Agency vs freelancer vs in-house: an honest cost and risk comparison
The choice is usually framed as a price comparison, which is why it is usually got wrong. A salary is not a cost, a day rate is not a cost, and what actually settles this is how much specialist work you have, how predictable it is, and whether anyone inside your organisation can tell good work from bad. Here is the arithmetic, the risks and a worksheet.
Three models, described without the sales pitch
Strip the positioning away and there are three ways to get specialist digital work done, distinguished by who carries which risk.
In-house means the capability is a permanent cost on your payroll. You carry recruitment risk, the risk of being wrong about the hire, the cost of absence and the cost of the skill going stale. In exchange you get context — someone who knows your product, your customers and your internal politics well enough to be useful in a meeting they were not briefed for.
Freelance or contract means you rent a specific person for a defined period or a defined piece of work. You carry no employment cost and no bench, but you carry all of the management and none of the redundancy. When that person is unavailable, the capability is simply gone.
Agency means you buy a team’s output rather than anyone’s time. Several disciplines are available at partial allocation, absence is absorbed by someone you never meet, and the supplier carries the cost of being wrong within whatever the scope says. You pay for that structure, and you pay for the sales, account management and idle capacity that keep it standing.
Two hybrids blur the categories and are worth naming. Staff augmentation is a contractor working inside your process, tooling and management line — commercially a supplier, operationally an employee, which is the exact ambiguity the off-payroll rules exist to police. Fractional leadership — a fractional CMO or CTO — is senior judgement bought at a few days a month, a different product from either execution capacity or a permanent hire.
Organisations that have done this a while do not pick one model. They allocate different layers of the work to different models, which is where this article ends up by a longer route.
What a salary actually costs
The commonest error in this decision is comparing a supplier’s day rate with an employee’s salary. They are not comparable quantities. One is a price for delivered work, the other is a single input to a cost base.
The statutory floor
Before any discretionary spend, employing someone in the UK carries obligations you cannot decline. Employer Class 1 National Insurance runs at 15% on earnings above the secondary threshold. Automatic enrolment requires a minimum total pension contribution of 8% on qualifying earnings, of which at least 3% must come from the employer. Employers with a pay bill above the levy threshold pay the apprenticeship levy at 0.5%. You must register with HMRC as an employer, run PAYE, and arrange employers’ liability insurance before anyone starts. Statutory Sick Pay is payable by you for up to 28 weeks of a single absence.
That is roughly a sixth on top of gross salary before you have bought a laptop. It is also the easy part, because it is knowable in advance.
The part nobody puts in the business case
The discretionary and hidden lines are usually larger than the statutory ones and are almost never modelled:
| Cost line | Why it is missed |
|---|---|
| Recruitment | Agency fees or the internal time of everyone who sits on an interview panel |
| Equipment and software seats | Per-seat licensing for analytics, design, crawling, rank tracking, ad management and project tools |
| Training and conferences | The only defence against a lone specialist’s skills quietly ageing |
| Management time | A specialist reporting to a non-specialist consumes more management attention, not less |
| Onboarding ramp | Full salary paid against partial output while the person learns your product and market |
| The cost of being wrong | Notice, and after two years’ continuous service, statutory redundancy pay |
Available days, not calendar days
The denominator matters more than most of the numerator. A five-day week gives 260 nominal working days a year. Statutory leave removes 28 of them for a full-time worker. ONS put average sickness absence at 4.4 days per worker in 2025. That is around 227 days before you subtract training, internal meetings, all-hands, appraisals and the administrative overhead of being an employee.
So the fully loaded implied day cost of a hire is roughly (salary + statutory on-costs + tooling + share of recruitment and management) ÷ about 225, not salary ÷ 260. Run that calculation with your own numbers before you assume a supplier is expensive. Sometimes it confirms the hire comfortably. Sometimes it closes a gap that looked enormous.
One nuance worth knowing: the Employment Allowance offsets part of an eligible employer’s National Insurance bill, but it is claimed once per business rather than per employee. Your first hire is therefore cheaper in NI terms than your fifth, which quietly flatters the business case for starting a team and understates the cost of growing one.
