Media department vs agency. Own the system or rent the hours.

Because brands in-housed the headcount, not the system. By the ANA's count, 82 percent of members now have an in-house agency, up from 42 percent in 2008. Yet 92 percent of the same marketers also work with external agencies. Fifteen years of firing the agency never ended the renting.

The numbers come from the ANA's “The Continued Rise of the In-House Agency: 2023 Edition,” a member survey fielded in early 2023 with 162 respondents.¹ The in-house line climbs steadily: 42 percent in 2008, 58 in 2013, 78 in 2018, 82 in 2023. And 92 percent of those with in-house agencies said they also work with external agencies, up from 90 percent in 2018.

Hold the two lines next to each other. The in-housing movement doubled in fifteen years, and the renting line barely moved. Whatever those brands bought back, it was not independence.

What they bought back was headcount. Desks, titles, salaries, a floor of the office that does marketing. The thing that turns a brief into finished media, the workflows, the templates, the accumulated judgment about what works for this brand, either never came in-house or walked out with each departure. So every time the internal team hit a capacity wall or a specialist gap, the retainer came back through the side door.

The renting is structural. It survives any org chart you draw, because the org chart was never the thing being rented.

You did not fire the agency. You hired its headcount and kept its invoice.

Which is why the consensus advice, keep strategy in-house and rent execution, has quietly inverted. Strategy was always the ownable part; it lives in the founder's head and it never scaled anyway. Execution, the part everyone tells you to rent forever, is the part that just became ownable software. The rest of this piece is about what that changes.

You are choosing who owns the system that makes your media, not who does the work. The agency retainer, the AI platform seat, and the lone hire's personal workflow are all rentals. The third option, a media department you own outright, is the one no ranking page names.

Search this exact question and the results mostly explain the media department inside an ad agency, the people who buy airtime. The decision pages on the adjacent searches all frame the same choice: which labor pool do you hire, theirs or yours. That is the wrong axis. Labor was never the scarce thing. The scarce thing is the system the labor runs on.

Put the usual three options on the ownership axis instead. The agency retainer is rent on someone else's system; you buy its output by the month. The AI platform seat is rent on someone else's system with a smaller invoice; the agents and prompts sit in the vendor's account. Even the solo marketing hire is, more often than admitted, rent on one person's private workflow. It lives in their head and their bookmarks, and when they resign it resigns with them. Three prices, one landlord structure.

The third option needs a name, because nobody on these pages gives it one. A client-owned media departmentis an agentic production system: the agents, the prompts, the pipelines, and the infrastructure, registered in your name, running on your own API bill, directed by your people. Not a vendor's tool you log into. Not a contractor's process you glimpse in a monthly report.

A working system that sits in your accounts the way your codebase and your customer list do. One that keeps working when any given vendor or employee walks away.

By the vendors' own numbers, a mid-market retainer buys one account manager, one content writer, bi-weekly check-ins, and a monthly report that arrives after the month it describes is already over. Every dollar of it is rent: stop paying and the output stops with it.

Take the pricing from the people selling it. Stackmatix, a growth-marketing agency publishing its own estimates, puts a typical full-scope retainer at $5,000 to $20,000 a month.² It prices a single in-house marketing hire at $125,000 to $200,000 a year fully loaded, and full four-to-six-person channel coverage at $600,000 to $1.2 million a year.

Enrich Labs, a vendor selling AI marketing agents, says the average full-service SMB retainer runs $5,000 to $15,000 a month.³ It typically buys one account manager, one content writer, bi-weekly check-ins, and month-delayed reporting. These are self-reported vendor figures, not independent studies. But notice they all price the same unit: hours, billed monthly, forever.

Enrich's own words on what that cadence costs you: “When a monthly agency report arrives on the 5th, the month being analyzed is already over. The campaign that underperformed cannot be course-corrected. The creative that was crushing it cannot be amplified.”³ That is not an indictment of any particular agency. It is what renting hours at a monthly cadence structurally delivers.

