Outsourcing Software Development: Freelancer, Agency, or Offshore? 5 Criteria

Three quotes for the same project are sitting on your desk. One agency, two individuals. The prices are so far apart that it feels like you asked for three different projects. All you wanted was one thing: to have software developed. Maybe an internal tool, maybe an app for your customers.

What you are about to do has a clunky name: software development outsourcing. It sounds like corporate territory, but it is exactly what you do the moment you have software built externally instead of hiring. And you have three routes, not two: a single expert, a local agency, or an offshore team abroad. The most common framing is the question “freelancer or agency for software development?”, but the third route belongs on the table too.

The honest short answer hangs on five criteria, and none of them is the legal form of the provider. Three carry the decision: parallelism, discipline breadth, coordination costs. Two secure it: guaranteed availability and the question of who owns the code in the end. Once you have worked through all five, your three quotes become comparable for the first time.

One thing upfront: I work freelance myself, so you are not reading a neutral voice here. Which is exactly why you will also get the cases where I would point you to an agency. And if you google the question, take a look at the imprint of the first results. As of today, they are almost all agencies and staffing marketplaces. Every source has an angle. Mine too. I am just telling you mine upfront.

What does it cost to have software developed?

The 2026 market ranges in Germany: agencies typically charge 90 to 150 euros per hour, freelancers 75 to 120 euros, senior consultants up to 180 euros. These are not official statistics, but several independent provider guides name similar corridors (Source: milkycode / Xmethod, 2026). The Freelancer-Kompass by freelancermap, by its own account the largest survey of independent professionals in the German-speaking market, measures an average of 103 euros per hour, slightly down for the first time since the survey began (Source: freelancermap, 2026).

The third route in numbers: nearshore teams in Eastern Europe run at 40 to 100 euros per hour, offshore teams in India at 20 to 50 euros. On paper, half the price. Why the final invoice often tells a different story comes up in the criteria below. For now, just this: on small projects, the rate advantage gets eaten quickly by distance. Language, context, and time shift cost you on every round trip, entirely regardless of how good the team is.

More important than the ranges is what is inside the rate. An agency rate usually includes project management, quality assurance, design, and backup staffing. That premium buys overhead plus real services. The only question is: Does your project need those services, or are you paying for them without using them?

And then there is the hourly-rate trap. A deliberately simplified example: two quotes for the same internal tool. Quote one: 80 euros per hour, estimated 50 hours, comes to 4,000 euros. Quote two: 120 euros per hour, estimated 25 hours, comes to 3,000 euros. The second looks more expensive per hour and is cheaper as a project.

To check this yourself, ask every provider for the effort estimate per work package, not just a total. A quote without an effort breakdown can only be believed. It becomes comparable once you see the work packages behind it. And anyone who cannot explain their own hours either has not thought it through or is padding.

Experience buys hours off your bill. Compare projects, not hourly rates.

Why the playing field just shifted

The old picture went like this: freelancer equals cheap but limited, agency equals expensive but complete. That picture comes from a time when building software was almost pure manual labor and capacity equaled headcount.

What changed is AI-assisted development. Around 90 percent of developers now work with AI assistance, according to the DORA Report 2025, the big annual study on software development with close to 5,000 respondents worldwide (Source: InfoQ / DORA, 2026). For the first time, the report measures a positive relationship between AI use and both delivery speed and product quality.

But, and this belongs in the picture: the core finding of the report is uncomfortable. AI does not fix a team, it amplifies what is already there (Source: Google Cloud / DORA, 2025). In the hands of experienced people with clean processes, it works like a lever. In chaotic processes, it amplifies the chaos, and without testing and review discipline, instability measurably rises.

The consequence in one sentence: What a single experienced person can deliver today matched a small team a few years ago. Not because that person works magic. Because the lever got bigger, provided the judgment is sound.

Which five criteria decide the right model?

Forget the labels for a moment. What matters is which capabilities your project needs and how they have to be organized.

A picture that helps me here: building a house. If you build an entire house, with twelve trades that need coordinating, you hire a general contractor. That coordination margin costs you, and it earns its keep. If you renovate one bathroom, you call the specialist trade directly: shortest path, direct line, no layer in between. Both are right. It depends on the building project, nothing else.

Translated to your software project, these are the five criteria. The first three make the decision:

1. Parallelism. Do several workstreams have to run at the same time, say app, backend, and data migration against a hard deadline? Then you need several pairs of hands at once, which speaks for a team or agency. If the project runs sequentially, one person carries it. And be honest: do the streams really have to run at the same time, or do you just want everything done sooner? A hard deadline is parallelism. Impatience is not. On top of that, this line is moving. An AI-native developer orchestrates several AI agents in parallel and keeps two or three workstreams going at once, where this used to strictly require several people. The threshold where you truly need many hands sits far higher today than it did two years ago.
2. Discipline breadth. Do you need branding, UX design, development, operations, and support as a full package from one source? That is agency terrain. Check what you already have first, though: if brand and design are in place and only the build is missing, the full package sells you services you own twice. If the core is clearly defined, an internal tool, an automation, an app with one purpose, one experienced expert with a network covers it. And here too, AI-native work shifts the line: quality assurance, tests, and documentation run alongside as agent work instead of being bought in as separate roles. I described how the single developer role turns into a hybrid of five roles in “The slow death of the programmer”.
3. Coordination costs. Every intermediate layer costs money and loses meaning in translation: project management, account management, and with an offshore team also time zones and language. On small projects, coordination can become more expensive than the development itself. On large ones, it is what holds the project together. Be honest with yourself: How big is yours? This is exactly where the cheap offshore rate most often loses its lead. What you save per hour, you pay back in alignment, specification, and correction loops when the project is small and the requirements keep moving.

