-by-case judgment, you leave it for later.
In a ten-person company you do not have an operations department. You have two or three people who, on top of their actual job, do the company's admin work: enter invoices, answer the same questions, move data from one place to another, put together a report on Monday morning. This work appears nowhere in the job description, and that is exactly why nobody sees it until the person goes on holiday.
When you start automating, the temptation is to begin with whatever annoys you most. That is the worst selection rule there is. What annoys you is often rare, complicated and full of exceptions. Here is the rule we use for triage.
The triage rule: frequency, clear rules, one owner
A process deserves to be automated first if it ticks all three: it happens at least weekly, it can be written down as rules a new hire would understand, and one person is responsible for it. If any of the three is missing, move to the next one on the list.
Frequency gives you the saving. A process that takes twenty minutes and happens daily eats more than eighty hours a year. The same process done three times a year justifies nothing, however unpleasant it is.
Clear rules give you an automation that does not break. The test is simple: ask the person who does the process to explain it to you as if they were training a new employee, in five minutes. If the explanation is full of "it depends" and "we'll see then", you do not have a process yet, you have a habit.
A single owner means you have someone who can tell you how the process really works, as opposed to how the boss thinks it works. Processes split across three departments can be automated, but not first.
What is not on the list: how modern it sounds. Nobody pays for "we have AI". People pay for hours that are no longer spent and mistakes that no longer happen.
See in 1 minute if it can be automated is a short questionnaire that applies exactly this triage to your workflow, without talking to anyone.
Three process types that come out first
These are not case studies and they are not clients. They are process patterns that show up in almost every small company in Romania and that pass the triage above more often than others.
1. The document that comes in and has to be transcribed somewhere
Supplier invoices, orders by email, delivery notes, receipts from the field team. They all have the same shape: a document arrives, someone looks at it, types five or ten fields into another system, then archives the file.
It passes the triage because it is daily, because the rules are the same every time, and because one person usually does it. It is also the kind of process where a typing mistake costs money directly, which means the saving is more than time.
2. The repeated answer on the same channel
Quote requests that ask for the same three pieces of information. Status questions from customers. Document requests from accounting. Anyone who has answered the same question twenty times in a month knows what this is about.
Here automation rarely means "a robot answers instead of me". It usually means the answer comes prepared, with the data already pulled from your systems, and the person presses send. The difference between five minutes and thirty seconds, twenty times a day.
3. The Monday morning report built from three sources
Someone opens the billing account, exports a file, opens the sales platform, exports another file, pastes them into Excel and sends a summary by email. It takes an hour and a half and happens every week.
It passes the triage with top marks: frequent, fully deterministic, one owner. It is also the easiest to prove it works, because the result can be compared directly with the report made by hand.
When NOT to automate
This part is missing from most material on the topic, because nobody sells their service by telling you when not to buy it. We would rather not build something that gets thrown away in three months.
When the process changes every month. If the rules get rewritten every cycle, the automation becomes a second job: you fix it more often than you use it. The right answer here is to wait until the process settles.
When every case needs judgment. Negotiating a price, deciding whether to take on a client, assessing a nuanced complaint. You can prepare the ground for the person, but the decision does not move to the machine, and whoever promises you otherwise is selling you a future problem.
When it happens a few times a year. It does not matter how painful it is. The math does not work out, and it is fine for it to stay manual.
When nobody can explain the rules. If two people in the company do the process differently and both are right, you do not have a process. You clarify it on paper first. That is an hour of discussion, not a project.
When you want to automate three things at once. In a ten-person company, three processes started in parallel usually means three unfinished things. One, taken all the way to production, then the next.
What the first step looks like with us
We start with a structured Technical Diagnostic (EUR 490): a full workflow analysis delivered in 5 business days. We evaluate your current systems and tell you exactly what can be automated in two-week engineering sprints and what is not worth touching.
If you want more than an opinion, the AI Diagnostic costs 490 EUR, takes five working days and is deducted in full from a build started within thirty days. From there, an end-to-end workflow costs between 700 and 1,500 EUR, and a production agent starts at 3,500 EUR. Prices are final: DeviDevs is not registered for VAT (art. 310 of the Romanian Fiscal Code), so no VAT is added on top.
The full details are on the pricing page, and the logic behind the numbers is explained in the automation pricing guide.
Frequently asked questions
What do I automate first in a small company?
The process that repeats most often, has clear rules and goes through the hands of a single person. Pick the most predictable one, which is often neither the most annoying nor the most visible. Frequency gives you the saving, clear rules give you an automation that does not break, and one responsible person means you have someone who can tell you how the process really works.
How many processes do I automate in the first wave?
One. One process taken end to end, in production, with error handling and with the people who use it every day. Three processes started in parallel usually means three unfinished things. After the first one runs for a month without intervention, you have both the confidence and the data to pick the second.
When is a process not worth automating?
When the process changes every month, when every case needs human judgment, when it happens a few times a year, or when nobody can explain the rules. If two people in the company do it differently and both are right, the process is not a process yet. You clarify it on paper first, and only then automate it.
Next step
If you already have in mind the process you thought about three times while reading this, that is the candidate.
Book a Technical Diagnostic. Architecture review, complete workflow mapping, and a direct engineering roadmap.