What a New Hire Actually Costs
A client called me on a Friday afternoon, maybe two weeks before he was going on holiday. He runs a B2B ingredients distribution company, around 65 people, clients are food manufacturers and catering businesses across Central Europe. He said he needed someone for customer operations because the team was drowning. Had a candidate lined up already, a referral from his business partner, wanted me to "just take a look" at the job description.
I asked what this person would do for the first two hours of every working day. He started listing. Order confirmations. Delivery status updates pulled from their ERP. Replies to the same client questions, over and over. He slightly stumbled at the fourth point and said "you know, mostly inbox management."
Right.
The full cost of a hire is not the number in the job posting. Gross salary, employer social contributions, equipment, software licences, the time a manager spends onboarding someone and answering their questions for the first two months. In Western Europe that adds up to somewhere between 60,000 and 90,000 EUR per year for a junior to mid-level role. And that assumes the person stays, because turnover in operational roles is real.
I am not writing this to scare anyone off hiring. Sometimes you just need to hire and that is the end of it. I am writing it because most of the business owners I have talked to never compared these numbers against the cost of automating the tasks the new person was going to do.
What AI Is Good At (and what I think about the AI market)
One thing before I go further: most of what is being sold as "AI implementation" right now is Make.com flows dressed up as a revolution. I understand why the market looks like this, it is hard to evaluate quality from the outside and demand is high. Worth keeping in mind when you search for a vendor.
What AI actually does well, and what I build day to day: email triage and replies connected via API to Gmail or Outlook with a classifier running on n8n, data extraction from PDF invoices and purchase orders into ERP or CRM, order status updates triggered by webhook when something changes in the order system, reports generated overnight and sent in the morning, payment reminders personalised per client. Boring things. Repetitive things. That is precisely why AI handles them well, not because it is smart but because they have consistent inputs and defined outputs.
Where AI does not belong. I watched a client try to automate complaint handling for their top accounts, the ones spending several hundred thousand a year. I told him not to do it. A client relationship at that level is not a process you can put in a config file. It is a specific person on the other end who knows that when they call, someone will pick up and already know the context. No chatbot replaces that. Does not matter how good it is.
Same applies to negotiations, team management, anything where the right answer depends on reading the room. Automation does not read rooms.
What the Numbers Look Like
I did this calculation properly with that distribution client in October, over coffee at their office. The scope for the planned hire: inbox management, order status updates, weekly client reports. That was it.
A hire: around 38,000 EUR gross salary per year, so the employer spends somewhere north of 46,000 EUR annually. Add onboarding, equipment, a few weeks of the operations manager's time: call it 50,000 EUR for year one.
Automating those same three areas: implementation cost was 5,800 EUR, infrastructure runs around 200 EUR per month. Year one comes out to roughly 8,200 EUR. Year two is just infrastructure.
The client said it sounded too good to be true. I understand that reaction. A month after deployment he wrote to say he wished he had done it a year earlier. I am honestly not sure we would have been technically ready a year earlier, but I understand the sentiment.
Three Questions I Ask Instead of Running an Audit
I stopped doing formal pre-project audits because they took too long and gave clients the impression something complicated was happening before we had even started. Now I just ask three things in a conversation.
First: what will this person do for the first two hours of every working day? If the answer is basically a list of the same things, there is a chance it can be automated. If the answer is "it depends what comes in," I dig into what specifically changes and how often. Usually "it depends" turns out to mean one of five scenarios, and all five can be described.
Second: could someone in your company describe this process on one page of A4? Not a document. One page. If yes, automation is technically feasible. If the description takes ten pages and still has exceptions, that is a sign you need a human. Not because AI cannot handle exceptions, but because the cost of describing and handling every exception stops making economic sense.
Third: what would happen if this work was done identically every time, with no initiative, no deviation from the pattern? Companies often pause on this one, because it turns out they actually want someone who will notice things that are not in the process. That is a completely different role and a completely different conversation.
The ISP Case: Two Headcounts That Were Not Needed
Regional internet service provider, around 80 people, mix of business and residential clients across several regions. Growing ticket volume, two people in the support team at capacity for three. They wanted to hire two additional coordinators.
Before we signed anything I spent a week going through their tickets. Their system was Freshdesk, fairly basic setup, no tags or categories, everything landed in one queue and the team sorted manually. Roughly two thirds of all tickets were six types: connection down, speed issue, billing question, address or account change, service upgrade request, planned maintenance notification. Each had a clear handling path. None required a decision.
We built a classifier on n8n connected to Freshdesk via API. It reads the ticket, assigns the type, generates a response pulling live data from their billing system, and either closes the ticket or escalates with a short context note. The first two weeks were rough. The model kept misclassifying tickets written in shorthand, missing punctuation, sometimes half in English and half not. One customer wrote "internet down again pls fix urgent its third time this week" with no other context and the classifier got it wrong four times in a row. We had to retrain on their actual ticket history, about two thousand examples from the previous six months. That took an extra week and a half that I had not planned for.
After that: roughly two thirds of tickets handled automatically, usually within a few minutes of arriving. The rest goes to coordinators with a context note already written. Response time on straightforward issues went from a few hours to a few minutes. Client satisfaction scores improved. The company did not hire those two people.
Total project cost: well under 7,000 EUR. Two hires for a year would have been well above 90,000 EUR. I do not know a business that would not appreciate that difference.
One thing I always say and often forget to write: not every company is ready for this. Some projects I walk away from after the initial conversation because hiring genuinely makes more sense. That is not a problem for me. Better to say it at the start.