In the Netherlands, 30% of businesses now choose automation over new hiring when facing talent shortages, up 5 points in a single year, according to Statistics Netherlands (CBS, June 2026). For the first time, automation ranked as the most popular response, ahead of salary increases and flexible contracts. Across the EU, that same pressure is intensifying: the European Commission estimates a shortage of over 500,000 ICT professionals across member states, and AI-related roles take an average of 5.2 months to fill in Europe according to IDC research. Five months of searching, and then onboarding begins. The question forces a concrete decision right now: open a job posting, or deploy a tool? Five questions help you make it clearly.
Why this question is more urgent now than a year ago
AI now completes entire workflows without supervision, not just assists with individual steps. Tools could summarize emails and draft copy a year ago. The shift since then is different in kind: Microsoft Copilot Cowork runs analyses while your laptop is closed. Claude agents run nightly reports. Salesforce Agentforce prioritizes invoices and drafts follow-up emails autonomously.
That changes the nature of the decision. It's no longer "can AI help my employee do this better?" The question is now "can AI do the work I'd otherwise hire someone to do?"
Two-thirds of European businesses report difficulty filling critical roles, consistent with Eurostat's 2025 vacancy data. The pipeline for AI-competent staff is especially slow. Recruitment costs and the productivity gap during that waiting period add up fast.
What does that new employee really cost?
The true annual cost of a mid-level knowledge worker runs €55,000-€75,000 in most European markets once you include employer contributions, workspace, training, and recruitment fees. Most managers calculate from the gross salary line. That's the wrong starting point.
Take a role at the EU average knowledge worker salary, roughly €36,000-€45,000 gross per year depending on country. Add employer social contributions (typically 25-40% on top of gross), and you're at €47,000-€63,000. Add workspace, hardware, training, and management overhead, and the real employment cost lands between €55,000 and €75,000 per year. Recruitment fees, averaging €3,000-€8,000 per role for SMBs, aren't in that number either.
What does an AI tool cost? A ChatGPT Team subscription is $30 per user per month, or $360 per year. Claude Pro is $20 per month. A robust professional AI setup, Claude Max at $200 per month plus a Cursor Business license at $40 per month, comes to roughly $2,880 per year.
For the price of one new hire, you can equip your entire team of ten with professional AI tools and still have budget left over.
Cost alone doesn't settle the question, though. The real issue is what that employee would do, and whether a tool can actually do it.
Can an AI tool handle this specific work?
Not all work is equally suited to AI. A simple framework helps you assess whether a task can move to a tool or needs to stay with a person.
AI performs well for:
- Structured, repeatable tasks (email triage, report generation, invoice processing)
- Text-based work with clear inputs and outputs (summaries, translations, proposal drafts)
- Data analysis and pattern recognition (customer segmentation, trend analysis, anomaly detection)
- Tasks where speed matters more than nuance (first-pass candidate screening, FAQ responses)
AI falls short for:
- Relationship-based work (client conversations, networking, conflict resolution)
- Contextual judgment (a difficult complaint that requires bespoke handling)
- Physical work
- Strategic decisions requiring deep organizational knowledge
Think of it as the difference between an assistant and an advisor. AI is an excellent assistant: fast, tireless, always available. It's a mediocre advisor. It doesn't know your clients, can't read the mood in a room, and doesn't know when an email is better left until tomorrow.
That line is moving fast, though. Tasks that only a person could handle a year ago, like analyzing an entire contract or drafting a full project proposal, are now within reach of current AI tools. If you want a systematic framework for deciding what to delegate, see the AI task decision model published separately.
Five questions to ask before you decide
Run every hiring consideration through these five questions. They act as a filter: after question five, you'll know whether you need a person, a tool, or a combination of both.
1. Is the work structured and repeatable?
If you can describe it as "every week, do step A, then step B, then step C," it's a candidate for AI. The more predictable the process and the more often it recurs, the stronger the case for automation. A monthly report that always processes the same source data? AI. A sales call with a new enterprise prospect? Person.
2. What's the acceptable error rate?
AI makes mistakes. Sometimes subtle, sometimes significant. For an internal weekly digest, a 10% error rate is manageable. For legal advice or a medical assessment, it's unacceptable. Ask yourself: if the output is 10% wrong, what does that cost your business? If the answer is "a few minutes of correction," AI is fine. If the answer is "a client complaint" or "a liability claim," you want a human in the loop.
