In-house vs agency, in numbers

TL;DR — Three median-salary engineers in Germany cost about €939,000 over three years once you add employer contributions, recruiter fees, and one replacement hire. That's roughly €504 per available person-day at steady state, which is cheaper per day than most Western European agency rates. In-house wins on raw price. It loses on everything else: you wait somewhere between six months and over a year for the first fully productive hire, and the €26,000 monthly bill arrives whether or not there's work that month. The decision comes down to one number, and it isn't the salary.

---

Who this is for, and what I assumed

You run a small or mid-sized company in Western Europe. You need a product built and kept running. Someone has told you to hire a frontend developer, a backend developer, and a DevOps engineer, and someone else has told you to just use an agency. You want to know which one costs more.

Every figure below traces to a published source or to arithmetic on one. Nothing is a vibe. The specific assumptions:

• Market: Germany, as the Western European anchor. It has the largest sample in the salary data and sits mid-range between France and the UK.
• Seniority: median, all experience levels. Not staff engineers.
• Currency: euros, converted at the ECB reference rate of 1.1460 USD per EUR (18 September 2026).
• Excluded: laptops, software licences, office space, training budget, management time, payroll admin, and employer's liability insurance. All of these make in-house more expensive, so the model below is generous to the in-house case on purpose.

If your market is more expensive than Germany's, scale the numbers up. Eurostat put average hourly labour cost in 2025 at €38.2 across the euro area and €47.9 in the Netherlands, so a Dutch version of this model runs meaningfully higher.

What three engineers cost before anyone writes code

Start with gross salary. The 2025 Stack Overflow Developer Survey reports median yearly pay by country and role, with 2,141 German respondents:

Gross salary is not what leaves your bank account. Eurostat puts non-wage costs — employer social contributions and employment taxes — at 25.6% of total labour cost in the euro area for 2025. Working backwards, that's a multiplier of about 1.344 on gross pay.

That's €25,926 a month in fixed cost. France would be worse — Eurostat puts French non-wage costs at 32.3%, the highest in the EU.

Then there's finding them. Recruitment agencies in Germany charge 15% to 30% of first-year salary, per Hiring Hub's European fee survey. At 20%, three placements cost €46,293.

Year one, all in: €357,403.

What you get for it: 206 days, not 365

A salaried engineer is not available 365 days a year, or even 261. Here's the subtraction for Germany:

Three engineers give you 617.7 available person-days a year, or about 51 a month. And "available" is doing some work in that sentence — it counts every hour spent in standups, retros, performance reviews, and onboarding the next hire.

The number that actually decides it

Divide loaded cost by available days:

• Year one, including recruiter fees: €579 per person-day
• Steady state, year two onward: €504 per person-day

Now compare that to what the market charges for the same day:

One caveat on that table: these are rates for individual freelancers, not agencies. Published agency rate cards are scarce, and a blended agency day rate sits above the freelance equivalent — it carries sales, project management, bench time, and cover when someone is ill. So treat €562–€800 as the floor of what an agency will quote, not the middle.

So the honest finding, and it's not the one an agency usually leads with: per day of work delivered, in-house is cheaper. A €504 internal day rate undercuts every observed market rate. If you can keep three engineers genuinely occupied, hiring is the better deal, and any agency telling you otherwise is selling.

The word doing the work is if. Here's the same comparison run backwards — how busy you'd have to keep them:

At a €700 blended day rate, in-house wins as long as all three people are productively booked more than 72% of their available days. Below that, you're paying for idle capacity.

For most SMBs the frontend and backend roles clear that bar comfortably. The DevOps role frequently doesn't. Once the pipeline is built and the infrastructure is stable, a small company's genuine DevOps demand is lumpy — heavy during a migration, near zero for the quarter after. I don't have data on what fraction of a year a typical SMB genuinely needs a dedicated DevOps engineer — but you can estimate it for your own company, and if the answer is under 72%, you're paying €115,657 a year for capacity you won't use, every month, regardless.

That's the real shape of this decision. Not "is the agency's rate higher" — it is — but "will I use what I'm buying."

The three-year picture

Extending the model over 36 months, with two adjustments: Eurostat's Q1 2026 figure of +3.2% annual labour cost growth in the euro area, and one replacement hire, because Ravio's 2026 European compensation data puts average engineering tenure at 2 years 11 months. Over three years, expect at least one departure.

At a €700 agency day rate, the same €938,799 buys 1,341 person-days — about 37 a month. The in-house team delivers 1,684 person-days across the same 36 months, once you subtract the two and a half months spent searching and the vacancy left by the replacement hire. That's 47 a month, and about 26% more capacity per euro.

Keep that number in mind, because the rest of this article is about why companies still choose the agency.

The lag nobody puts in the spreadsheet

Ashby's recruiting benchmarks, drawn from 54 million applications across 93,000 jobs between January 2021 and March 2026, put the median time to fill a technical role at 75 days. Senior roles run 37% longer than junior ones.

That's per role. You need three.

If each search independently has a 50% chance of closing by day 75, the probability that all three close by day 75 is 0.5³ — 12.5%. Searches aren't perfectly independent and a good recruiter improves the odds, but the direction is unavoidable: a team is complete when the slowest hire lands, not the median one. Plan around the median of a single search and you will be wrong roughly seven times out of eight.

Then add notice periods. Statutory minimums in Germany, the Netherlands, and France start at around a month and rise with tenure, and employment contracts for experienced people routinely specify longer. Check the actual figure for your country and the seniority you're hiring — whatever it is, the model above ignores it entirely.