Where freelancers win, and where they break
Freelancers win on three things, and they win decisively.
The first is senior skill without overhead. You pay for one person’s time and nothing else — no account manager, no new business function, no bench capacity sitting idle between projects. For a bounded specialist task, that is the cheapest senior hour on the market.
The second is speed of start: no requisition, no notice period, no onboarding programme.
The third is directness. You brief the person doing the work, so nothing is lost between the meeting and the execution — a larger advantage than it sounds.
They break in equally predictable ways. Availability is binary: illness, a better offer or a larger client’s escalation removes the capability entirely, with no cover. Capacity has a hard ceiling, and hiring a second freelancer at short notice restarts the search. Breadth is limited — one person rarely does strategy, execution and measurement to the same standard, and the weakest of the three is where the work quietly fails, with no peer to catch it. And the management burden lands on you: with an agency you brief an account team, with a freelancer you are the account team.
Two commercial points deserve attention. Settle the off-payroll working position in writing before work starts if the person operates through their own company. For medium and large private-sector clients, HMRC guidance places responsibility for determining the worker’s employment status on the client, who must produce a status determination statement including the reasons for the determination; where the client is a small business, the intermediary determines it instead. Getting this wrong is not a paperwork problem.
The second: ask directly whether the freelancer subcontracts. Plenty do, entirely legitimately. Some sole traders are a front for a team you have not met, and some agencies are a single person with a company website. Neither is disqualifying. Not knowing which you have bought is.
What you are actually buying from an agency
Not hours. If you are buying hours, you have bought the wrong thing and should have hired a contractor.
What an agency can structurally sell that neither other model can is partial allocation across several disciplines. Real digital work rarely needs one full-time specialism. It needs a third of a technical SEO, a fifth of a paid media specialist, a quarter of an analyst and occasional access to a designer. You cannot hire those fractions. That is why the honest comparison is rarely one salary against one retainer.
You are also buying continuity through absence, pattern recognition from seeing the same failure across many accounts, and — where the contract is written properly — the transfer of estimation risk. If a fixed-scope statement of work takes longer than expected, that is the supplier’s problem, and the price of that protection sits inside the rate.
And you are paying for things that appear on no deliverable list: new business costs, account management, tooling licences, and the gap between a person’s cost and what they must be billed at to cover non-billable time, holiday, training and the bench. That multiple is the honest answer to why a day rate looks high next to a salary. It is not a markup on like-for-like work, because the work is not like-for-like — but it is real, and a supplier who cannot explain it plainly is worth a second look.
Retainer or statement of work — a real disagreement
The industry does not agree on this, and articles that pretend otherwise are selling something.
The case for retainers: digital marketing is compounding work. Search visibility, content authority and account structure improve over quarters, not deliverables. A retainer keeps a team in your context, preserves institutional knowledge and lets the mix shift as the situation changes. Chopping the same work into projects means paying for the ramp repeatedly.
The case against: a retainer with no defined output pays for presence rather than progress, and the incentive to demonstrate value decays once the payment is automatic. Fixed-scope work forces both sides to say what “done” means.
Both arguments are correct about different failure modes, so the resolution is not choosing a side but writing the contract properly: a retainer needs named deliverables, a named allocation and a review point, or it becomes a subscription to a relationship. The Cabinet Office’s guidance on contracting for agile work makes the parallel point for build projects — specify the outcome rather than the outputs, and never tie payment milestones to the delivery of the process itself, such as completing a set number of sprints. If you are running this procurement now, the twenty questions worth asking before you sign cover the contractual half.
The in-house tipping point
There is no revenue figure at which marketing should come in-house, despite the number of articles that supply one. The tipping point is structural, and there are four honest tests.
- Is the work continuous or episodic? A content operation shipping every week, or paid search management at meaningful volume, needs attention most days and is cheaper owned. A quarterly campaign and an annual audit are not a job.
- Is domain knowledge the hard part? If understanding your product takes months, you are paying that ramp cost every time anyone changes on either side. Owning the knowledge internally stops the meter.