And when the rented hours go wrong, they go wrong opaquely. As one commenter, zerotousers, put it on the Indie Hackers thread “I regret my $46k website redesign”: “the cost of failure is usually higher when you hire an agency vs. a person.”⁴You do not know who exactly is doing the work, or when. A retainer is a permanent line on the P&L that builds no equity. Twelve months in, you own twelve reports.

AI can replace the execution layer of most agency retainers. But swapping a mid-band $10,000 retainer for a $299 to $499 platform seat, Enrich's own pricing, only changes the landlord. You still own none of the agents, prompts, or infrastructure, and cancellation still turns everything off.

The 2026 replace-your-agency wave is real, and its math is not wrong. Enrich Labs claims its platform runs $299 to $499 a month against that $5,000 to $15,000 retainer, a 60 to 80 percent cost cut, with a 30-day agency-to-AI migration.³Take the vendor's numbers at face value and the execution layer of a mid-market retainer is clearly replaceable.

What is nowhere in the pitch is ownership. The agents are theirs. The prompts are theirs. The infrastructure is theirs. Your brand's accumulated context lives in their database, at their pricing, under their roadmap. And a shared platform writes toward a shared middle, the same pull we traced in why all AI marketing content looks the same.

So the trade on offer is agency rent for SaaS rent. Cheaper rent, often much cheaper, but the structure from the first section survives intact. The system that makes your media still belongs to someone who can reprice it, sunset it, or get acquired out from under you. You did not exit the landlord relationship. You refinanced it.

What still works the day you stop paying?

There is one question that sorts every option on this page, and none of the ranking pages ask it. Call it the cancellation test. Cancel the retainer and the output stops. Cancel the platform seat and the login stops. If the honest answer is “nothing,” you are renting, whatever the invoice says.

Whether an agent can actually carry the production is its own question, with its own judgment problem, and we answered it in can an AI agent run your content production. Keep the cancellation question in hand; the next section is what happens when people run it too late.

Usually nothing. When the agency created the ad account, Google's own documentation says the creating manager account owns it. Meta ad accounts cannot leave the Business Manager that owns them. The warm audiences and optimization history tied to those accounts stay behind, and breakup guides document exit fees up to 50 to 100 percent of the remaining contract value.

This is not a horror story; it is documented mechanics. Google's manager-account (MCC) documentation states that when a manager account creates a client account, the creating manager “will automatically become the owner.”⁵ Meta ad accounts cannot change the Business Manager that owns them, as the agency iDimension lays out in its breakup guide.⁶ So if your agency spun up the accounts, a common onboarding pattern, the accounts were never yours.

Everything welded to those accounts goes with them. As the consultancy North Country Consulting puts it: “your conversion history, your audience lists, your two years of bidding data, gone.”⁶And the agency-breakup guides from Market Correct and Clicks Geek document exit charges ranging from two to three months' retainer up to 50 to 100 percent of the remaining contract value.⁶ That is the custodyfight, in the platforms' own words and the practitioners' own numbers.

Here is the strange thing about the search results on this topic. The divorce evidence, the how-to-get-your-accounts-back guides, lives on entirely different pages from the marriage advice, the how-to-choose-an-agency listicles. Nobody connects them. The exit checklist people read after firing an agency is the buying spec they should have read before hiring one.

And the checklist should run past accounts and audiences to the production system itself. Who owns the prompts that write in your voice? The templates that carry your look? The pipeline that turns a brief into a finished piece? If the answer is “the vendor,” you already know what the cancellation test will return.

Cancel every contract and watch what keeps working. Whatever survives is what you own.

No. The seven-figure number comes from a vendor modeling an enterprise ML engineering team; Docket's own build-vs-buy analysis puts that at $1.4 to $1.6 million a year. A fixed-scope agentic build on commodity models, with a written definition of done, lands for a founder-led company at roughly the price of a few months of retainer.

The scary number deserves its context. Docket, a vendor selling pre-built AI marketing agents, estimates in its own build-vs-buy analysis that an in-house enterprise AI marketing agent runs $1.4 to $1.6 million a year, plus roughly six months of engineering runway.⁷That is a model of an enterprise hiring ML engineers to build custom agent infrastructure from scratch. As an argument for buying that vendor's product, it is doing its job. It is not a model of what a founder-led company needs.

