In-House Team vs Agency vs Freelancers: The Real Cost Comparison

What does an in-house engineer actually cost beyond the salary line?

The figure in the offer letter is the smallest number in the calculation. Most in-house team vs development agency comparisons fail at the first step, when someone sets a $140,000 salary against a $22,000 monthly invoice and concludes the agency is expensive. One is a wage; the other is a delivered-capacity price with employment costs, management and tooling already inside it.

The statutory portion is not negotiable and it is knowable. A US employer pays 6.2% Social Security on wages up to the annual cap, 1.45% Medicare uncapped, federal unemployment tax at an effective 0.6% on the first $7,000 in most states, and state unemployment insurance that varies by state and by claims history. A UK employer pays National Insurance at 15% above the £5,000 secondary threshold, plus pension auto-enrolment of at least 3% of qualifying earnings.

Add those together and you reach the multiplier most finance teams already use: total cost of employment runs about 1.25 to 1.4 times base salary before any management overhead. On a $140,000 salary that is roughly $175,000 to $196,000 a year, plus a one-off fee of $21,000 to $35,000 in year one if a contingency recruiter found the candidate.

Why a contract rate and a salary are not comparable numbers

An employee is paid for 52 weeks. Subtract annual leave, public holidays, sick days, onboarding, company meetings and the hours spent interviewing other candidates, and productive engineering weeks land somewhere in the mid-forties. That is the denominator that matters, and it is the one nobody puts in the spreadsheet.

The second missing line is bench time. Headcount is a standing commitment: when a project ends, the developer who built your iOS app stays on payroll while you work out what they do next. Over two years those gaps are often the largest single source of waste in an in-house team, and they never appear as a cost because they look like a planning failure.

The third is management — one-to-ones, performance reviews, hiring loops, cover for parental leave. Real work, expensive work, and in the agency model it sits inside the rate rather than inside your calendar.

Contract pricing varies enormously by geography and specialism. As observed market ranges, not anyone’s rate card: independent contractors doing product engineering in the US or UK commonly sit between $60 and $150 an hour; onshore agencies in those markets between $120 and $250; nearshore firms in Latin America or Eastern Europe between $45 and $90; offshore firms in South and Southeast Asia between $25 and $60. What pushes a project to the top of a band is consistent — regulated data, a hard deadline, a team of five rather than two, deep integration with systems you do not control, and changes of direction after the architecture is set.

In-house team vs development agency vs freelancers, side by side

In-house team vs development agency: which one starts faster?

Speed decides this more often than cost does. Hiring runs through a job specification, sourcing, screening, two or three interview loops, an offer that may be countered, then a notice period — two weeks under US at-will employment, one to three months in the UK and much of Europe. Two to four months from decision to first commit is a realistic plan, and that is before ramp.

Ramp is a separate cost, and the honest measure of it is the first independent change to production, not the first pull request. In an unfamiliar domain and codebase, four to eight weeks is normal, and throughout it the new hire consumes attention from people who are already productive.

An agency’s advantage here is not that its people are better. The forming cost has already been paid by somebody else: the pairs have worked together, review conventions exist, CI is a template rather than a decision, and someone already knows how to wire up the observability stack. You are buying a team, not individuals who will become one on your budget.

What happens when the person who knew everything leaves?

Every model has a bus factor problem and each one fails differently. In-house, the knowledge leaves permanently and the replacement starts from the code. With an agency, continuity is contractual, but the agency’s own attrition is invisible to you until it is not — the person who designed your event model can be rotated onto another account unless the contract names individuals. With freelancers the risk is highest: no bench, no cover, no colleague who read the code last week.

The mitigations are identical across all three models, cost almost nothing to insist on at the start, and are close to impossible to retrofit at the end.

• Require every non-trivial decision to land in an architecture decision record in the repository, not in a chat thread that expires with the retention policy.
• Own the cloud accounts, DNS, repositories, CI provider and app store listings yourself from day one — never in a vendor’s or contractor’s name.
• Define infrastructure in Terraform, Pulumi or CloudFormation so that no individual laptop is the deployment mechanism.
• Verify quarterly that CI can build, test and deploy from a clean checkout with no undocumented local steps. This single test surfaces most hidden knowledge.
• Contract for a paid handover window rather than relying on goodwill after the final invoice.
• Have two people review anything load-bearing, so that no subsystem has exactly one reader.

When is hiring an agency the wrong choice?

There are situations where bringing in a firm — including ours — is the worse decision, and saying so is more useful than pretending otherwise. The clearest is when the software is the company. If the product is what customers pay for and will be built continuously for the next five years, the compounding value of domain knowledge held by permanent staff beats any rate arbitrage.

The second is when you cannot yet articulate what you want. A firm will happily run a discovery process and bill for it, but if the specification changes materially every week because the business model is still moving, the contract converts that churn into invoices.

The third is small, continuous work. A few hours a week of maintenance is badly matched to minimum team sizes, retainer floors and the cost of context-switching a shared team into your codebase. A part-time contractor who already knows the system is cheaper for both sides.

The fourth is regulatory. Some environments require named, vetted, geographically constrained personnel, or prohibit third-party access to production data outright. No commercial argument overrides that.

Where freelancers win and where they quietly fail

Freelancers are the best value in the market for bounded, specified, single-discipline work. Implementing a design system in an existing component library. A Stripe integration including webhooks, proration and dunning. A one-off data migration with a verification script. An accessibility audit against WCAG with a remediation list. In each case you can describe done and evaluate the result without having to trust the process.

They fail on the inverse. Work spanning backend, frontend, mobile and infrastructure at once becomes an integration problem you now own, because nobody else holds the whole picture. And the review step quietly disappears: an individual merges their own work, so the second pair of eyes that catches the concurrency bug or the missing index does not exist unless you supply it.

That is why quality variance is widest in this model. There is no institutional standard behind the person and no colleague who will refuse to approve the pull request. Two contractors quoting the same rate can produce work that differs by a factor most buyers would not believe.

A decision rule you can apply this week

Work through these in order and stop at the first clear answer. Most teams find they were arguing about cost when the binding constraint was calendar or accountability.

1. Will customers still be paying for this capability in three years? If yes, its core belongs in-house whatever else you decide.
2. Does the work have a defined end date? If yes, permanent headcount is the wrong instrument — a fixed cost against a variable need.
3. Can anyone internally judge the output technically? If not, an agency with a named architect accountable for quality beats individuals who each grade their own work.
4. Is the scope one discipline or four? One is a contractor. Four is a team, and assembling one from contractors makes you the integrator.
5. What does a month of delay cost in revenue, runway or a missed contractual date? If that number is large, the agency premium is the cheapest part of the decision.

Most companies do not pick one model and hold it. The arrangement that works is a small permanent core owning product decisions and the differentiating parts of the system, an agency team for surge capacity, and contractors for bounded tasks. A common sequence is an agency building version one against a fixed scope while you recruit, then a structured handover into permanent hires it helped onboard.

We work both ends of that arrangement, and if you are weighing the three models against a specific build, it is worth a conversation about which one your constraints actually point to.