- Can you fill the role at the seniority you need? The classic failure is replacing a senior specialist supplier with a junior generalist because the salary fits the old retainer line. Output does not drop proportionally; it drops off a cliff, because the missing thing was judgement.
- Can you evaluate the work? This is the one that decides it. Hiring a specialist nobody internally can assess produces a year of activity and no result, and neither party will notice until an external review says so.
There is a counter-argument that in-house advocates rarely engage with: lone specialists flatten. Someone doing organic search alone for two years, with no peers and no exposure to other estates, stops learning at the pace the field moves. That is survivable, but only if you budget for training, external audit or peer review deliberately, rather than discovering the drift when rankings do.
Maturity also cannot be recruited. Nielsen Norman Group’s research notes that lifting an organisation by a single maturity stage can take several years, and cautions against trying to leapfrog one. A first hire arrives into the culture that already exists and inherits its ceiling.
Hybrid models, and why most settled teams end up there
The arrangement that survives contact with reality is almost always a split by layer rather than by channel.
In-house owns direction, context and measurement. Strategy, prioritisation, brand knowledge, the relationship with sales, and the numbers. This is the smallest defensible internal footprint and the one organisations most often skip. The Cabinet Office calls it the intelligent client function and is explicit that you need it in addition to whatever delivery capacity you buy. Without it you are not managing a supplier; you are hoping.
Outside owns specialist execution and peaks. Disciplines needed fractionally, work that spikes seasonally, and anything requiring tooling or skill depth you cannot justify permanently.
Three variants of this are worth distinguishing, because they fail differently:
- Internal strategist plus agency execution. The most robust. Fails when the internal role is filled by someone junior enough to be captured by the supplier’s plan.
- Fractional leadership plus internal doers. Works when you have willing people and no direction. Fails when the fractional lead never has enough hours to own an outcome, and becomes an expensive advisory presence.
- Staff augmentation. Contractors inside your process. Works when your process is genuinely good. Fails when it is used to avoid the discipline of a scope, in which case you have taken on management overhead and the off-payroll question in exchange for nothing.
Keep measurement and analytics internal wherever you can. It is the one function where independence changes the answer, and the marginal cost of owning it is low relative to what it protects.
Risk: key-person dependency, continuity and knowledge loss
Every model carries key-person risk. They differ only in who absorbs it.
With an employee, the risk is concentrated and slow-moving: notice periods, and after two years’ continuous service, statutory redundancy pay banded by age and capped. With a freelancer, it is concentrated and fast-moving — the capability can disappear between one week and the next with no contractual remedy worth exercising. With an agency, it is diffuse but invisible: your team can change without anyone consulting you, which is why naming individuals and their allocation in the statement of work is worth more than most clauses people negotiate harder.
Knowledge loss is the greater cost in all three cases, and the mitigations are identical regardless of model:
- Own every account. Search Console, analytics, ad accounts, tag manager, the domain registrar, the code repository, the CMS. In your organisation’s name, with you as owner and the supplier as a delegated user. This single item causes more damage at handover than everything else combined.
- Insist on a decision log, not just deliverables. What was changed, when, and why. The reasoning is the expensive part to rediscover; the artefacts are usually recoverable.
- Make documentation a deliverable with an acceptance criterion. Documentation promised as goodwill is documentation written badly at the end, if at all.
- Apply the exit test quarterly. Write down what you would take with you if the relationship ended tomorrow. If the honest answer is “nothing”, the dependency is the product.
- Settle data responsibilities in writing. If a supplier handles personal data on your behalf, you remain the accountable party, and that arrangement needs a contract rather than an assumption.
Speed to value against long-run cost
These trade against each other almost perfectly, which is why cost-per-day comparisons mislead.
| Time to first useful output | Marginal cost at high volume | Cost of being wrong | |
|---|---|---|---|
| Agency | Fastest — the team and tools already exist | Highest | Low: end the scope |
| Freelancer | Fast once found; finding is the slow part | Moderate, but capped by one person’s capacity | Low, but the gap is immediate |
| In-house | Slowest — recruit, notice, then ramp | Lowest | Highest: notice, redundancy, and a repeat search |
The decisive variable is rarely price. It is opportunity cost — what a quarter of not doing this actually costs you. If the answer is modest, take the slow cheap route and build the capability properly. If a quarter of drift costs more than a year of supplier fees, the expensive fast option is the cheaper one, and the spreadsheet that says otherwise is measuring the wrong thing.