Then add ramp-up. A Swimm survey of 80-plus engineers and engineering managers found 3 to 9 months to full productivity, with a meaningful share of companies reporting a full year. It's a small, older sample and I'd treat the range as indicative rather than precise — but nobody who has onboarded an engineer thinks the honest number is two weeks.

Stack it up:

2.5 months to find them (sourced) + 1 to 3 months' notice (assumed) + 3 to 9 months to ramp (sourced) = roughly 6.5 to 14.5 months before your first hire is running at full speed.

You are paying full loaded salary for most of that. And the trio has to learn your codebase, your domain, and each other simultaneously — a frontend dev and a backend dev who have never worked together spend their first months negotiating an API contract that an existing team would have settled in an afternoon.

An agency team doesn't eliminate ramp-up. It has to learn your domain like anyone else. What it skips is the search, the notice period, and the part where three strangers figure out how to work together, because that part already happened on someone else's budget.

The team doesn't stay

Average engineering tenure in European tech is 2 years 11 months, per Ravio. Netherlands attrition ran 14% in 2025. Those aren't alarming numbers — they're normal — but they have a specific consequence for a three-person team: every departure removes a third of your capacity and all of one specialism.

When your only DevOps engineer resigns, you have no deployment expertise for the 75 days of the search plus the notice period plus the ramp. That's not a cost line, it's a single point of failure. An agency of any size absorbs the same event by reassigning someone on Monday.

This is the argument that actually carries the decision for most SMBs, and it has nothing to do with day rates.

When in-house genuinely wins

I'd hire, not contract, if any of these are true:

• The software is the business. If your product is the code, the knowledge belongs inside the company. Any cost model that ignores strategic ownership is answering the wrong question.
• You can keep them above 72% utilisation. Look honestly at your backlog. Is there 12 months of work for all three, or three months of work and nine months of maintenance?
• You have someone to manage them. Three engineers with no technical lead is not a team, it's three contractors with worse incentives and no exit clause. If nobody in-house can review architecture decisions, you're buying supervision from the agency whether you notice or not.
• Domain knowledge is the bottleneck. In regulated or deeply idiosyncratic businesses, the learning curve is the project. That knowledge is worth keeping on payroll.
• You're funded for the lag. If you can survive 6 to 14 months before full velocity, the three-year economics are on your side.

When the agency genuinely wins

• You need output this quarter. No hiring process compresses to weeks. This is the single most common reason, and it's a legitimate one.
• Your demand is lumpy. Especially for DevOps. Buy the days you need instead of carrying the headcount.
• You can't yet write the job description. If the stack isn't settled, hiring for it locks in a decision you're not ready to make.
• One departure would stop you. Three-person teams have no redundancy. If that risk is unacceptable, don't build one.
• You want the option to stop. A contract ends. A German employment relationship, past probation, does not end quickly or cheaply.

Where this model breaks

Redo the arithmetic before you trust it, because four inputs move the answer a lot:

1. Seniority. These are medians. A senior-heavy team can cost 40–50% more, which pushes the in-house day rate above most agency rates and flips the conclusion.
2. Country. Germany sits mid-range. Stack Overflow's 2025 medians put a UK back-end developer at $108,913 against France's $71,929 — a 51% spread for the same role.
3. The excluded costs. Equipment, licences, workspace, training, and management time are all real and all omitted here. Adding them only moves the number one direction.
4. Utilisation. This is the input with the widest range and the biggest effect. Everything else is a rounding error next to the difference between 50% and 90%.

The spreadsheet is three lines: loaded cost of the team, available days, agency day rate. Compute your own break-even utilisation and answer honestly whether you'll hit it.

The short version

Hiring three engineers in Western Europe costs about €939,000 over three years and gives you 51 person-days a month once they're all in place. Per day, that's cheaper than any agency will quote you — but only if you use all of it, and only starting somewhere between month six and month fifteen.

Buying the same capability costs more per day and starts in weeks, with no search, no notice period, and no single point of failure when someone resigns.

Neither is the right answer in general. The question that resolves it isn't "which is cheaper," it's "how many days of work do I actually have, and when do I need them." Most founders who regret hiring got the first number wrong. Most who regret the agency route got the second one right and then never brought anything in-house afterwards.

---

Disclosure: I run AlgoCore, which does exactly the kind of full-stack work described above. I've tried to give the in-house case its strongest form — the model excludes real costs that would only make hiring look worse, and the per-day comparison favours hiring outright. If your utilisation clears 72%, hire. That's the honest answer, and it's the reason I published the break-even table instead of just the totals.

---

Sources

• 2025 Stack Overflow Developer Survey — Work — median salaries by country and role
• Eurostat: EU hourly labour costs ranged from €12 to €57 in 2025 — non-wage cost share of total labour cost
• Eurostat: Annual increase in labour costs at 3.2% in euro area (Q1 2026) — labour cost growth
• Ashby: Recruiting Operations Benchmarks, 2026 Talent Trends Report — time to fill, technical roles
• Hiring Hub: Recruitment Agency Fees Across Europe — recruiter fee percentages by country
• Ravio: Employee tenure trends — European tech tenure and attrition
• Destatis: Staff on sick leave — average sick days per employee, Germany
• freelancermap: The IT Freelance Market in Germany, Market Insights 2026 — DACH freelance hourly rates
• Malt: Grille des tarifs 2026, développeurs freelances — French freelance daily rates
• Swimm / HackerNoon: Engineer Onboarding — The Ugly Truth About Ramp-Up Time — ramp-up survey
• ECB euro reference exchange rate: USD — 1.1460, 18 September 2026