The agency terms in our keyword data are falling faster than almost anything else — marketing agency and app marketing agency are both down ten percent, among the steepest declines in a fifty-term set where the average is around minus five.
People are still searching. They are searching less, and they are searching while the tooling that makes in-house viable gets dramatically cheaper. This post is about how to decide, with the numbers laid out in the way they are usually not.
The comparison as it is usually presented
An agency quotes a monthly retainer. You compare it against a salary. The salary looks smaller. You hire.
Then the actual cost shows up over the following year, and it is not the salary.
What in-house actually costs
Salary, loaded. Base plus employer taxes, benefits, equipment, and software seats. The multiplier on base varies by geography but it is never one, and it is usually meaningfully above one. Budget accordingly.
Ramp. A new content hire produces very little for their first stretch, because they are learning your product, your customer, and your voice. This is not a performance issue, it is the nature of the work — and it is a real cost that never appears in the comparison.
Management. Someone directs this person, reviews output, and makes calls on strategy. If that someone is you, the cost is your time at whatever your time is worth. Founders systematically price this at zero and it is frequently the largest line item.
Coverage risk. One person means output stops when they are sick, on holiday, or interviewing elsewhere. An agency has bench depth; a solo hire does not.
Departure. When they leave, everything they learned about what works for your audience leaves with them — unless it was written down somewhere structural. Almost never is. You then pay ramp again.
That last one is the expensive one, and it is almost never in the spreadsheet.
What an agency actually costs
The retainer, which is the visible number.
Your time anyway. Briefing, review, approvals, feedback rounds. Less than managing an employee, not zero, and considerably more than agencies imply during the sale.
Voice distance. An agency is modeling your voice from the outside. Some are very good at it. The gap shows up in the content that requires actual product knowledge or genuine opinion — which is increasingly the only content that works, since the generic tier is being absorbed by generated answers.
Attribution fog. You will not cleanly know what the retainer bought. Agencies report on what they control (output, reach, engagement) rather than what you care about (revenue), because the connection between the two is genuinely hard to establish and reporting it honestly makes the retainer look worse.
Switching cost. Changing agencies means re-teaching everything. The knowledge lived with them, not with you.
The line that actually decides it
Strip away the spreadsheet and the question is: does the knowledge accumulate on your side of the boundary?
Content marketing produces two things. Published content, and knowledge about what works for your specific audience. The content has a short useful life. The knowledge compounds indefinitely — it is what makes the second year cost less and produce more than the first.
Agencies deliver the first reliably. They deliver the second to themselves.
That framing changes the decision:
Agency makes sense when you need volume now, the work is genuinely commoditized (localization, ad variants, production), you lack the internal knowledge to direct someone, or content is a supporting function rather than a channel you are betting on.
In-house makes sense when content is a primary channel, the work requires real product depth, you are in a niche where credibility is the product, or you intend to still be doing this in three years.
Neither, for a lot of small teams. The default is now a small amount of your own time plus tooling, and that default got much stronger recently. The reason is not that AI writes well enough to replace a person — it does not. It is that the bottleneck moved. Production used to be the constraint; now it is knowing what to produce. That is a measurement problem, and measurement is much cheaper to tool than writing was.
If you go in-house or solo, build the record first
The single highest-leverage thing, whichever way you go: make sure the knowledge accumulates somewhere that is not a person's memory.
Concretely, that means every published piece has a durable record of what it was trying to do, what tone it used, and what happened. Not a dashboard you glance at — a structured record that the next round of production actually reads as input.
Do that and the agency-versus-in-house question gets much less scary, because the expensive failure mode in both directions is the same one: the knowledge walks out the door. An agency takes it when the contract ends. An employee takes it when they resign. A record does not leave.
This is the thing we built clownfish101 around, so treat that as a disclosed bias. But the principle holds regardless of tooling — a disciplined spreadsheet beats no record at all by an enormous margin.
A reasonable sequence
If you are genuinely unsure, do not decide in the abstract.
Run it yourself for one full content cycle — long enough to see results, which on most platforms is longer than people think. Keep the record. You will learn what the work actually requires and roughly what a competent version of it produces.
Then hire or contract against that knowledge instead of against a guess. You will brief better, evaluate better, and know within weeks rather than quarters whether it is working. And if you decide to bring it in-house later, the record is already there.
clownfish101 is a content ops loop: generate on tone, publish semi-automatically, recover results by screenshot, feed them back into the next batch. Local-first — the record of what works stays on your machine, not an agency's. See the loop →