3. Does the work require a relationship?
Some work exists because of the specific person doing it. An account manager who has worked with the same client for five years. A project lead who knows exactly how the executive team thinks. A colleague who senses when someone is stuck. Those aren't tasks, they're relationships. AI can't provide them. If your answer is yes, hire.
4. How quickly does the output need to improve over time?
An AI tool delivers consistent output from day one. A new employee needs three to six months to reach full speed, but improves every year, takes on more responsibility, and grows with the business. AI also improves, but only if you actively update the prompts and workflows. Need someone who hits the ground running right now? AI. Need someone who's your right hand in two years? Person.
5. Can you verify the output?
This is the question most teams skip. AI output needs review by someone who understands the domain. Domain expertise matters: a marketer reviewing AI-generated copy produces better output than a junior staffer who approves everything without reading carefully. If no one on your team can evaluate what the AI produces, the tool is worthless, regardless of its price.
When do you always choose a person?
Three situations make AI a non-option, regardless of cost.
You're building a team, not a production line. A growing business needs people who think independently, take initiative, and shape the culture. AI doesn't contribute to the Monday morning brainstorm. It won't say "wait, this plan doesn't hold up" when everyone else is excited about a flawed idea.
Your clients expect a person. In healthcare, legal services, and financial advice, personal contact is a core part of the service. An AI chatbot handling initial triage is useful. Your clients still expect a person for the substance of the conversation.
The role requires physical presence. Technicians, healthcare workers, construction crews, hospitality staff: physical work is not replaceable with a language model. Industrial robotics is a separate conversation, but that's not the AI this article covers.
What if you combine both?
The pattern emerging across high-performing teams isn't "person or AI" but "fewer people, higher output per person, every team member working with AI tools."
Eugenia Kuyda, founder of chatbot platform Replika, described her hiring process recently in Platformer. She screened 120 candidates to hire one exceptional Swift developer.
"I'm not hiring people anymore for these junior jobs."
Eugenia Kuyda, founder of Replika, in Platformer
Her model: ten to fifteen senior people, each working with AI tools. Enough, she argues, to build a billion-dollar business. That's a Silicon Valley extreme. But the underlying logic applies beyond tech startups.
Instead of hiring three mid-level staff for administrative work, you hire one senior person and give them Claude, Copilot, and automated invoice processing. The one person costs more per hour, but the total employment cost is lower and the output is higher. Research from S&P Global indicates that employees working with AI tools save an average of 5.6 hours per week on repetitive tasks. That's nearly a full workday per person. Multiply across a team of five, and you have the productive capacity of six people at the cost of five.
According to our AI Workforce Statistics, that shift is already accelerating across European industries. For context: the productivity gain only sticks if you deliberately redirect freed hours to higher-value work. Left unmanaged, they disappear into slightly longer meeting prep and admin. The explanation for why this happens, and how to prevent it, is in why AI-saved hours tend to evaporate.
The hybrid model works when you're intentional about where the freed time goes: client contact, quality control, strategy. The repetitive work goes to AI. The thinking stays with people.
What can you do with this before Friday?
Three concrete steps that take under two hours total.
Step 1: review your open roles. For each vacancy, write down the five most important tasks the new hire would handle. Run each task through the five questions above. You'll quickly see which vacancies allow a hybrid approach and which genuinely require a person.
Step 2: calculate the real cost. Take the true annual employment cost (gross salary plus employer contributions, plus workspace, plus recruitment fees) and compare it against equipping your existing team with AI subscriptions. The gap is almost always larger than managers expect.
Step 3: run a two-week pilot. Pick one open vacancy and hand the underlying work to an AI tool for two weeks, with a team member reviewing the output. After two weeks, you'll know whether the tool can handle it based on real output in your actual business context, not a vendor demo. Give the tool enough context about your business before you start, as covered in the guide on giving AI the right company context. Otherwise you're comparing apples to oranges.
The choice between hiring and automation isn't ideological. Some organizations search too long for scarce talent while a tool could already handle the work. Others deploy AI for tasks that only a person can do well. Both mistakes cost money. The five questions help you land on the right side of that line.