The frontier models are a commodity now. You rent intelligence by the token and own everything built on top of it.

We can give you the first-hand math, because building these is what we do. A paid pilot week runs $2,500 and credits into a build sprint that starts at $10,000. The build has a fixed scope and a written definition of done. The system runs in your accounts on your own API bill, so there is no seat pricing and no landlord.

This journal is itself the output of the system we run in production, and that system runs on named gates. Before this article went out, it passed a thirteen-claim verification pass, every printed number traced to its primary source, and four human approval gates: topic, angle, copy, render. Each gate closes with a stamp only a human writes. And the marketing hire you make later inherits that working department instead of a blank page. Their judgment goes into a system you keep.

You do not have to take a builder's word that this is no longer exotic. Benjamin Gibert, CMO of Base Operations, a Series A startup of about 40 people, runs marketing with no dedicated content team. Over 14 months he built, “mostly through painful trial and error,” a content system centered on Claude Code, as he describes on the GTM Strategist newsletter.⁸ One marketer, one owned system, built and refined over more than a year.

You do, by design. Direction and the approval gate stay human, and the care plan is optional: cancel it and the system keeps working. In our own builds, that human-owned loop is what keeps a system in production.

An owned media department is not a machine you switch on and leave. Two things stay human on purpose. Direction: a person decides what the system makes and why, which was always the ownable part. And the approval gate: nothing goes out without a human approving it. That standing read is the eval loop. It is how the output gets better instead of merely more plentiful, and it belongs to you, not to us.

The honest objection arrives here, and it deserves the corrected numbers rather than the scary paraphrase. Gartner predicts that over 40 percent of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls.⁹Canceled is Gartner's word, and it is an analyst forecast, not a measured failure rate.

Per McKinsey's State of AI 2025 survey, 62 percent of respondents say their organizations are at least experimenting with AI agents.¹⁰ Only 23 percent report scaling an agentic AI system anywhere in the enterprise, and the 62 includes the 23. Read together: most organizations are experimenting, a minority have something running for real, and a forecast says many open-ended projects will be shut down.

Gartner's three reasons are Gartner's. What we can add is first-hand: in our own builds, what keeps a system in production is the opposite of that list. A fixed scope. A definition of done. A named human who owns the output. The deliverable is never “transformation.” It is a production system with a defined output and a person at the gate.

As for maintenance, we offer a care plan, and it is deliberately optional. Run the cancellation test on it: cancel the care plan and the system keeps working, because it is yours. That is the whole difference, applied to us.

The questions founders actually ask when they weigh this choice, answered straight.

An agency is still the right call for one-off campaigns, specialist paid-media execution, and speed before your offer is validated. A system is only worth owning once the work is recurring.

This piece is an ownership test, not a case against agencies, so the concessions are real. A one-off deserves a specialist: a rebrand, a launch film, a campaign you will run once and never again. Paid media at serious spend rewards a buyer who lives inside the auction platforms every day; per Glassdoor, that expertise runs roughly $97,000 a year in salary alone if you hire it instead.¹¹

And before your offer is validated, speed beats equity. Renting a proven team for a few months to find out whether anyone wants the thing is a better trade than building a department for a product that may not survive. The dividing line is recurrence. You do not build a factory for a prototype. You build one when the same work comes back every week, and content, for a founder-led company, is exactly that kind of work.

One more distinction, because the own-versus-rent frame is older than the technology. Duct Tape Marketing frames the choice as renting your marketing, an outside agency or consultant, versus owning it, which it defines as hiring a marketing person internally.¹² It is a fair frame, and until recently it was the only one available.

But hiring a person is not owning a system; it is renting a workflow that resigns when they do. Owning means the department outlasts any single tenure, and the hire you make directs it rather than carries it.

So run the test on everything you currently pay for: the retainer, the platform seats, the freelancers, even the hire. What still works the day you cancel? Where the answer is “nothing” and the work is recurring, that line item is a candidate for ownership. Naming those line items is exactly what our free audit does. It is where this argument stops being prose and becomes a list with your name on it.