These two secure it, whichever way you decide:

4. Availability and backup.How bad is one week of standstill for your business? If the answer is “barely noticeable”, a single person is perfectly fine. If the answer is “existential”, you need team structures with backup staffing.
5. Code ownership and handover readiness. Who owns the code, how well is it documented, could an outside expert take over? This is the criterion almost nobody checks upfront and the one that decides everything when it matters. It gets its own section below.

Between the poles, by the way, there is a model that often gets overlooked: tech leadership on demand. I explored how far this borrowed judgment can carry you in my article on the fractional CTO.

When is an agency the better choice?

An agency is the better choice when your project needs real parallelism, many disciplines, or guaranteed availability. Concretely, that looks like this:

• Several sub-projects with hard deadlines at once. A launch with an app, a portal, and integrations in four months: at that breadth, even the best single-person agent setup runs out of road.
• Full discipline breadth as a package. Brand, UX research, frontend, backend, operations, support hotline. The well-rehearsed multi-person setup of a good agency is honestly worth more here than any individual.
• Guaranteed backup. If your business depends on someone always being reachable, through sickness and vacations, a team has a structural advantage. The downtime-risk argument is legitimate. How to secure it with individuals comes further down.
• Formal requirements. Some clients require certifications, compliance documentation, several named roles. That is agency or systems-house terrain.

If you recognize yourself in these points, an agency gives you exactly what your project needs: coordination, breadth, backup.

And the offshore route?

Offshore outsourcing fits when three things come together: a large, long-running project, crystal-clear specified requirements, and someone inside your company who can manage international teams and has the time to do it. That is the honest checklist, and note what it measures: it says nothing about the quality of the teams, excellent developers exist everywhere in the world. It says something about the prerequisites on your side. In a company of five to twenty people, usually none of the three boxes is ticked. The half-price hourly rate is real. So is the management capability it requires.

When is a single expert the better choice?

A single experienced expert is the better choice when your project has a clear core and you want direct access to the judgment that makes the decisions.

• A clearly defined core. An internal tool, an automation, an app with one purpose. One workstream, a manageable set of disciplines.
• A direct line.You talk to the person writing the code, without translation through a project manager. From my own work, this is the advantage almost nobody prices in: every translation layer is a place where your “roughly like this” turns into something you did not mean. And you only notice misunderstandings once they are built.
• Senior judgment without the senior-team price.The same person who makes the architecture decision also builds. The amplifier finding from the DORA Report works in your favor exactly here: experience plus AI leverage delivers the output of a small team without you paying for that team's overhead.
• Budget logic.Your money flows into development instead of structure. On small and mid-sized projects, that is the difference between “doable” and “postponed”.

And since I told you my angle upfront, let me say this directly too: an AI-native freelancer is, in my view, the most interesting option in this field right now. Much of what used to strictly require an agency, parallel workstreams, built-in quality assurance, clean documentation, is now part of the toolkit of a single person with a good agent setup. It is exactly how I work myself.

And one point that never shows up in a cost table: it is also about who you talk to when things get difficult. Every project has a moment where something jams. The question “Who picks up the phone then, and does that person understand my business?” is a real decision criterion, not a soft one.

How do you secure your project, whichever way you decide?

Which leaves the one worry everyone has about individuals: What if that one person drops out? The worry is legitimate. And it has a contractual answer. This is criterion five in detail: three safeguards you can demand, from agencies and offshore teams just the same.

1. The code belongs to you. Your own repository on your own account, from day one. No vendor lock-in, neither with an agency nor with an individual.
2. Documentation and tests as part of the delivery, not as an extra. An outside expert must be able to take over the project without doing archaeology. Automated tests are the handover documentation that cannot lie.
3. Clarify operations before go-live. Who deploys, who monitors, who gets called? These questions decide the second week, no matter who built the software.

I wrote down nine questions that tell you whether software will actually hold up, no matter who built it, in “Production-ready software”.

Finally, the context for why so many companies face this question in the first place: in Germany alone, more than 100,000 IT positions remain unfilled (Source: Bitkom, 2026), and 70 percent of companies name the skills shortage as a digitalization hurdle, 66 percent a lack of time (Source: Bitkom, 2026). Hiring your own developers simply is not a realistic option for many small companies. Outsourcing here is above all one thing: access to capabilities the job market barely offers. So your team can stay on what it was hired to do.

Five criteria, one sheet of paper

Freelancer, agency, or offshore: the decision is made by your project, along the five criteria: parallelism, discipline breadth, and coordination costs make the call. Availability and code ownership secure it.

My suggestion for the next step: take your project and work through the five criteria in writing. Half an hour, one sheet. With the answers, quotes suddenly become comparable, because you know which services you need and which ones you would be paying for without using them.

And if you get stuck on one of the five criteria, that is exactly what a good first conversation looks like. Not with someone who wants to sell you a model, but with someone who tells you what your project needs. Even if the answer is “agency”.

The market is re-sorting itself right now. The line between one person and a team moves a little every year. The labels age, the five questions stay. And if you have answered them and still are not sure which model carries your project, you are not missing knowledge. It is time for a conversation.