A scoring worksheet for your situation
Score each row 1 to 5 for your own situation, then read the pattern rather than the total. This is a thinking tool, not a formula, and it works best filled in by two people separately and then compared.
| Dimension | Score 1 | Score 5 |
|---|---|---|
| Volume and continuity | Occasional projects | Daily, ongoing work |
| Disciplines required | One specialism | Four or more at partial allocation |
| Domain knowledge depth | Generic best practice suffices | Months to understand the product |
| Internal ability to evaluate | Nobody can assess the work | An experienced practitioner is in post |
| Demand predictability | Highly seasonal or spiky | Flat and forecastable |
| Cost of a gap in cover | Tolerable | Revenue stops |
| Data or confidentiality constraints | None | Restricted, cannot leave the building |
| Appetite for fixed cost | None; budget must flex | Comfortable with permanent headcount |
Reading it honestly:
- High on continuity, domain depth, evaluation ability and fixed-cost appetite points in-house. All four, not three.
- High on disciplines required and demand spikiness, low on evaluation ability points to an agency — with the caveat that the low evaluation score is the thing to fix first, whatever you decide.
- High on one specialism, low on breadth, moderate on continuity points to a freelancer, provided the cost-of-gap row is not a 5. If it is, a single person is not an acceptable answer.
- Mixed scores — the usual outcome — point to a hybrid, and specifically to owning the rows you scored highest on evaluation and domain depth.
The row that overrides the others is internal ability to evaluate. Score it a 1 and every model degrades: the hire cannot be managed, the freelancer cannot be checked, and the agency cannot be held to anything meaningful.
Moving between models without losing momentum
Transitions destroy more value than the wrong model ever does, usually because they are run as a cost decision rather than a knowledge-transfer project.
Agency to in-house. Hire first, then serve notice — not the reverse, however uncomfortable the overlap looks on a budget line. Scope knowledge transfer as paid work with named outputs, because unfunded handovers get done in the last hour of the last day. Confirm account ownership before notice is served. Where budget allows, keep a reduced external scope for review, so the new hire has a peer rather than an echo.
In-house to agency. Usually triggered by a resignation, which means it happens at the worst moment with the least documentation. The defence is to write things down while the person is still there and happy.
Freelancer to team. The hardest, because there is often nothing written down at all — one person held everything in their head, entirely reasonably. Buy documentation as a discrete engagement while the relationship is still good.
In every direction, avoid switching across your peak trading period. The saving is small and the exposure is not.
When we tell people not to hire an agency
We would rather say this early than be six months into an engagement that was never going to work.
When the problem is not a marketing problem. If the product does not fit the market, or the pricing does not stand up, more traffic converts a slow problem into a fast one. Fix the thing first.
When you cannot measure anything yet. Buying execution before measurement means paying for work nobody can evaluate, including us. Analytics, conversion tracking and clean attribution come first, in whatever order is cheapest.
When the budget only covers one channel at token volume. Spread thinly across three channels, it achieves nothing anywhere. One freelancer’s focused attention on one channel usually beats it.
When there is no internal owner. An agency with no counterpart drifts towards whatever is easy to report, and nobody is at fault.
When you can genuinely do it yourself. Google’s own guidance for site owners says that a small local business can probably do much of the search work itself, and it is right. The same applies to research: Nielsen Norman Group’s finding is that around five participants surface roughly 85% of usability problems, and that three small rounds beat one large study — which means the first useful round of user testing is something you can run yourself, this month, with people you already know. You will learn more from watching five customers use your site than from most reports you could commission. If you are still evaluating suppliers afterwards, the red flags in an SEO pitch are a faster filter than another round of proposals.
The version of this decision that ages well is not “agency or hire”. It is: which layer of this work must we own to remain a competent buyer, and which layer is genuinely better bought? Answer that first, and the rest is procurement.