For twenty years, marketing leaders have oscillated between two options, neither of which works, and each of which looks attractive mainly because the other one has just disappointed them.
Option one is the full-service agency: one partner, one contract, one throat to choke. Option two is in-house: bring it all inside, own the capability, stop paying a margin on your own strategy. The pendulum swings every few years, the trade press announces a new era, and marketing performance does not noticeably improve. This should tell us something.
The same failure, wearing two costumes
The reason both models disappoint is that they share a hidden assumption: that a single organisation can be genuinely excellent at eight materially different disciplines simultaneously. Performance media, lifecycle automation, brand strategy, local search, conversion design, creative, analytics, and the applied statistics that underpin all of it. These are not variations on a theme. They attract different people, reward different temperaments, and are learned in different places.
The full-service agency solves this by being adequate at all of them and excellent at whichever one the founder came from. Everything else is staffed to the standard of “good enough that the client does not leave”. This is not cynicism; it is arithmetic. No agency can afford to hold the best performance-media practitioner in the market on the bench waiting for a client who needs them.
The in-house team solves it by being excellent at one or two things — usually whatever the head of marketing personally cares about — and structurally weak everywhere else. Worse, it has a recruitment problem it cannot fix: the best specialists in any narrow discipline do not want to be the only person doing that discipline, in a company where nobody above them understands it, with no peers and no career ladder. They go where the other specialists are. That is not a failure of budget. It is a failure of gravity.
The best specialists go where the other specialists are. An in-house team can outbid an agency on salary and still lose the hire.
What a network is, precisely
A network is not a holding company, and the distinction matters. A holding company owns agencies and extracts a margin; the agencies compete internally, hoard clients, and share nothing that would weaken their own position. The synergies are announced in the annual report and are absent from the work.
A network is a federation of independent specialists that share the things which are expensive to build and stupid to duplicate: measurement infrastructure, data, tooling, and a common definition of what counts as a result. Each agency stays narrow enough to be genuinely excellent. Each engagement can draw on the whole.
The critical design decision is what gets shared. Share too little and it is a holding company with a website. Share too much and the specialists get flattened back into generalists. The answer, in our view, is that the measurement spine must be shared and everything else must not. Every agency in a network must define a conversion the same way, attribute the same way, and be graded against the same holdouts — otherwise the network cannot compare its own agencies, and cannot tell the client the truth about which discipline actually moved the number.
Why this is slow, and why that is correct
There is an obvious objection: if a network is so clearly better, why does everyone not simply announce one? The answer is that announcing one is trivial and building one is not.
A specialist agency is only worth having in a network if it is actually excellent, and excellence in a discipline takes years of accumulated, unglamorous practice. You cannot recruit your way to it in a quarter, and you certainly cannot press-release your way to it. A network of nine agencies stood up in eighteen months is nine mediocre agencies with a shared logo — which is to say, it is a full-service agency, and it has reproduced the exact problem it claimed to solve.
This is why Fabulous.Media has two agencies operating today — Infabio in AI and predictive optimisation, IntentConvert in performance media — and seven in formation. We would rather be honest about a twelve-year build than dishonest about a two-year one. The temptation to imply otherwise is considerable and we are declining it, on the straightforward grounds that a claim like that is trivially disproved by a single search, and a firm caught overstating its own scale should not be trusted to report on anything else.
The question a marketing leader should actually ask
Not “agency or in-house”. That framing has produced twenty years of expensive oscillation.
The better question is: for each discipline I depend on, do I have access to someone genuinely excellent at it, and can I prove what they contributed? For most organisations the honest answer is that they have excellence in one or two disciplines, adequacy in the rest, and proof in none — and no amount of restructuring between two flawed models will change that.
A network is not a magic third option. It is simply the only structure that does not require anyone to pretend they are good at everything.
Also
The Price of Execution Is Collapsing. Most Agencies Are Priced on Execution.
Generative AI has not made marketing cheaper. It has made a specific part of marketing — the part most agencies bill for — close to free. That is a different, and much harder, problem.
Proof Is Now the Scarcest Signal in Marketing
When anyone can generate a persuasive claim in seconds, a persuasive claim stops being evidence of anything. What remains valuable is what cannot be generated.